{"_id":"q-en-kubernetes-8e3c1b5434af24681841213f17bfec0a8ba7d713665fc9b8d51a3bd3e2bb8fcb","text":"A temporary workaround is to seed our random number generators. But really, apiserver should assign a guaranteed-unique identifier upon resource creation.\nInternally we use RFC4122 UUIDs for identifying pods. Any objections to making this part of the pod setup? I guess it would really be a string (like \"id\") but with the strong suggestion that it be an encoded UUID. Or we could use docker-style 256 bit randoms, but that might get confusing. If we further lock down container names to RFC1035 labels, we can use <uuid.<nameas the docker container name, which seems much nicer than the current dashes and underscores :) What think?\nBoth points sound great to me. These encoded names are just plain ugly:) Id is also a required field (network containers break otherwise), so it seems weird to have them marked as 'omitempty'. This is probably more an issue for config files than anything else.\nWas going to get familiar and try to fix this - sounds like the suggestion is to add \"ID string\" to Container, and fail PodRegistryStorage.Create() if Container.ID is empty? Or should PodRegistryStorage.Create() populate unset DesiredState.Manifest.Containers[].ID that are empty? Latter seems more flexible (server controls default UUID generation for clients)\nI started in on some validation logic this morning. It's not clear whether unique ID is something users should have to spec or not. I lean towards not. That means the master (api server?) Has to generate a uuid upon acceptance. That uuid would have to flow down to Kubelet. Is that in the same vein as you were thinking? On Jun 25, 2014 6:35 PM, \"Clayton Coleman\" wrote:\nAs an API consumer I like not having to specify things that the server can do for me - having to generate a UUID on the command line to curl a new pod into existence feels wrong. I started but didn't pass down to kubelet yet.\nOne thing to note with global unique identifier is static containers. Today we setup cAdvisor as a static container and have no way to assign it a unique ID globally. , Clayton Coleman wrote:\nStatic as in \"defined on each host via a config file\"? Would it make sense for the Kubelet to auto assign a UUID for containers pulled from files based on the host MAC and the position in the file (or a SHA1 of the contents of the manifest plus the host MAC)?\nAre you setting it up by config file? Can we generate a new uuid when we write the config file, or do we want it to be identical across machines (google prod style)? We cod do something like: if Kubelet finds a config without a uuid, it will assign it a uuid and log it. Internally we go one step further and define \"master space\" where each configuration originator can choose an ID and then manage ids within that space, meaning the master doesn't have to generate a UUID at all just a unique masterspace + pod name. Do we need to go that far? Uuids are notoriously human-hostile. On Jun 25, 2014 7:38 PM, \"Victor Marmol\" wrote:\nYes, it is a static file we leave on the machine. I don't think we have the ability to assign it a unique ID (outside of running some custom init script). Having the Kubelet assign it a UUID seems reasonable to me. , Clayton Coleman wrote:\nTo be clearer on masterspace. Each new pod gets fields such as: Masterspace: Name: \"a name unique within this masterspace\". I am sultaneously trying to argue that new should be DNS compatible, which lits them to 64 lowercase alphanums, and gives names semantic meaning. Making then bad as unique IDs. Need to think more about this... On Jun 25, 2014 7:46 PM, \"Tim Hockin\" wrote:\nI want the container names from the different sources to have different namespaces, so there won't be any collisions. E.g., kubelet prepends/appends \".etcd\" (or something) to etcd-sourced containers, \".cfg\" to containers from the config file, \".http\" to containers from the manifest url, etc. Then it is up to each source to stay unique. Api server can stay unique via uuid or counting up. Config files & manifest url stay unique by humans not screwing up; kubelet rejects them otherwise. This has the nice effect that the container name produced from a config file is predictable without having to do a lookup. This would be good for our own container vm image and anything like it.\nThis started structured and turned into a stream0-of-consiousness, sorry. I agree that unique names are desirable, but I'm not sure \"etcd\" and \"httpd\" are sufficient. The reason we have masterspace internally is because we also need to attach metadata about the master that created a pod (e.g. which cluster it thought it was in, what magic number it had (generation number) and so on). Here's what I think I have convinced myself of, so far. 1) All pods have a required name string, which is human-friendly and DNS friendly (probably rfc1035 subdomain rather than label). 2) All pods which are running have a uuid (suggest RFC4122 but could be something else). This of this as a cluster-wide PID. It has no semantic meaning and changes whenever a pod is started on a new host. 3) The kubelet API includes this uuid. If the uuid field is not specified, the kubelet will assign a new UUID. This does not have the property that Daniel wants - predictable container names. The problem is that you putting semantic meaning into the unique ID. There's a reason that databse best practices involve surrogate keys and that UNIX syscalls operate on PIDs rather than command names. Consider what happens if we get a phantom instance on a split network - both pods end up with the same name - not unique any more. That said, I could maybe be convinced. If we put the rule that the pod name had to be unique within a masterspace, we could sort of punt the problem a bit, for now. E.g. Pod { masterspace = \"\" id = \"\" containers [ { name = \"apache\" be created with container name If the apiserver did not care about phantoms, it could leave off the id8675309 noise. for Google people, this should look very familiar. I still have a vague foreboding about making the uniqueness be the master's problem, but it would get rid of the need for opaque UUIDs. Or would it? We need to handle static pods (e.g. cAdvisor). If they are allowed to collide, then some hyopthetical cluster-data aggregator can not use ID as primary key, and we will need to disambiguate queries by hostname. Blech. What if the rule is that masterspace is optional. If not specified, kubelet will make something up derived from the source. So cAdvsor would say: Pod { #no masterspace id = \"cadvisor\" containers [ { name = \"cadvisor\" be created with container name <host_fqdnThis still sort of sucks in that you can't aggregate by masterspace, e.g. \"give me a list of all containers in the cadvisor masterspace\". But maybe that's better served by labels anyway. Yes, I think so. Thoughts? I could go wither way (UUIDs or masterspace + unique name). Or does someone have other ideas? I feel like this got complicated pretty fast. Tim , Daniel Smith wrote:\nJust to be clear, I only think that the property of predictable names is desirable for containers that come from the manifest url and maybe the container, because without that property, is unhelpful. But you've got me somewhat convinced that maybe I shouldn't care so much about that. I don't know if we want to wait to solve naming in general before we accept a better solution than what we do currently.\nI could put my weight behind either approach. The problem with UUIDs (be they RFC4122 or SHA or whatever) is that they suck for humans. The problem with !UUIDs is that we have to count on the \"master\" to get uniqueness right. , Daniel Smith wrote:\n\"I feel like this got complicated pretty fast.\" Yup:) [I think I've read & parsed all of the comments, but I could be mistaken] I like the idea of DNS-styled hierarchical namespaces. I'm not super worried about multiple masters scheduling over the same cluster of minions, but it does make multi-tenant masters easier (i.e. a provider is running a master, and customers run their minions). It's also something that people are used to. I think it's also reasonable to require that users produce unique names for managing in the system. Machine-generated unique IDs are more useful for running containers (i.e. a global PID). Having this be globally unique is really handy for things like log aggregation, etc. Why not always have the kubelet generate that? The master can learn it when it lists the running containers. It's worth noting that the only restriction of docker names is that they are unique to the host. We can still encode useful human data (manifest+container names) in addition to the unique ID into the if we only parse out the unique ID! FWIW, I'm just polishing my ID cleanups to remove dead (or dieing) code. My goal was to normalize on docker IDs for many/most things inside the kubelet. I should have it out for review tomorrow.\nAlso, clarifying some terminology might be handy. Here's how I think of things (could easily be wrong!): is a collection of . are an instance of a . They can be running (or not). They are assigned to a host. Things that confuse me: have an Id. have a name. have an Id. Seems like the manifest should only have a name (but this gets to the discussion above about names vs ids). has a which really just wraps . Seems like overkill.\n, Justin Huff wrote: I'm more worried about \"masters\" that are config files crafted by humans without coordination. Those should not accidentally collide :) Here \"users\" == apiserver? Yeah, this is doable (and is in fact closer to what we do internally) I think it is useful, but not critical that docker-reported names be human-friendly. I'm keenly interested in this, and I'd like to stabilize it all ASAP. Please do not merge this change until I have had time to feedback on it. Caveat: I am part-time the rest of this week.\n, Justin Huff wrote: The names as defined in pkg/api are somewhat confusing. It all depends on perspective. We really have two APIs here (user -master; master -kubelet), and we're cramming the spec into one file. We should consider whether we really want to do that. From Kubelet's POV ContainerManifest == Pod. Yes, let's settle the discussion about UUIDs vs Names before we rename things. and then lets rename things. I think this was to leave room for growth. Kubelet should do the same and define a Pod type that has an api.ContainerManifest and maybe the UUID (if we do that). BTW: I have a change pending (not sent yet) to do validation of a ContainerManifest - so it will reject manifests that, for example, do not have an ID. The core of it is done, but it needs a test and has some open questions. Hope to have it out tomorrow.\nWhew! This is long. This issue started with no context about what we're trying to do. Starting with the end: Yes, we should clean up the identifiers. Currently, every object/resource includes JSONBase, which has an ID (which probably should be Id). ContainerManifest also has an Id. Container has Name. Port has Name. I'll point out that Container and Port are not standalone resources right now. ContainerManifest is the way it is due to compatibility with the container-vm release. What are we trying to do? I saw several things mentioned in this issue: Human-friendly, memorable, semantically meaningful, short-ish references in operations Predictable references in operations and/or configuration files Unambiguous references in space, such as pods/containers created by multiple masters or by static configuration Unambiguous references in time, such as for logging Idempotent creation () Auto-generation of DNS names () Identification of sets for aggregation Both unique identifiers and human-friendly names have value. Human-friendly names can only be unique in space, not in time. We should use \"Id\" for the former and \"Name\" for the latter. Names could be used to ensure idempotent (at most once) creation, though a non-indexed resource value could be used for that, also. The replicationController would treat \"Name\" in the template as a prefix and would append relatively short random numbers for uniqueness. Static pods could be treated similarly. Label selectors should be used for set formation / aggregation. If users are providing Names, we could also use them for DNS (). I'd use \"domain\" rather than \"masterspace\". Argh, my laptop needs to be rebooted...\nContinuing... I agree we want a mechanism that permits unique id allocation by Kubelets and that doesn't require centralized and/or persistent state. I like UUIDs. There are the issues of ensuring unique MAC addresses in VMs and/or namespaces, and determinism for testing, but I think we've found ways around these issues. I've considered not having human-friendly names for pods before and just using labels instead. Labels are predictable and human-friendly, but don't require uniqueness, and don't require concatenating lots of identifying info together in order to ensure uniqueness, which users WILL do for names (and they'll want to parameterize them). They aren't short, though. Also, I guess part of the problem is that Docker doesn't support labels. We should push on that. Idempotence could be ensured by a client-generated cookie, such as a fingerprint/hash or PR number. It wouldn't be required for static configs. DNS names for instances aren't super-useful if they aren't predictable and DNS-like human-friendly names aren't that friendly if they are long. Short nicknames don't have to be predictable (so they could have a short uniquifying suffix) and don't even need to be semantically relevant -- just memorable (hence Docker's silly auto-generated names, I guess). Services need predictable DNS names, OTOH, and ports need predictable names (for DNS SRV lookup or ENV vars or whatever), and pod-relative hostnames of containers should be predictable, so they can communicate with each other, though I don't know that they actually need to be FQDNs. Other types of services (e.g., master-elected services) and groups will need predictable DNS names, also. Is the main motivation for names for pods to be consistent across all resource types? If so, I could buy into DNS-like hierarchical names for them. I'd use something like \"domain\" or \"namespace\" instead of \"masterspace\".\nPod-relative hostnames and stable internal names are definitely valuable - and limiting \"name\" to rfc1035 subdomain has been extremely valuable in practice to us on OpenShift. Quasi-uniqueness I assume? How predictable? As a concrete example, with something like Zookeeper you need the container to have an identifier/name that is stable across restarts / reschedules in a shared config (so not a pod ID). I'd assumed you'd model this with a set of replication controllers (vs a shared controller) so you had 3 replication controllers with 1 item each, and you'd be able to either set an ENV per pod, or use the name in order to apply that. If name changes over pod instances that rules out the reuse there.\nOne last thought from me; it occurs to me that docker already generates a long container id. Perhaps we can consider using that directly as our spatial/temporal unique identifier, and use this hypothetical dns-style solution as the friendly human name. We'd need to investigate just how unique docker's id's are, and we may not want to depend on docker for that, but it would reduce the number of IDs needed. Also, as a footnote, if we end up with both pod.ID and Manifest.ID, IMO they should be the same identifier.\nFor clarity, let me suggest the convention that \"name\" means a friendly mostly human readable string, possibly in the style of dns names, and that \"id\" means an opaque, machine generated identifier, guaranteed unique at some level of resolution. Maybe everyone except for me is already using this convention. :) But I want to talk about ids and names in general without referring to a particular implementation. I think, in that case, the rep. controller itself has a name, which it can use to make names for the pods it creates (prepend/append indices or something). IDs for the pods could still be generated by the apisever or wherever we decide they need to be generated. I was trying to say that since there's a 1:1 relationship between pods and manifests, we shouldn't make different IDs for each.\nTrying to collect thoughts and ideas into a proposal. It's tricky considering it from all angles (users, apiserver, replication controllers, kubelet). My background is node-centric, so I am counting on people to smack me if I say something dumb for higher-levels :) I have to run out right now, but I'm hoping we can distill the discussion into a design O(soon). NB: We spec names as RFC 1035 compatible, though we might extend that to allow purely numeric tokens (e.g. ) but I forget which RFC that is. Docker names allow underscores, which kubelet reserves for use in infrastructure containers. From kubelet's point of view: 1) All pods have a namespace string, which is human-friendly and DNS friendly (rfc1035 subdomain fragment). For example: \"\". This is used to indicate the provenance of the pod. If the namespace is not specified when creating a pod, kubelet will assign a namespace derived from the source. For example, a file \"\" on host \" \" might get masterspace \" file-\" (the big mess is a base32 encoding of the path). 2) All pods have a name string, which is human-friendly and DNS friendly (rfc1035 subdomain fragment). For example: \"\". These names are typically NOT user-provided, but are generated from infrastructure. For example, if the user asked for a job named \"myjob\" the apiserver must uniquify that into a pod name. 3) The namespace + name pair is assumed to be unique. This provides simple idempotency, for example when re-reading a config file. Open: Do we need UUIDs at all? For what purpose? The only argument I can come up with is to protect against accidentally non-unique names at cluster scope (e.g. in a giant database of namespace, name, uuid, stats) you could disambiguate colliding namespace+names pairs with uuid. Other justifications? From the apiserver's point of view: 1) The apiserver has a configured namespace. 2) All incoming pods have a name. I don't know what the formatting and uniqueness rules are here. 3) Upon acceptance, a pod is assigned a unique ID of some sort. 4) Upon assignment to a minion, the name and unique ID become the pod name. Open: Should the unique ID persist if the pod is moved to a new minion? Maybe we need two IDs - one that sticks across moves and one that does not? , Daniel Smith wrote:\nNumeric host names:\nRegarding the open question: is a pod on a new minion the same as the old pod? Is a move an action that is logically part of kubernetes, or is only \"remove\" and \"create new\" available? If it's the former, it seems like the ID should be the same, if it's the later it seems like it should be different. We have a use case for being able to move a pod from minion to minion (and any volumes that come with it) - however, since this requires the volume data on disk to be in a resting state, it's not an operation that seems to lend itself well to the replication controller (since the move is inherently stateful). So I would expect this to be managed as an operation above the replication controller, vs part of it. One namespace per apiserver can be limiting if the namespace is automatically bound to DNS and you're dealing with very large numbers of containers, but it doesn't sound unreasonable if you are using wildcard DNS.\nBTW, it looks like docker IDs are more or less just 32 random bytes: They enforce them to be unique per-machine:\nUnique ids disambiguate among multiple instances reusing the same name over time, such as when polling termination status. Beware of moving unique ids. Once we do that, they are no longer unique, such as in the cases of live migration or even just preloading. If we do move ids, we should have 2 ids, the movable one at the cluster level and a non-movable one at the node level, and the movable ids would correspond to zero or more of the non-movable ids.\nOK, collecting thoughts again. If this is acceptable, I (or Clayton) can write up a short .md file covering identifiers. There are still a few FIXME comments in here. Please tell me if I am capturing any of this incorrectly. Global: We spec most names as RFC 1035/1123 compatible. Docker allows container names to include underscores, which we reserves for use by infrastructure. From kubelet's point of view: 1) All pods have a namespace string, which is human-friendly and DNS friendly (an rfc1035/1132 subdomain). For example: \"\". This is used to indicate the provenance of the pod. If the namespace is not specified when creating a pod, kubelet will assign a namespace derived from the source of the pod. For example, a file \"\" on host \"\" might get masterspace \" file-\" (the big mess is a base32 encoding of the path). (FIXME: do we need this if we have a UUID as spec'ed below?) 2) All pods have a name string, which is human-friendly and DNS friendly (an rfc1035/1132 subdomain fragment). For example: \"\". These names are typically NOT user-provided, but are generated by infrastructure. For example, if the user asked for a job named \"myjob\" the apiserver must uniquify that into a pod name. 3) The namespace + name pair is assumed to be unique. This provides simple idempotency, for example when re-reading a config file. 4) When starting an instance of a pod for the first time (i.e. not restarting an existing pod), kubelet will assign an rfc4122 compatible UUID to the pod. This provides an ID that is guaranteed unique across time and space. If the pod is stopped and an identical pod (same namespace + name) is started, a new UUID will be assigned. 5) Kubelet will use the aforementioned identifiers to produce unique container names, for example \"\". NB: The UUID is per-pod, not to be confused with Docker's own container IDs which are per-container. From the apiserver's point of view: 1) The apiserver has a configured namespace, for example \"\". 2) All incoming pods have a name assigned by the originator of the pod (be they human users or other infrastructure). Pod names must be unique in space, but not time. 3) Upon acceptance, a pod is assigned a unique ID (FIXME: just spec it as rfc4122 again? could maybe be simpler here?). This provides an easy way to disambiguate successive pods with the same name. This ID will persist for the lifetime of the pod, across restarts and moves. (FIXME: is this really needed or is name good enough?) 4) Upon assignment to a minion, the name and unique ID become the pod name. 5) The namespace + name together must be less no more than 255 characters long. If DNS service is to be configured automatically (not a feature yet, but has been discussed), the pod namespace + name will already be DNS compliant. , bgrant0607 wrote:\nI think this seems like a good outline! Two comments: config files, maybe use a hash of the file vs a base64 encoding of the path as the component of the namespace. Just seems more natural. \"These names are typically NOT user-provided, but are generated by infrastructure.\" I'd consider being more explicit about that behavior, even if it's as simple as a \"unique prefix will be to the name\". This sorta touches on the whole id/name discussion....\nAnother positive attribute of unique ids: they are typically more concise than names, and are therefore more efficient to use in cross references in logs. Based on experience in Omega, I think there's no question that we need unique ids. Also, if you play around with Docker for a short while, you'll find a pile of old objects that only have unique ids and can no longer be referenced by name. How do you query them, resurrect them, or delete them once their names have been recycled? Is there a compelling reason to use a different kind of unique identifier in the apiserver than in the kubelet? I don't think so. We could probably shorten it (e.g., remove the locator) when we combine it with the name to form the name for kubelet.\nDefinitely prefer hash of something in the config file (maybe hash of the JSON of the pod's manifest or something) for use in the namsepace component.\nI'm happy to write this up as part of (along with some more tests and the apiserver implications) once folks reach closure.\nOne minor note regarding Docker ecosystem - \"name\" is currently being used in Docker for a lot of lightweight integrations (linking, sky dock dns, hostname in container). A side effect of any generated name for a Docker container is that those lightweight integrations may become more difficult for end admins. Is there a practical way to make the name appropriately unique on the minion without breaking those potential integrations (making the name a subdomain fragment by omitting '.' for instance)\nI would also like to see us drive labels all the way down into Docker. That would enable us to abandon much of what we are encoding into the name today. Brendan On Fri, Jun 27, 2014, 8:39 AM, Clayton Coleman wrote:\n+1 to what brendanburns@ wrote. Labels provide much cleaner solutions for many use cases.\n, Justin Huff wrote: This is an implementation detail - human-friendliness of docker's names is a nice-to-have not critical, I think. Yeah, where is the line between \"ID\" and \"name\"? The apiserver COULD just send a UUID for the name, but I think we want the structure and content of a pod spec to stay largely the same throughout the entire system, at least for now. I don't think we've got any consensus on diverging it, anyway. For that reason, I think \"name\" makes sense. It is more human readable than not.\n+1 to labels , brendanburns wrote:\nWhat would break on those? We're not proposing to do anything with name that is not totally valid as per standalone Docker? , Clayton Coleman wrote:\nIt's not that the names aren't valid for Docker, its that something consuming those names (as a dns prefix a la skydock, or as a hostname) would break due to length or format on the Kube generated name. I don't think that is a blocker to the proposed naming patterns above, but it's a consideration when thinking about how other software that plays well with Docker might react. Concrete ideas might be to reduce the generated names' length and avoid using '.' as a separator.\nI would argue (for the sake of arguing) that anyone who was making assumptions about container names deserves to be broken. But yeah, let's keep an eye on it. , Clayton Coleman wrote:\nAgree, naming is hard\nWriting up a summary md - does this belong in api/doc, , or another location?\nI would go for or something , Clayton Coleman wrote:\nReferenced pull includes a summary of this discussion, open questions: implication 3-1 (agree/disagree)? I defined \"pod unique name\" and \"pod unique identifier\" to represent thockin's kubelet items 3 and 4"}
{"_id":"q-en-kubernetes-fa1ebef63f76d15ef1c28364f7447772418ff92e79eefd59fa95d78576bc8628","text":"If you are using a private docker registry, you first need to call something like and provide a user/pass/email. Any further calls to will pick up your identity from a file in $HOME/.dockercfg. Before a kubelet starts a container, it calls the equivalent of . Unfortunately, the auth information is not picked up from $HOME/.dockercfg, and there is no alternate way to configure it. The kubelet needs to support docker registry auth.\nHere's where the auth should be getting passed to the docker client:\nI'm not certain, but it looks like authenticated docker pulls would have worked before this patch:\nSee .\nah, search skills failed me again"}
{"_id":"q-en-kubernetes-24df73a6db05c45f0cac4e827cf56a2dc89af2bfb599ee7ed6ea71c4b1f5c613","text":"Should be able to tell from logs if a minion isn't present because cloud provider didn't tell us about it, or if it's failing health checks."}
{"_id":"q-en-kubernetes-823174cd7a80c2aabbd968d214ca3d07426d76d7cc3a412fb0316f4befd8fbf6","text":"https://travis- Seems surprising-- I will look into it later if no one else does first.\nGot another flake. Looking into it."}
{"_id":"q-en-kubernetes-41a4a9d51023611803efc4b60b5824c1d1fbca7fe22577f19d7270fd74d5ef0d","text":"if you delete a service and restart it the port is no longer accessible To reproduce: 1) start a pod 2) start a service All works fine port is assigned - traffic flows 3) delete service 4) start service the port is now not assigned Here are sample configs: pod: Service:\nI thought this was a known issue, but I could not find another issue on the same topic. Note that all containers started after the first instance of the service was created will have environment variables referring to that instance and won't see environment variables referring to the new instance until they restart. This is a fundamental shortcoming of the current approach.\nI understand what you just replied but that is not the issue here. The problem is that when you delete a service , no other service can resuse that port successfully. The port is removed from a netstat listing. When a new service tries to use that port it gets allocated but you can not communicate to the pod through it. It is hosed until I re-install the cluster. (I have not tried rebooting the cluster) BTW this is a bug not a design issue.\nAh. The previous issue I was thinking of was . It sounds like the issue hasn't been completely resolved.\nI will take a look at this today.\nSorry for the problem, has the fix. --brendan\nThanks! On Aug 19, 2014, at 2:26 PM, brendandburns wrote:"}
{"_id":"q-en-kubernetes-99e02bbd6dfd36028985d05c078261611bf34deff196750e9774a0cbdb8c8a20","text":"We have an e2e test that runs the guestbook example. We should also have one that runs through the other example."}
{"_id":"q-en-kubernetes-09fca157bd43c35013d3772434f9998a9623afc2f92a2c7f553134f7bdea5afe","text":"Running vagrant from scratch gives the error: For vagrant cluster, the last minion cannot be detected. The /tmp/minions file only shows two minions, but after a while, using the same command, i.e. will get three minions. There is a race between kube-up and validate-, we should wait a while before panic.\nI can look to fix this."}
{"_id":"q-en-kubernetes-b9fbe44fc01f631378bc467f73c902ae0e65f3298439f7434f4b3f0d323b9e8e","text":"As a sysadmin When using the Vagrant cluster provided And logged in as the vagrant user to the master or minions I would like to be able to call docker commands wihout requiring the 'sudo' prefix. For Fedora =19 and RHEL=7 this can be enabled by adding the vagrant user to the docker group in /etc/groups\nThis should be trivial, but its tricker to do because of the whole salt configuration process as the docker group is only created on each kubernetes-minion once docker is installed by the salt-minion. To do this properly, we would need a salt state change for this, and new flag in to let you specify additional users that should be in the docker group."}
{"_id":"q-en-kubernetes-a987b5e3b3353d21d7a8120b4ee9be1106a65ade9430df500d1a08c675d1ca1a","text":"When creating replication controller (e.g. frontend php-redis from guestbook example), scheduler gets stuck and never recovers. This is a race, I have to run several times to see the error (A script to delete controller and pods, then create new controller, leave enough time in between for k8s to react). I've created PR for an attempted fix.\nI saw this too while following the guest book and creating the php/frontend step. In my case, I had 2 minions while the replication controller configuration had configured 3 replicas, so 2 pods got scheduled -each on another minion, the third pod had empty Host (saw that when doing 'list pods'), and the 3 of them were stuck in \"waiting\" status. I deleted the controller and the pods, fixed the configuration in json file to be 2 replicas and recreated the controller and everything worked. Does your fix take care of this use case? (too many replicas, too few minions/one of the pods is not scheduled)\nThis is different from your use case. If you have more replicas than minions, then the scheduler will get stuck for sure (with the error msg probably being 'failed to find a fit for pod'). AFAICT, it's not support yet.\nIf you remove the HostPorts, it should work for you. thanks for report.\n, thanks for your replies. I understand why it isn't working, but I'd expect one of the 2 options to happen instead of the current situation: the creation of replication controller if there are less minions than replicas, AND there is a hostPort property. In other words - validation of the content during creation. least schedule all replicas that can get a minion (in my case 2 out of 3) and leave only one at \"waiting\". Is there an issue open about the use case I describe?\nshould be true today. is less possible than it seems like-- minions can come and go, and become over or under loaded in the time between creation and when the pod shows up at the scheduler. So even if we check up front, we still have to check again later, and sometimes the first check will pass but not the second."}
{"_id":"q-en-kubernetes-4bb1462c8ab8a913aa37a2b81374b4383daa9203c50249aab71f7ef3e87e35d5","text":"the file and line for these log statements is in the That's not helpful I1031 11:09:06. ] 'foo' has no storage object I1031 11:09:06. ] 'pods' is not a redirector I'd rather see Maybe use runtime.Caller in httplog to somehow fake the log to come from the caller's file/line.\nYeah-- actually this should be a feature of the glog package. Would be awesome to do:\nFollowup: glog is an opensource copy of a google-internal library; there's already something enabling this internally; we'll try to get a new version pushed externally in the next week or so. Then we can fix this.\nAwesome. On Oct 31, 2014 2:53 PM, \"Daniel Smith\" wrote:\nshould meet your requirements.\n+1, I would love to have this."}
{"_id":"q-en-kubernetes-cbed3b419c4c061a7654b2a666cebe928bfe0af38a566faf794d9f808fca93d4","text":"I suspect this is dues to the recent change to container VMs as the base image? ifconfig docker0 - exists but should not. iptables-save | grep docker - 4 docker rules that should not exist.\nIs this still relevant?\nwas still a containervm turd the last time I checked. It's just not causing any harm.\nYeah, no harm, just ugly. , Zach Loafman wrote:\nI think we should remove the docker0 from minion nodes, but do we still need this for master node? cc/\nThis is required by running etcd server as pod with flannel on gce nodes. Please sync your cluster with this fix."}
{"_id":"q-en-kubernetes-3a47714aea9d63183fb02abaef06e69a4e2b29d7c7283e834a03eb83ec0694c5","text":"It looks like we always try and hit the metadata server even if we aren't on GCE. We then print an error out. This should either be silent (assume after X attempts that we aren't on GCE?) or should only be activated with an option. Email thread: !topic/google-containers/w-ei5Xs6K0Y\ncc:\nFYI, the way the credentialprovider logic is set up, providers are asked once per keyring creation whether they are enabled. If they are enabled, then they are asked to provide their contribution to .dockercfg each time the lazy keyring is accessed. For efficiency, there is a caching provider, which composes with heavier providers and stores their .dockercfg contribution for a predetermined TTL. What was happening here was that \"Enabled()\" for our GCE-metadata implementations was logging about errors fetching \"http://metadata\", where the entire \"Enabled()\" check is \"err == nil\".\nI believe that this was fixed by"}
{"_id":"q-en-kubernetes-142e8decd8d12f564814b0a08026ec0c2500ddcd8040f0182edc6a7f51b222f8","text":"0/8 is with a company's intranet, so it can be insecure. This is similiar to \"For GCE, allow insecure registries anywhere in 10.0.0.0/8\" Thanks!\nLinking to and .\nThis was"}
{"_id":"q-en-kubernetes-c802a37aad85dfdce2d6232f30f1e039273d58e19f8aa63b8acb725eeb2b6040","text":"it's impossible to GET replication controllers on v3. on v1 on the same server it works fine. example request: GET on http://localhost:8080/api/v1beta3/replicationControllers returns 404 http://localhost:8080/api/v1beta1/replicationControllers returns 200\nv1beta3 moved to all lower case URL Would give that a try first Sent from my iPhone\nlooks like it's working. but it breaks backwards for clients who worked with v1. For instance, in my gem I can't now smoothly work with both versions because I need to treat the name of the resource differently. why don't at least redirect those who call replicationControllers to the lowercase too? Also, there's no way to know which resources are available and their names. A lot of REST api on their most top level at least publish the links to main resources, but if I call: http://localhost:8080/api/v1beta3/ it returns 404\nBecause we are actively trying to drop support for mixed case prior to 1.0, I did not continue supporting the old url paths. If you look at the go client, we have a \"legacy\" flag we use to control this and a few other behaviors.\nwell yes, but breaking backwards creates more work for the users... more \"if this version ...\" kind of stuff.\nWe got v1bet1/2 wrong and don't want to carry it forever. Stopping it early before v1 is our priority.\nok :( are the REST apis going through any kind of design review or code review?\nv1beta1 is really an alpha quality API. It grew organically and did not go through any kind of design review, as evidenced by the many issues you've filed. We want to get rid of it ASAP. We intend v1beta3 to be the \"release candidate\" for the v1 API, which will be the stable API. As for knowing what exists, you can now browse or GET . Additionally, pull is in progress, and will return a list of valid paths to GET .\nthanks. will check swagger for now"}
{"_id":"q-en-kubernetes-988dd0ceccf3dcde8b8ea1670f138b61545b1a8185bf62f096fafbe67e654232","text":"For example, kubernetes-ro service appears with empty selector (this service was created by the hack/local cluster) - is that ok? What meaning such service has if it has no entity to fwd traffic to?\nkubernetes and kubernetes-ro are currently \"virtual\" - they are not backed by running pods, but by the master itself. , abonas wrote:\nif that's internal impl, why is it exposed to user? let's say a client pulls all entities and builds a UI around it. then a service with no selector arrives, and it's internal to kube - how the client should know not to present it/touch it? (similar to the \"net\" container in pods)\nIt's exposed exactly FOR users - so they can find the apiserver. It has a corresponding Endpoints object, it gets a portal IP and a kube-proxy rule, and it gets service env vars and a DNS name. The fact that it is virtual is an implementation detail. That it exists is all users should need to know. , abonas wrote:\nAnd FWIW, the master components may one day actually be pods... , Tim Hockin wrote:\nthanks for the explanation. so any general rule how to distinguish kube pods/services from user defined?\nDo we need to distinguish them? We could add labels, I guess, if it is really important to distinguish them.\nsometimes users would want to see only the things they defined and not implementation details of the system.\nOne could argue this is why the users should be using labels. I'm not against defining labels for \"special\" stuff, but it's new territory, so we should enter with caution. , abonas wrote:\nI think the right answer might be to put them in a special \"cluster\" or \"system\" namespace. , Tim Hockin wrote:\nSee If we put them in a \"kubernetes\" ns, the DNS name would be :) I'm not against segmenting them into a different namespace. We just need to decide on what that namespace is called, plumb through that configuration, decide on compatibility for a short while or not, and decide if that namespace is \"special\" or just \"different\" - i.e. should the skydns, heapster, etc pods go into this namespace? , Brendan Burns wrote:\nAs said, the services are there for applications that need to contact the apiserver. Additionally, nil selectors can be used by users to import external services. Adding labels to the kubernetes services won't help. Users need to manage their objects using labels. /cc\nYeah, lots of consumers will expose \"special\" services. IN fact, in many clusters the \"special\" services will outweigh the unspecial ones. At this point, the namespace they are in is completely arbitrary, and I think because it is a name, it should be free for the end user to choose where they are surfaced.\nPer the proposal here: , I think the namespace should continue to hold these services, and they require no special labels. I sense comment is made without thinking through what Kubernetes operators do to bootstrap a cluster. I think it's reasonable that namespace holds everything you need to run, and operators may or may not choose to segment their data in other namespaces. If they want to not see this information, its an exact reason for why they should use their own namespace. I don't think we need anything new here.\n- I'm not clear on your meaning. Did you mean that we should make the \"cluster admin\" namespace a config param that everyone knows about and can be different cluster-to-cluster? Or did you mean that default is as arbitrary a name as any other? I'd be fine moving things out of default into \"kube\" or \"kubernetes\" or \"cluster\" or something, but I am less convinced that a flag is a good thing. , Clayton Coleman wrote:\n- wherever we stick this stuff, it should be the namespace that is initially bootstrapped, which today is . I continue to believe that is as good a name as any other. I do not see a need to do more here.\nI think as soon as you bring naming into the picture, making decisions based on names gets very tricky (what if I want my logging service to be everywhere?). If there is a magic namespace that exposes things to the rest of the cluster, we should ask whether there's only one magic namespace, or multiple? Are services supposed to cross namespaces? It's great that pods can find the master... but that's not the only thing they'll need to find in each namespace. And baking in special names (or customizable names) into your app code is fragile, because then your code only works inside Kubernetes. I'm just trying to pump the brakes on special services prior to us designing them... :) ----- Original Message -----\noh, to be clear, I don't want \"special\" services - I am merely asking if moving non-user-defined objects (where that is defined very loosely) out of \"default\" is worthwhile. I tend to agree with Derek that it's not worth energy. , Clayton Coleman wrote:\nagree with you that lots of consumers will expose \"special\" services We should not use the (namespace + k8s-service + skydns) mechanism for publishing the location of those special services. We should decouple the choice of namespace to run the pods of the service in from the well-known dns address of special services. Otherwise, it will be hard to move the pods of the service to another namespace, which people will definitely want to do as they grow. Instead, we need a place to publish well-known services which is not tied to the namespace of the pods of that service.\nyou open a whole new topic. I've noodled with the idea of a Name resource type. We would generate a Name for each Service as \"service.namespace\", but users could also add their own Name resources. , Eric Tune wrote:\nI'd like to understand better your comment regarding nil selectors used to export external services - can you elaborate? give an example? my understanding is that service redirects traffic to pods based on labels. with nil selector service does nothing, so why is it a service in the first place? I'm pro decoupling special services/internal implementation services from user defined entities. whether by separate namespace, or by saying that the default namespace is reserved, or by any other solution. because otherwise internal implementation and user defined content are mixed and that's no good for a proper management. just tagging you so you could see this issue and the related discussions.\nServices abstract backends - they don't have to abstract pods. Some examples: You want to have an external database cluster in production, but in test you use your own databases. You would have a service for both and use the environment variables, but the production service has a nil selector with its endpoints set to the database cluster. You want to have an empty service because you're reserving a name (that isn't accessible yet) but will be in the future You want to point your service to a service in another namespace or on another cluster. You set the endpoints for your service to the publicIPs of the other service. ----- Original Message -----\n, thanks. regarding the last point - I was under impression (based on some other discussion) that there would not be cross namespace interaction? that entities can can access only entities in the same namespace?\nCross-namespace is currently allowed and I think has to be allowed, though perhaps (in the fullness of time) that can be policy based rather than wide open. , abonas wrote:\nplease see this correspondence. my understanding is that there is isolation\nResponded there. I think there's some less-than-crystal-clear words in play. , abonas wrote:\n+1 to Eric's comment about ensuring there is a consistent manner for system services like kubernetes-ro, if that's not fixes for all clusters it's going to be a nightmare to write reusable containers for multiple different clusters. I don't care if we do that by using a constant namespace, or by giving system services two different names, but there needs to be at least one name that is constant. Regarding default vs. cluster namespace, I think the issue is one of clutter as a user is starting to understand k8s, you turn up a new cluster, launch a pod, or service, and it's hard to find your one object on the sea of system created pods and services. It's visual clutter that confuses people as they onboard. On Jan 23, 2015 3:01 PM, \"Tim Hockin\" wrote:\n- can you drop me a line (my github name at google) - I want to ask you something, and you have no contact info :) , abonas wrote:\nsorry for naive question but why would I care as an user that wants to run its existing workload on Kubernetes about kubernetes-ro ? If I write an application that uses kubernetes-ro then I'm locked into Kubernetes. I understand the need for kubernetes-ro being accessible by cluster level services such as monitoring, logging. I also can imagine that a power user might want to start its own cluster services that use kubernetes-ro. I don't see clearly why would majority of users care about kubernetes-ro being visible from their container. I see a reason for the cluster services being visible and accessible for Kubernetes developer, but from an user perspective such services only make the system more complex: why do I see these services (pods,rc) ? are they related to my application ? should I care about them (especially when I use hosted Kubernetes) ? why does my container get all these environment variables for cluster services ?\nI will update file to explain why one would have service without a selector.\nRight now, every service goes everywhere for simplicity. Name spacing and filtering probably make sense in the long run, but it's not a high priority for v1.0 Brendan On Mar 4, 2015 5:02 AM, \"Rafał Sokołowski\" wrote:"}
{"_id":"q-en-kubernetes-e9e89a316cb39ee57596a3170474f837d1fc36ca6f45ccf3902663a7708546fe","text":"As long as we have it in the API, it should maybe be renamed to gcePersistentDisk ?\nSGTM"}
{"_id":"q-en-kubernetes-7289a91ca3371479da7fb74d0ceced629b874ff7b3b9f9b433395853883821d6","text":"This whole set of tests (if it can't be run) should be omitted from the test execution, otherwise folks running on Mac can't verify everything else is clean in a sane fashion. Passive aggressive assign to Tim, I may get around to fixing this before he does.\nLast I checked Travis + Go didn't work. We should check again, and enable OS X builds in Travis if it's possible.\nYou could argue that if the unsupported mount () is being called in a test, there is a problem anyway - it has to run as root. Will investigate. On Jan 25, 2015 9:09 AM, \"Brendan Burns\" wrote:\nWithout a Mac to test on, I feel like this would be better fixed by someone who uses Mac... Probably: just rename to I just discovered \"implicit build constraints\", sigh. Go is crossing from user-friendly to sort of absurd, IMO."}
{"_id":"q-en-kubernetes-d0f38e058dd83a69064953b4bc0c2c3ffa6cae2ccfd3a071336c45d6b39f4940","text":"wrote: I just ran across this the other day. but I am seeing the error in the service. Here is a sample of the output. I am running v0.8.2. I can get the behavior to go away temporarily by rebooting the node or restarting the service but it always seems to come back eventually."}
{"_id":"q-en-kubernetes-4be2baac657b91c8150c9eaeb075c4073519ae9c3e811b2f182b2a22b3be9af8","text":"We found yesterday that cAdvisor wasn't being started properly in 0.9.0 and 0.9.1 due to an issue with manifest file format changes (, fixed in ). If any of our e2e tests even checked that cAdvisor was running or exposing metrics on all nodes, it would have been caught before release. We should add at least some coverage for it to catch bugs like this.\nIf any of you are up for adding an e2e test for cadvisor, that'd be much appreciated!\nI will work on adding one for cadvisor. , Alex Mohr wrote:\nWe already have a e2e test, which validate monitoring of Clusters using cAdvisor. We should figure out why the test passed without detecting such failure, and improve that test.\nThe existing test will succeed even if some of the nodes do not have cadvisor running. We can enhance the monitoring test to specifically check for cadvisor's health as well. , Dawn Chen wrote:\nThis test seems to be failing deterministically.\nThis was after fix to retry, but we're not exactly sure why.\nOnce I have emerged from the e2e test for cluster level logging with Elasticsearch quagmire (soon I hope -- perhaps by Wednesday) then I might also be willing to lend a hand with this since some of the experience might be relevant.\nActually latest error looks like So it's still not coming up after 5 minutes. The question is how long should it be taking to come up, given that the cadvisor containers get scheduled basically on node creation. The test finished in < 1s on my e2e cluster that has been up for awhile.\nOr there's an actual bug here. can you look into it?\n(I looked at the Jenkins log, and this test ran a good 10 minutes after startup, so it probably had 15 minutes to finish coming up. That seems wrong, given that cAdvisor is actually the first container we ever schedule, and it gets scheduled raw.)\nSorry, I hadn't seen -- not sure if that fixes this?\nNo, there are failures after as well. On Feb 20, 2015 5:39 PM, \"David Oppenheimer\" wrote:\nI don't see any issue with the test itself. The test attempts to access a REST endpoint ('/stats') via the proxy on the api-server. Is it possible that the api-server doesn't get to a stable state for a while?\nThis test executes after several others that hit the apiserver? On Feb 22, 2015 4:15 PM, \"Vish Kannan\" wrote:\nNoob question: How can we tell if it runs this test after the other tests?\nYou can't tell deterministically (i.e. it's not hardcoded). I was merely commenting that in several runs where this fails, it has run after many other tests. We've been randomizing the spec order for a while. You can look at the Jenkins console logs to get the random seed for a given run: And the console logs have the order just by manually scanning the horizontal bars, but it's kind of pain. (That seed was off of go/k8s-test/job/kubernetes-e2e-gce/2638/consoleFull)\nThe reason for this failure is . I am closing this issue since the test itself is fine."}
{"_id":"q-en-kubernetes-9c9d5bfd1ce2491596aef565e138702c57296cb976abe91eb5a2e9202f8205ed","text":"W0128 18:44:26. ] Pod from test failed validation, ignoring: namespace: required value '' E0128 18:44:26. ] Could not construct reference to: '&api.BoundPod{TypeMeta:api.TypeMeta{Kind:\"\", APIVersion:\"\"}, ObjectMeta:api.ObjectMeta{Name:\"foo\", Namespace:\"\", SelfLink:\"\", UID:\"foo\", ResourceVersion:\"\", CreationTimestamp:util.Time{Time:time.Time{sec:0, nsec:0x0, loc:(time.Location)(nil)}}, Labels:map[string]string(nil), Annotations:map[string]string{\"\":\"\"}}, Spec:api.PodSpec{Volumes:[]api.Volume(nil), Containers:[]api.Container(nil), RestartPolicy:api.RestartPolicy{Always:(api.RestartPolicyAlways)(0xe3d670), OnFailure:(api.RestartPolicyOnFailure)(nil), Never:(api.RestartPolicyNever)(nil)}, DNSPolicy:\"ClusterFirst\", NodeSelector:map[string]string(nil), Host:\"\"}}' due to: 'selfLink was empty, can't make reference'. Will not report event: 'failedValidation' 'Error validating pod foo.<empty-namespacefrom test, ignoring: namespace: required value E0128 18:44:28. ] Unable to read config path \"randomnonexistentpath\": path does not exist, ignoring --- FAIL: TestReadFromFileWithoutID (0.20 seconds) Expected update, timeout instead"}
{"_id":"q-en-kubernetes-48b6be9a789fe76ef913476cd055a1bb3da8978c4034069a469636df6a1d0695","text":"writes: As we start looking into etcd and event memory pressure [], we should look into switching over to etcd 2.0 which was released today.\ncc/\n+1 On Jan 28, 2015 5:29 PM, \"Dawn Chen\" wrote:\nMy long-running cluster tests resulted in a hard-crash on etcd: We should move to 2.0 ASAP in the hopes of (a) fixing the issue and if not (b) prioritizing the fix since we'll be complaining about head. Abhi, is this something you can pick up and drive to land in O(days) please?\nCalling this P0 as it blocks further long-running cluster stability testing.\nI will do this today. , Alex Mohr wrote:"}
{"_id":"q-en-kubernetes-618a17ee1a2bfcaa30e720bd6c59c3ac91ee900494c2c92ac1b2427757bb09d9","text":"reported a variety of failure modes in .\n+1\nI figured I'd add to this issue rather than create a new one. I attempted to create a vagrant cluster today (dev tree is up to date) with 1 Master and 1 Minion and the cluster creation is stuck in the 'Validating master' loop ( for over 10 minutes in the validating loop ). salt-master and salt-minion logs seem to be empty ( they don't appear to be running, in fact ).\nI still have the cluster running in case anybody wants additional information.\n/cc please let me know if there is any other information about my vagrant setup that I need to post here.\nI have been working to get more stability in this environment over last few days. Can you try again from HEAD after this merges:\nCan you also verify that you built a release prior to running cluster/Kube-up?\nyes, I started from scratch and built a release before running cluster/kube- - cluster/kube- seems to fail fairly early in case there is no release build available.\nIt looks like got merged. I'll try again and get back to you.\nThe vagrant cluster seems to come up ( completes master and minion validation ). However, there is new credentials related error that is showing up at the end of running cluster/kube- :\nI'll add one more problem I've encountered. Runing from head:\nNew to me. Will investigate more tomorrow. Sent from my iPhone\nAlso new to me :-) - will see if I can reproduce either. Sent from my iPhone\nYour credendials and certificates are not working with new environment. Just delete your .kubernetesvagrant* files and recreate the cluster. It would be nice to automate this task, if you tear down the cluster.\nI am able to reproduce the 502 Bad Gateway error and am working on determining what changed in order to provide a resolution over the last couple days. Please assign this issue to me.\nThe issue appears to me that the kube-apiserver is not getting provisioned by salt correctly. Parsing Salt logs to determine why.\nLooks like the file is not getting laid down by salt for some reason ;-)\nThe Jinja template was changed as part of identifying the cluster. Fixing up now, and will send a PR to close this issue.\nthat was it, thanks!"}
{"_id":"q-en-kubernetes-1898b360a97767b2693ea582431a164ade42bd7fbf892fdd653b38d513260c90","text":"contains function get-password. This uses a template which gets the current context. But a user who has not yet used the kubectl contexts feature has no context set. This causes a nil pointer deref while evaluating the template.\nI think kube-up creates the needed entries in the .kubeconfig file. But it calls get-password before it does that. Need to allow the kubectl config view command to fail.\nokay, I have a fix."}
{"_id":"q-en-kubernetes-2d8db3db60a856068241611a4e608d0c0eb92592804309234e8facba95234683","text":"I have been following the instructions here: to try to explore the kube UI. by executing: I get the following which shows it successfully started: When I try to access the page I get a 404:\ncould you try running ? I believe is a typo of\nAlso, try accessing instead.\nthanks"}
{"_id":"q-en-kubernetes-33e3eb0cb35f85a754b3ee3944c38e72c05cb6c9cd3f09473af5448c1ac03388","text":"I tried to create a pod without containers to see if it's supported or not. (does a pod without containers have a meaning in k8s?) I was getting \"conflict 409 already exists\" error although I used a new pod with a new name/id, so it's not clear whether it's supported or not, and whether this error is a mistake (409 instead of 400)\nWhat is the reason to create pod without specifiing image / creating container? Please provide a use case when such functionality is necessry.\n- to me it's not necessary. However if it's not supported, a validation should prevent it not to allow corrupted entities. And if it is supported, it will be good to understand why. For instance, service with nil selector is supported, though it might not make sense to some users. see this:\nYes, it should be allowed, and I'm pretty sure it used to work. Could you please post your object schema here or somewhere?\nwhat do you mean by object schema? an example json of such pod or something broader? and btw what meaning a pod has without containers?\ncc\nI'm fine with banning this for now. The use cases where it would be useful (e.g., reserving resources, staging deployment, prefetching images, ...) aren't currently well supported. When we add the features that could take advantage of this (e.g., pod-level resources, additional of containers during update, container volumes, pod/volume init hooks), then we can make it legal again at that point.\nin that case, a clear error message coming from REST would be in place when a pod with no containers is recieved by the server :)\nI'll look into this. AFAIU we want to return meaningful error if user specifies pod with no containers. Is my understanding correct?\nCorrect. We should do it in validateContainers:\ncc\nshould we document this somewhere (e.g. ). Validation is not versioned, so any change would apply to all existing APIs.\nYes, it should be documented, in the field descriptions, such as:\nI'll send a PR for this.\ncc"}
{"_id":"q-en-kubernetes-8bb108261f9ea31f2e1442aeb830b4a2731925b51c70e95fc60efadb69675c81","text":"We use capitals for Pod fields (ex: Status.Host) and camelCase for Event fields (ex: involvedObject.resourceVersion). We should be consistent. Should it be capitals? camelCase? Or should we be case insensitive?\nIMO, we should use camelCase for pod fields, to match the way it's serialized in JSON.\nI agree with The goal here is to be consistent with the API, not with the internal Go representation.\nI'd like to fix this before v1beta3 is cast in stone.\nok. Event field selectors are correct then. I will update the selectors for pods to make them camelCase.\nI am wondering if I should change this only for v1beta3 or for 1 and 2 as well? This will be a breaking change for beta1 and 2 if we decide to do that.\nv1beta3 only, though that's hard, we should reconsider. We're contemplating another \"breakage day\", and I doubt this is widely used."}
{"_id":"q-en-kubernetes-f30562564cb56eba266b9f312736a73e9a741562f4b1725bf5d9e724708ac4cc","text":"As Satnam mentioned in PR , our Fluentd pod tends to not collect the first chunk of logs from each container. I've noticed that it seems to miss different amounts of logs depending on how quickly the container generates them, so it may be a timing issue. cc\nSee"}
{"_id":"q-en-kubernetes-055ba99112a20fcfdd2bfe16fad9fecdba21385acc682dc56530d866ddb24088","text":"Follow up from\nBumping this up to P1 given that 0.12.2 wasn't able to create a service with an external load balancer and we didn't notice before cutting the release.\nNo, to be clearer, 0.12.2 wasn't able to create a service with ELB on GKE. That was on our side entirely.\nBut it doesn't invalidate the need for this.\nThe example is using ELB.\nare you actively working on this? If not, I'd be up for doing it today.\nMaybe not actively but I know what to, however I won't be able to finish it today, so if you think you can do it today please move forward.\nWe run the e2e tests on GKE, so if we had a test for creating an ELB we would have known that the 0.12.2 release wouldn't work on GKE. I don't see how your point applies. The e2e tests are in place so that we don't have to test each feature manually to verify that it is working, and we only found out that this feature was broken on GKE after we pushed 0.12.2, hence the need for an e2e test.\nI think and you are in fact in agreement\nright now, k8petstore tests the external load balancer by curling down the amount of transactions ingested from the REST API. im playing with porting it as a e2e test now. The code in the shell script is a little raw, but you can see that it specifies a PUBLIC_IP and then curls down from it at the end ().\nthat sounds great. What I'd really like to see here though (and what is going to do early next week) is a very simple test that isolates creating a service with an ELB to verify that creating a load balancer works in isolation from a complex test of other parts of the system. This will give us a really clear signal when the test starts to fail instead of needing to debug lots of moving parts when we see a test failure."}
{"_id":"q-en-kubernetes-8b90e794a405b333388e6df8baae871c4efe69af9084fc8c039b202f7304993d","text":"We already \"healthcheck\" kubelet and docker periodically, but miss one for kube-proxy.\nIt would make sense to do (only bind to localhost) at the same time we add the healthchecking configuration."}
{"_id":"q-en-kubernetes-d1e28174cedbb4bdd2cadfc4485cd269e35cba0fa67cd75c639195b8163f8326","text":"stack trace here: it's due to a shared variable:\nHi thanks for reporting the issue with a patch! Do you want to submit a PR so that you can take the credit for your fix? :)\ndone , Yu-Ju Hong wrote: 585.241.9488 (voice) 650.649.6071 (fax)\nFYI, I think the same bug may still exist in your new PR.\nThere are multiple p\nThere are multiple PRs in flight, so I don't know which one you have in mind. One of them removes probing logic from controller altogether.\nI was referring to , but I didn't read that PR closely so the bug may longer exist there. Again, just a FYI. :)"}
{"_id":"q-en-kubernetes-1ff2227b578e0892bac06941b2b21a44da67a8fc722cc77d63c9a7ab123d81a8","text":"Appears to deadlock or otherwise hang after a few seconds and then hit the 5-minute timeout. From start of log: followed by stack traces of zillions of goroutines that mostly report being blocked for 4+ minutes. Recent instances: https://travis- https://travis- cc\nI tried to reproduce the issue by run kubelet unittests 200 times (100 for v1beta1, and 100 for v1beta3), but failed to reproduce the issue. I poked around, and found the only difference between my local machine and test machine is that I am running go 1.4.2 and test machine is running 1.3 (both failure cases posted by are with 1.3.0). I remembered from online I saw a deadlock issue reported against old golang version, but haven't found that issue yet. I will continue trying to reproduce the issue.\nSaw this in travis runs too: https://travis-\nYes, there have been multiple failures today. Check shippable, also.\nFWIW, these are not all kubelet tests failing, but all of them have many kubelet go routines in the trace - mostly stuck in managePodLoop().\nI reran a couple of thousands time last night before I leave, and all passed. Yu-Ju just pointed out that the failure posted above by is go1.4. Shippable is in a really bad shape today, tons of different tests are failed, which should be handled separately.\nOk, with some hacky debugging message into kubelet package, even I couldn't reproduce the very failure reported here, I made some progress on understanding the issue here. Basically there are two issues: 1) The flaky tests reported here are TestServeExecInContainer. All reports posted here have the same stack trace below: 2) There are tons of goroutines, especially one created for each pod. Kubelet leak those goroutines. One can see when test initial run, there are 106 goroutines. Test itself create another 41 goroutines, but none of them are created for podWorkder.\nFWIW, the tests in were creating go routines it didn't need at all. I just removed them in\nThanks for cleanup. Now the number down to 91:\ncc/\nDid something change related to this recently? I haven't heard any flakiness reports for a long time, and now they're starting up again. Any ideas? Sent from my iPhone\nI didn't see any changes related to this feature. Actually I even couldn't reproduce it on my desktop. Please note that the comment I made above all those leftover goroutines are separate issue thought.\nThis has exposed a deadlock in spdystream. I'll work with to come up with a fix.\nUpstream issue\nCool ncdc@ ! dchen1107 by the way, I've run across a few similar bugs recently where their reproducibility hinges mainly on how the test cluster was brought up. If the test cluster comes up right, then the tests pass 100% forever. If the cluster comes up bad, the tests fail, sometimes forever. Were you rerunning your tests against the same test cluster? Or rebuilding your test cluster between every test invocation?\nThis is unittests, not e2e. We saw test flakiness through travis and shippable, but never on my desktop.\nThis will hopefully fix the deadlock: I will do a bump commit once it's merged.\nThanks for your quick fix. I will clean up goroutines after that.\nit's entirely possible there's some bugs in my exec & port forwarding code around leaking goroutines. Are you going to take an initial look to see what's sticking around?\nYes, your test leak some goroutine, but other tests too."}
{"_id":"q-en-kubernetes-fd524e6826dc66b2f73439575ef5cd8d63e3e397dffa3cbe39f71d1f2888b60e","text":"goroutine 11 [running]: net/http.(ServeMux).Handle(0xc20800a8a0, 0x9e9490, 0xd, 0x7f7d12f190a0, 0xb0e7c0) +0x244 net/http.(ServeMux).HandleFunc(0xc20800a8a0, 0x9e9490, 0xd, 0xb0e7c0) +0x6d net/http.HandleFunc(0x9e9490, 0xd, 0xb0e7c0) +0x48 ·001() +0x58 created by (*SchedulerServer).Run +0x1c7\ncc"}
{"_id":"q-en-kubernetes-859a8b847337ca2cf9879726f3bdb71f0db5559ee2589591ceae612e41f71a1b","text":"There is a funny subtle behvaiour in get namespaces, in which if i call more than one time, it seems that the removal from Terminating state is accelarated. It seems this might indicate a bug, or maybe its not a bug and the eventual semantics of is a little funny sometimes. in any case, here's what we are seeing : 1) when i delete a namespace the first time usning kubectl, we get a state change to . that is correct, of course. 2) but ... if make the call a second time, then the same status line is immediately removed from . Expected behaviour : A second call to \"kubectl delete ...\" shouldn't have any effect on the timing of removal of a namespace from 'Terminating' state*... So i guess there is some \"eventual consistency\" subtle bug that is resolved by repeated calls to\nsuggests this may not be e bug, maybe its just the etcd eventual deleter thingy.\nLooks to definitely be a bug. The 2nd delete seems to have interrupted the delete process and artifacts (like events) from the namespace still exist in etcd.\ninvestigating...\noops, see the error, will submit a fix.\nI have a fix, but need a test case, will submit PR tomorrow AM"}
{"_id":"q-en-kubernetes-0c0bb3600e99710565f99e8be4bd669de365f6aed46898aff519521ca79dfa0d","text":"GCE has a length limit of 63 characters for its load balancer components (target pools and forwarding rules), and AWS has a length limit of 32 characters. Our current load balancer name construction of clustername-namespace-servicename can exceed both limits, especially AWS's. We should use a shorter name construction method less likely to hit these limits.\nThanks for filing this Alex. My suggestion would be to use a simple UID, as per utils.NewUID(). We'll need to store the mapping from {clustername, namespace, servicename} -UID somewhere stable like etcd. Using anything based directly on {cluster, namespace, name}, for example a stable hash of each, has the problem that the aforementioned tuple is not unique over time. This raises the possibility of e.g. the following scenario: creates mycluster/mynamespace/myservicename, which creates external load balancer named stablehash(mycluster,mynamespace,myservicename) deletes mycluster/mynamespace/myservicename, but at the time MyCloudProvider is down, so kubernetes can't delete ELB named stablehash(mycluster,mynamespace,myservicename). Kubernetes keeps trying in the background to delete the ELB. the mean time user creates a new service named the same as her old one mycluster/mynamespace/myservicename. Kubernetes tries to create an associated ELB named stablehash(mycluster,mynamespace,myservicename), but this fails, because the ELB exists already due to 2 above. ... etc. Badness ensues.\nAssigning to Chao Xu, as this makes for a good starter bug.\nHmmmm.... can't seem to assign to caesarxuchao yet. Not sure why.\nHe has to have write privileges on the repo before issues/PRs can be assigned to him.\nBut we have many contributors with issues assigned to them that are not in the kubernetes-write group (that only has 48 members). I must be missing something. How do I grant him write privileges on the repo?\nWe do? That's not how I understand it to work: If you want to add him, go .\nhas been because the load balancer name changes if the service's UID changes (e.g. if it's deleted and recreated). Think we can change this? A hash of cluster name, namespace, and service name should get around it, but there may be better ideas as well.\nSure. I'm not confident that I can fix this quickly. If it's urgent, maybe you can assign some one more experienced to fix it? One question: I guess the ELB's name should stay unchanged as long as the \"zone+namespace+service name\" doesn't change, am I right?\nI think I'm the guilty party here. I was unaware that when we create a new service with the same name as a previously existing service, that it needed to keep the same load balancer name and IP. This seems broken to me, but I may be missing the larger context. We can quite safely roll back the change while we decide on the best alternative solution. Q , Chao Xu wrote:\nI think the real breakage is that we don't have any way to claim an LB-IP at a larger lifespan than a single service. Alex, can you go back to OP and get a sense of what's happening there to trigger this? , Quinton Hoole wrote:\nWill do. Even if the name is kept consistent as it was before, with the way everything currently works the IP won't be consistent.\nyeah, I'm not positive that the name has to be consistent. What we need is a way to claim the same public IP address, and we already have that (at least for GCE) Assigning to me, and I'll validate the length requirements, make sure public ip assignment works for AWS and close out this issue.\nNote there are two separate things here: (1) a purely-name-based construction can exceed length limits given a single service and (2) some form of across-service consistency is apparently needed. (1) is trivially solvable by either always using a hash or alternatively using a two-stage process where you generate a desired name and only hashify if it's longer than the length limits by truncating it and appending a hash of said name.\nAWS's load balancer hasn't been merged yet, so there's not much to check out there. I'm going to be reducing the churn of load balancers later today, which should help prevent IP address changes for services that are modified.\nI think we need to be able to figure out which Service an LB is attached to from the Service, so that LBs can't get orphaned. regarding being able to claim an LB IP. I don't think that is covered in revamp... , Brendan Burns wrote:\nI think that we need to roll back and think this issue through a bit better. See comment on that PR."}
{"_id":"q-en-kubernetes-d40e4a6c562fc50ef12d5d4b4bfe6a402a7cb5ec672dfc45502a752fbe0291f6","text":"Rackspace support has not ben updated in a few months. cluster/rackspace/cloud-config/minion-cloud- is using etcd_servers for kubelet and kube-proxy, but support for them has been removed since v0.9, I think. I'll remove it in a few days if no-one makes a counter-offer to get the config working with latest release.\nThere's been some work on this but it is blocked while we figure out some load-balancer design stuff. , Eric Tune wrote:\nOkay, I will remove the link to rackspace until it is fixed (). Leaving this issue open to remind us to delete it in, say, a month, if no progress is made.\ncc\nSee also\nI'd prefer to try to get things updated and keep support. However I don't want to continually become stale. every few months either.\nHi all, I want to use kubernetes on my existing rackspace infrastructure but I don't understand if rackspace suppor was removed or not. In Doc, under \"custom cloud solution\" it's linked but I wasn't able to deploy it neither with install method nor with . With i got a error (I have 16GB or RAM!). thanks in advance, Fx\nIt looks like no support has happened on Rackspace for a year.\nYou should not expect it to work, and we should delete rackspace support.\nACK.\nUnfortunately I think we'll need to drop support for Rackspace as a deployment target since I can't commit to keeping the deployment scripts up to date. Rackspace support as a cloudprovider should be maintained.\nI would be willing to support the Rackspace deployment target, if there is interest to keep it alive.\nThere would be definitely an interest from our side - sad to see that Rackspace has no interest in maintaining it\nNo problem, I will take over the maintenance. It will take me a bit of time to get caught up, but I will get it done as soon as possible.\nDo you think it would be hard to get the of LBaaS working ?\nSorry it took me a bit to respond. there is a quick update as to where I am. I was busy with some stuff, that prevented me from working on this. But that has been resolved, and should have updates by the end of next week. However what's causing the delays is the lack of a good API for Rackspace Cloud. I've used the gophercloud, it works but I'm not a fan. Additionally gophercloud doesn't have full API support (one of the things I'm not a fan). This would make it hard to implement LBaaS. I should have a better idea of your request sometime next week.\nI have seen this api docu: - is it the same as you saw? I think it should be doable with it."}
{"_id":"q-en-kubernetes-3e41c7a0a85635ba91742439b913fa4ee706ea1bfde3bc5546521b3318c03960","text":"If another process is altering iptables rules at the time k8s attempts to add iptables rules it will fail with a locking error. k8s should use -w and retry after a delay. Below, an external process (openshift-sdn-node) modifies iptables at roughly the same time and k8s fails without retrying. Related to\nSorry, I misunderstood the -w flag, with it retry logic wouldn't be necessary as it'd wait. Perhaps simply retrying is a better solution as -w flag may not exist everywhere we'd like.\nLooks like -w was to iptables 1.4.20(August 2013).\nYeah, we have to be pretty back-compatible with iptables. We always retry after a few seconds - is there really a bug here? , Scott Dodson wrote:\nIt definitely seems like there is some sort of a problem. We see this randomly on system startup. In our environment OpenShift is launching kubernetes around the same time other systemd units are modifying the firewall config. If one of these others units has the lock kubernetes will report the error Scott mentions and then will never create the required nat'ing for services. Here's how to reproduce this: At this point if you run you will see the exact same error kubernetes reports. It's probably not obvious why you have to set the breakpoint at 257 but if you set the breakpoint at 260 and then step through the code it appears that the compiler has optimized lines 263 and 257 to be the same. In any case running will verify if you have the lock or not. Once you've verified you have the lock just restart OpenShift (running kubernetes v0.14.1-582-gb12d75d): At this point you can press in gdb to continue the iptables invocation and release the lock. Afterwards I will see output from cadvisor every once in a while, however the will show the firewall is still blank: I initially restarted Kubernetes with a blank firewall at 11:21 and yet at 11:38 I still have nothing:\nI probably should have made it more clear, I removed the lock by pressing 'c' in gdb to continue. That's when I started waiting. I'll update my comment to make that more clear.\nSo something for a holding the lock? I am out of your office for the rest of this week, but maybe you can figure out - is there a stale process holding a lock? What kind of lock is it? Flock or semaphore or ... ? On Apr 29, 2015 8:44 AM, \"Brenton Leanhardt\" wrote:\nWhat we're saying is that on startup it's fairly common for multiple systemd units to make firewall additions. If one of them grabs the lock (rarely for more than a few seconds) it will cause kubernetes to fail to initialize iptables and it will never fix itself. The lock only need be grabbed the microsecond kubernetes tries to initialize iptables for this bug to present itself. We know exactly which program in our environment was concurrently making iptables updates. The problem is that kubernetes doesn't appear to be retrying if it fails to initialize iptables.\nOn RHEL the lock is the xtables unix socket lock in iptables to allow multiple programs to update iptables concurrently. Other platforms may use a flock since that is the default upstream. To solve this kubernetes needs one more more of the following: pass the flag to as scott and the logs mention (possibly only available on some platforms). If kubernetes desides to wait for the lock it will need a timeout. retry the firewall initialization quickly and maybe backoff if it still can't get the lock It's probably worth noting that if iptables on a platform doesn't have iptables lock support concurrency is likely broken already in other ways for that platform.\nIt does retry all up tables ops. Unless it somehow gets pathologically phase locked with your agent, or there is some deeper bug. We have e2e tests that prove you can delete tables and it will recreate them. Are you saying that it never retries (did you set -v=4?) or that every retry fails? On RHEL the lock is the xtables unix socket lock in iptables to allow multiple programs to update iptables concurrently. Other platforms may use a flock since that is the default upstream. To solve this kubernetes needs one more more of the following: pass the -w flag to iptables as scott and the logs mention (possibly only available on some platforms). If kubernetes desides to wait for the lock it will need a timeout. retry the firewall initialization quickly and maybe backoff if it still can't get the lock It's probably worth noting that if iptables on a platform doesn't have iptables lock support concurrency is likely broken already in other ways for that platform. — Reply to this email directly or view it on GitHub .\nI can grab the lock for just the second that kubernetes needs it and then let it go. The kube-proxy nat'ing will never be rebuilt without intervention (restarting kubernetes). That seems like a bug to me and is what my original comment reproduces. In addition, once the environment is in this state it won't even pick up new service nat rules. Other nodes in my environment will pick them up but this one is dead. I highly suspect it has to do with the fact that iptables initialization is failing. We test all the time that once a system is up and running you can clean iptables and it will come back. This is a slightly different case.\nThis seems like a v1 blocker - we see this in every realistic environment we put the Kube-proxy in. Disagreement?\nI still don't understand this issue. It is demonstrably true that the kube-proxy retries all of the iptables logic periodically. It's possible (given the mention of \"won't even pick up new service nat rules\") that this locking tickles a bug in kube-proxy, but I am having trouble reproducing it. Can you run your kube-proxy with -v=4, trigger the bug, and then take a look at the logs? If this is real, yeah it's a 1.0 blocker.\nThe problem is in the initialization process. If you hold the lock while the kube-proxy tries to initialize itself then it will be wedged forever. What you're showing is the kube-proxy can repair the iptables rules after the initialization code has run. The logs I pasted earlier show that after the initialization fails at it will never try again. I tend to run at loglevel 4 by default but I'll try to find time today to run through this again.\nI have a fix coming - can you test the PR before commit?\nI ran the exact same test as mentioned in my original comment. While the debug lines have changed the end result is the same: grab the lock (at this point even will fail) start the kube-proxy release the lock (at this point works) wait.. try creating another service wait.. The rules never come back. Here are the logs:\nSure, I'll test the PR. I'm betting it will apply fairly cleanly on to my OpenShift codebase. If not I'll launch Kubernetes directly.\nMany of those log lines are not part of upstream. In fact, what upstream has is a glog.Fatalf() which will kill kube-proxy, and when it restarts it will try again. Did you guys break it by not actually exiting?\nLooks like we did. We need to put it into a restart loop. ----- Original Message -----\nis teh fix, though I assert we should not need it, and if you need it this fix may not help you anyway. I am still OK to commit this.\nAlthough NewProxier sucks because it doesn't return a usable error. I'll take the todo to clean up that part. ----- Original Message -----\nhey, it logs its error - nothing wrong with that :) , Clayton Coleman wrote:\n:)\nOpened and will move the rest back to openshift.\nI restarted docker and it worked then"}
{"_id":"q-en-kubernetes-83d108f0f5dc476eb3ccc1bd3ac7ca0c88769da6703901d0bcbe64c096876bb1","text":"Following on from the Kubelet should accept YAML pod specifications. Currnelty it supports only JSON pod specifications."}
{"_id":"q-en-kubernetes-528257a8d86f973b92cc017b02ead0ef3ea1934cb7f6381428f555668a8c1a21","text":"As the title suggests, I can find no documentation on KUBEADMISSIONCONTROL (from the apiserver configuration), and I don't know golang well enough to determine the answer (yet) from the source code. Google search yields only 3 hits today, and all of them are copies of the configuration file itself with no commentary. As an admin, I'd be grateful for simple documentation on all the variables, their possible settings, and what those really mean. If that already exists, please identify where it exists. Though my formal issue is the documentation, the problem that leads me here is that my pods can't query the apiserver. Given the following, my connection is reset: Any assistance would be greatly appreciated, though my formal issue remains the lack of thorough documentation on the configuration. I'll continue learning golang in order to try to figure out what my problem may be.\nI agree a document is needed, will look to put something together describing what each option does.\nThanks! :)\nThat being said, I recommend we scope this issue to just documentation being needed, and if you need assistance with your other topic, treat it as a separate issue.\nQuite reasonable! If you think that documentation will be before the weekend, I'll wait until that's available and see if I can figure it out myself; otherwise, I'll make a separate issue for that."}
{"_id":"q-en-kubernetes-5cad7f0b9d7423b15ef1aff818cb7d7539e6f84907a939c490fb9042ee93c984","text":"We seem to have an occasional consistency issue between the client and server versions on our Jenkins GCE e2e project. Build 6295 failed this morning because the client version was newer than the server version: The server version matches the build was was immediately running in 6294, which was aborted early. I wonder if (somehow) the cleanup didn't work properly? (Which is a bit worrying if so.) Here's the version from 6294: A short while later, build 6297 failed, this time with the client version older than the server version: This one makes a little bit more sense: there was a new build published between downloading the client/test tarballs and setting up the cluster. I'm guessing that somewhere in the cluster setup scripts the latest CI release file was read again, rather than using the version read for the client/test tarballs. (I haven't yet worked out where exactly it's read, but it should be an easier fix.) cc\nA side issue: we should probably figure out why the build git client is being marked dirty. Since we're checking out the head version from GitHub, it should ideally be a clean client.\nThe -dirty bit is now fixed by .\nThis is caused by different Jenkins jobs uploading to the same GCS bucket and clobbering each other. In we see: Over in\nThis also explains the issues seen in ; the job is also running in the same project.\nAh, good sleuthing. We avoided this issue in the PR builds by setting (used ), but maybe we should always be setting to include the Jenkins job name (and optionally executor/node number).\nThe original idea was to run every e2e job in it's own project, to avoid these sorts of name clashes. But we seem to have regressed and now have many jobs sharing the same projects. Is project-per-job not a better approach? It also helps with resource leaks and the likes. Just a thought.\nI'd argue this is actually uncovering a (dev) deployment bug; I should be able to deploy different versions of Kubernetes to the same project. I know you and I disagree (and probably always will) on the strategy here. :smile:\nfwiw I think I agree with that if project-per-job avoids this issue and helps identify resource leaks, it's a pragmatic near-term improvement in the state of the world that would result in more benefit than the cost of changing all the everythings to support multi-job-in-project. Re: devs deploying different versions of k8s to the same project, P3 feature request?\nI was just about to suggest the same thing. I will file a separate issue, and fix this in the short term an easier way. is out to make it easier to isolate a run to a project.\nAlso it seems like a reasonable thing (higher than P2) to want to deploy multiple clusters to the same project safely in prod, not just dev. Disagree?\nMultiple clusters in prod using well-defined build versions such as v1.1.4 = P0. Multiple clusters in prod using our cooked-up test-building system = non-goal?\nSGTM; this is prod code, though, not test code; it's used every time we call , regardless of whether it's a CI build or released version.\nFYI: kubernetes-e2e-gce-reboot/ has one reboot test failed due to kube-proxy image change. But that build doesn't include pr yet.\nlooks like it. That run used version , but the first build with was ."}
{"_id":"q-en-kubernetes-3aad09bd638f4d0e6738065aebd7ae97fbb23d73b8dd9d1472c634994380848d","text":"We need to be squelching the auth variables on upgrade anyways, but this complaint gets issued (and it proceeds anyways): cc"}
{"_id":"q-en-kubernetes-661345ebb8c9ce30f7455d4bf53da418af157a4bd40a657033e168d627bd4150","text":"In case of cluster crash and status recovery, the persistentVolumeClaimBinder would not re-add available volumes to its internal index. Users would not be able to claim those volumes.\nThis is really bad. I think we should just nix Pending phase and always consider every Available volume for indexing - map lookups are cheap."}
{"_id":"q-en-kubernetes-a52b59c7d2fad26006f4e403c6ccd7d769148b1428731cb4cfa9b4cdf94eb1c4","text":"The \"\" E2E test will sometimes fail because it fails to delete the PD used in the test after the test is finished. The error: Detailed test log: The disk that fails to delete still shows up in : My attempts to manually try to delete the PD fail with: (the work around for ) doesn't change anything here:\nis the disk still mounted ? It looks like the detach call did not succeed for some reason\nI'll have to get another repro and verify.\nokay cool. If you get a repro check to see if the disk is still mounted. If so then the cleanup code is not being called. If not then the clean up started, detach was called but failed for some reason; look in the logs for something like \"googleapi: some error about detaching\". It can also just be a race condition. That is, detach was called but is taking too long; if the disk does not detach then you cannot delete it."}
{"_id":"q-en-kubernetes-4a5f8a117915b8e2720ab3fe6bd6b6e9a9116553d4b98258a619803800705175","text":"If the container_probe test runs on a node without the nginx image, it sometimes takes 90s to get the pod in running. Most of this time goes in pulling the image, so I'm assuming it's registry related hiccup. Using might make it more reliable. Example failure: Note the following event in the logs:\na) we should not be pulling from docker in our tests - too hit-or-miss b) we should replace nginx (132 MB) with (4.5 MB) whenever possible. , Prashanth B wrote:"}
{"_id":"q-en-kubernetes-e173b509db75d449d82cb27f028365138ce8184a0f33391f61ddb69043f21d65","text":"I'm conducting PoC about distributed build using Kubernetes. Jenkins master distribute jenkins slave pods through Kubernetes API Server. When build is done, jenkins slave pod is also deleted by kubernetes API Server. But, kubernetes didn't delete container. (Pod is deleted well) After searching, I found kubelet’s containerGC option. So, I adjusted below options to kubelet. --minimum-container-ttl-duration=0s --maximum-dead-containers=0 After adjusting both 2 options, container was deleted, too. But, kubernetes containerGC didn’t delete host volume path. (/var/lib/docker/vfs/dir/..) Due to aufs’s low performance, I used data volume of host instead of container inside. So, data volume was accumulated and host's inodes was fulled, too. Is there any solution to clean the containers wreckage? My environment was.. Docker: v1.6.2 Kubernetes: v0.18.2 (container based multi clustering)\nare you using the host volumes you created yourself? If this is the case, kubelet does not clean the volumes since they are external to kubelet. If not, kubelet should clean the volumes before container GC kicks in.\nNo.. volumes were created by kubenetes API server. LGTM. Will it be adjusted next release?\nYou said \"I used data volume of host instead of container inside\" which says you used a hostPath mount - correct? That is explicitly NOT managed by kubernetes - you've gone to the host, you get to manage it. But you also said \"Due to aufs’s low performance\" - an emptyDir volume is NOT stored on aufs, but it IS managed with the pod. , gghonor wrote:\nThe issue as i understand it is dangling volumes, that is volumes that are created via a VOLUME directive in Dockerfile with no corresponding kubernetes volume in spec which therefore has no host path & is not cleaned up by the kubelet. These volumes reside in /var/lib/docker/vfs/. This waste of disk space has the potential to cause out of space errors on the node & therefore make the node unstable.\nAs the volume dir you mentioned is I assume you didn't specify any volume in your Kubernetes pod spec but did use the directive in your Dockerfile? As mentions I guess this isn't a Kubernetes managed volume as that would be in the Kubernetes volume dir, right? That dir is used by Docker for volumes specified in Dockerfiles or a docker container that is run with with no specified host path. The latter doesn't happen with Kubernetes managed containers AFAIK.\nThis makes sense. Thanks for the explanation :) I see the point of removing the dangling volumes. Even though kubelet doesn't manage those volumes, it doesn't hurt to remove them.\nDo we have a full understanding of their lifetime, though? We don't control them, I don't see how we can know when to delete them. Docker's volume implementation is just too under-defined to actually be useful. , Yu-Ju Hong wrote:"}
{"_id":"q-en-kubernetes-5e141256b52005c28fd1b24c7f05dfac8f617bee1bc04c7a011b4ab9fa62a6c6","text":"The file: has instructions for AWS, GCE, and VMware. I'd like to see documentation for OpenStack as well. I'd be happy to do the work for this as I don't think it would take too much time. I'll add some instructions to this issue if its something that people think would be helpful.\nI would greatly appreciate any work you can do with this. I've been trying, unsuccessfully, to get the OpenStack integration working.\nOk, I created instructions in this gist: If it looks good to you I'll create a pull request and merge changes in the document after the GCE section (as it appears we're going in Alphabetical order? )\nSending a PR sounds good."}
{"_id":"q-en-kubernetes-0899495702d6e9ebf67f2fa89ae62937548eeddaa12f77ce123a9a812c22bf72","text":"If is passed to the kubelet, container metrics are not included in the Prometheus metrics endpoint."}
{"_id":"q-en-kubernetes-0f36d617029b75ecdfb03cc044087fba40537f8a434f7633b301501da951ce4d","text":"The first line logged from the integration test is the last line from the service account token test is then the test panics with: I think the tests are just running too close the the timeout. On successful runs, the tests take ~115 seconds, which is ridiculously close to the threshold. For example: see\nok, do you want to send a PR to increase the timeout to 3 or 4 minutes?\nSure, I can do that tonight"}
{"_id":"q-en-kubernetes-ae486ffe03a734aa443d42b073f2c6b23bd4d201c949d5702210dc739a9b466f","text":"Nodes Network when a minion node becomes unreachable [replication controller] recreates pods scheduled on the unreachable minion node AND allows scheduling of pods on a minion after it rejoins the cluster(Failed 17 times in the last 30 runs. Stability: 43 %)Failed with 0 retries\nfor triage\nI'll look into this.\nI talked to about this and he had a pretty good guess in about two seconds about what the problem is. ReplicationController sync method does this: but DaemonSet controller doesn't. (I guess when DaemonSet was copy-pasted from ReplicationController, this was accidentally removed.) Pod reflector and syncer are running in separate goroutines, and if pod reflector doesn't run then syncer will think there are no pods on the node.\nTo be clear: this is about the test Controller Manager should not create/delete replicas across restart which is the one that is actually failing.\nI'm confused you're saying that this is about , but you're updating . I thought they are about two different tests...\nYes sorry, I put my comment in the wrong issue. Reassigning this one to you, and I'll copy my comment into the other one.\nFirst failure: We check if all namespace are deleted. If other tests failed to delete their namespace this test will fail. I'm going to delete this logic as we already check if we have successfully deleted our namespace. See also for some more details.\nSecond failure: We disable network by running the following command: I checked and it seems that it results in ing master to hang. I suspect that Kubelet hangs on updating node status and depending when it breaks the connection (by timeout) it will fit in test timeouts or not.\nWhat is the default timeout for connections using client libraries to apiserver? I'm checking the code, but can't find it.\nI'm not sure if we set one, but we should. I guess I would expect to see that in the config, w/ default set by this function: , Filip Grzadkowski wrote:\nSo: let's add a timeout, and revert .\nI think that adding a timeout and reverting are orthogonal. I we want to simulate network problems with node-master communication then using REJECT is better anyway. At the same time adding timeout makes sense for many other reasons.\nOK then, let's add a timeout and test both DROP and REJECT-- no sort of network problem should hang a component.\nI think it didn't hang. It just that the kubelet was retrying too slow and sometimes it fit test timeout, sometime it didn't.\nCould we reduce the timeout to make it always fit the test timeout? , Filip Grzadkowski\nI guess we can. I'll try tomorrow.\nIt seems we have a new type of failure: It seems that kubelet on a machine where network is blocked doesn't notice new pod. After reading logs I assume that kubelet doesn't notice that the watch is broken and doesn't try to relist everything. It notices new pod only after watch timeout.\nUgh, I hope we don't have to start sending watch heartbeats.\nIt seems that all recent failures are with the following error: which means the reason is here: - can you please take a look or delegate?\nI did some more debugging of it. I was looking into this failure: There is something strange happening here: run performTemporaryNetworkFailure: succesfully block the load from kubelet: the succesfully wait for node to become not ready: some time nodecontroller successfully removes the pod: replication controller creates a new pod: here is something I don't undertand - because of some reason, the test doesn't observe that the pod actually disappeared. - any thoughts on it?\nBTW - this is exactly the same situation in:\nOK - I think this is related to default 30s deletion grace period.\nYes - it seems that this is exactly the case. I will send out PR fixing this out for review.\nI'm reopening it for tracking. If there won't be any failures within next few days - I will move the test back from flaky and close this issue.\nThanks for fixing, .\nIt seems that the main issue is fixed. However, there was one failure in the last 30 runs: in which one pod simply didn't started (and it was \"preparation\" for the test, not the test itself).\nDuring next 2 days there was exactly one more failure of this test: it is exactly the same as the previous one. However, those are duplicates of So I'm going to move this test out of flaky suite."}
{"_id":"q-en-kubernetes-13f89768690da6fc511a0bb03e5ffac897995a7d8142248427cada30c91df786","text":"runs foo and then waits for it to be ready or terminated. If the pod is terminated, it prints the log, and if it's not terminated, it attaches the output. However, it does not handle the case that the pod terminates between returning the ready status and attaching, as can be the case in the : Options to fix this: Fix kubectl: If attach fails, fall back to printing the logs If attach fails, check the status again and if it's terminated, print the logs Fix the flaky test: Prevent the container from exiting (e.g. `sh -c 'echo running in container && sleep 30') Prevent the container from entering ready state (e.g. )\nThanks for debugging. I think we should probably do both: If attach fails and the container is now terminated, fall back to printing logs Make the e2e more robust by not exiting for a while to ensure time to attach. I'm happy to review PRs"}
{"_id":"q-en-kubernetes-9b5fe03698914c610d32064544928591f0a004f7e1f0038b0626d93de3549b78","text":"I'm seeing a bug where the network plugin fails to configure networking, and yet the containers for this pod are still deployed (without networking configured). It looks like the following is occurring: 1) Kubelet attempts to create infra container in 2) Network plugin fails, returns error code. 3) Kubelet detects error and skips pod creation but does not tear down the infra container 4) On the next call, the kubelet detects that the infra container is running and deploys the other containers in the pod. The fix here is for the kubelet to clean up the infra container in when it receives the error from the network plugin and before it returns. Here are some logs illustrating the steps I described above:\nCleaning up infra container on network plugin error sounds reasonable/"}
{"_id":"q-en-kubernetes-bc481a4627aea373f2de65a2d965f4d31ac5088d41d6a75d1a607aee3be72971","text":"Unit tests on our CI have apparently been failing for 2 weeks and no one noticed (oops!)... The PR that seems to have caused the failures is . The failing test is - . Full Test Log: I'm not sure why this is failing on our CI and not on shippable. Has anyone else seen something similar? That test only seems to fail on the second run through with . Our unit tests run in a container built specifically for testing k8s-mesos: The Dockerfile for the env: Is anyone else able to repro with the above env or their own?\n/cc\nthat test was around a week ago, it's doesn't seem possible to me that it was failing since 2 weeks ago. Anyway, we have few duplications of this issue: this one And the PR where is working on: It's a little bit difficult to me to follow all the conversations in all the differente places about the same bug, could it be possible to unificate some? Since you provided the environment I will try and take a look to reproduce it locally. Thanks!\nI think I've got a working patch in\nClosing in favor of &"}
{"_id":"q-en-kubernetes-5d8539be161cacf223528437869590f13423a31b1176c381211bcba6b79e0cca","text":"When trying to edit multiple resources, for example: will panic. Ref as the possible cause cc\nif was the cause, any fix needs to happen in 1.1 as well, was already cherry picked there\nAlso shows that we need a test for this case.\nis the cause. When we edit multiple resources (i.e., list-type objects), the list's Accessor.annotations (type ) is nil and will panic when UpdateApplyAnnotation tries to SetAnnotations ()."}
{"_id":"q-en-kubernetes-fc39169146c9455c303f0970f8a8e8f06dd10bfe4e4173704fe0272d20d254f8","text":"We have to spend times on figuring out what are the mount error code means from issues and . In , the error message is and is great for diagnosing mount problem. The console message better be returned rather than just error code."}
{"_id":"q-en-kubernetes-42a2dcc81f7ae4dc8763c9cd41f1b3c90a02e5ad103b4424eb3978e5cae8304c","text":"Heapster pod: And logs: cc p0 for triage\nDid the required node scope change?\nIts either the API on the console or monitoring write scopes on the node. , Mike Danese wrote:\nI think this affects only GCE\nI will fix this issue since is on vacation.\nNeed to cherry pick so i'll leave open\nis out for review.\nhas been closed and the cherrypick here () has been merged so I'm closing this issue. Thanks everyone for the debugging."}
{"_id":"q-en-kubernetes-f4ab746f40c555f197660ba2d46053a482240358ec0954f57c2d8c122630e391","text":"has TODOs for how to bring up cluster addons like monitoring and logging. We should fill these in. requested by user:"}
{"_id":"q-en-kubernetes-cc8f7dd240ce03e902a3a6988aef93cbe00593e0f5081bd8380b06a0a728c5de","text":"Currently watches only log when they complete, which can make analyzing kube-apiserver log files a bit tricky. (did component x never start watching after that list or did apiserver shut down?)"}
{"_id":"q-en-kubernetes-1b90ac0e27eb7112950da48e761bdf56abe3dcf0ad8451e1546669339b5f8d17","text":"We have found that on 1k nodes cluster scheduling takes 60-200 ms and becomes the bottleneck. We want to have a standard way to reproduce scheduling latency metrics result and benchmark scheduler as simple and fast as possible. Only with that can we: 1. figure out underlying behaviors and potential bottlenecks quickly; 2. publish results in a standard way and prevent future performance degradation. We have the following goals: Goal 1: Save time on testing. More accurate scheduling latency. We mock up kubelets and only measure scheduling latency and do not care about the latency to start the pod. This is different from Kubemark -- scheduled pods won’t be run by mocked kubelets. We only need to set up the master components. Goal 2: Profiling runtime metrics and find out performance bottleneck. Write scheduler integration test. Take advantage of go test and profiling tools, like cpu-profiling, memory-profiling and block-profiling. This is very fine-grained metrics. Goal 3: Reproduce test result easily. We want to have a known place to do the performance related test for scheduler. Developers should just run one script to collect all the information they need. We have the following plans for writing scheduler tests: density test (by adding a new Go test) schedule 30k pods on 1000 (fake) nodes and 3k pods on 100 (fake) nodes print out scheduling rate every second let you learn the rate changes vs number of scheduled pods benchmark (by modifying the existing benchmark in integration or simply add a new one) make use of and report nanosecond/op. schedule b.N pods when the cluster has N nodes and P scheduled pods. Since it takes relatively long time to finish one round, b.N is small -- level of 10. We are new to k8s codebase. We also want to know where is the best place (directory) to commit in these tests? Any suggestions are welcomed!\n/cc\nThis extends\nSo density already does this... the one option it looks like you may want is custom control , real or fake is irrelevant as the e2e's just need to point to an api-server. re: benchmarks, seems like we could simply add to the unit-tests.\nAfter chatting with lets add this as an agenda item for the SIG meeting tomorrow.\n+1 for adding benchmark (howeveer this should be integration test, not unit-test in my opinion)\ncc The next SIG on scheduler meeting is on 12.07, isn't it? If there is one tomorrow, can I have the information please?\nretract, +1 to . I was adding this to the list given the context, but it crosses over into both.\nSlides from today's meeting:\nHere are are the action item notes from the SIG meeting: Everyone agrees this is a good idea and long overdue, but the question is where would this live: New category of tests? (isolated/sandbox’d/component) The current testing structure allows to easily append. create new directory (component) + hack/script to execute Future \"we could\" gate on known good data, and fail performance regressions\n/cc\n/cc"}
{"_id":"q-en-kubernetes-8a1317e390e9571143fd6d82d56b72619936bbd931877828728e077c4c54d6b4","text":"I'm running application with several env variables. One of them has a key ending with ==. Another variable has a comma separated list of email addresses. While perfectly valid on *nix and windows, we can't pass these values into Kubernetes run command. It fails with \"error: invalid env:\" Example command: The parsing in should be more robust\nI volunteer myself to fix this.\nMaybe they can be passed from file? I mean s'rsly.. you need to lits 20 variables by hand?? I've specified them in yaml file already, but kubectl-run have no way to specify it.. without this, deployment is totally impossible when you need to run migrations first!"}
{"_id":"q-en-kubernetes-716ca225a252153ae82ca602b3860ed212a55f6220331030249e6fc7ae53f83d","text":"Have not reproduced locally even after a few tens of runs.\nwhich version of Go are you using locally? This probably only repros on 1.5.1.\nThis was from Jenkins runs.\nThis is caused because an attempts counter is being read/written to without synchronization. I think this is caused by the event creation order not being guaranteed. I am looking a little deeper to track this down.\nGot a fix with 130k+ runs and no failures. Will send out PR shortly."}
{"_id":"q-en-kubernetes-e505295a05058bec1cba89d2e6a4b223262484cf137dc1e912f1c7462280605a","text":"Currently scheduler latency are observed regardless of failure. For example, binding latency is like: This could be a potential confusion if latency spikes happen because of failure. We should better handle it like separating error case from normal\n/cc\nI completely agree - are you going to change this?\nYes. I will do it."}
{"_id":"q-en-kubernetes-4030aa11852f25ec651e1fc5dd667b4297fe2300b704273273cd2ebda6e02fe8","text":"I'll take a look at the failed builds and triage.\nI went through the recent reboot failures in kubernetes-e2e-gke-ci-reboot and here is a summary. [6481, 6486, 6488] Kubelet never came back from the reboot, and stopped posting node status. Throughout the test suite, the failed node stayed silent: It's hard to dig further due to the lack of clarity. Gathering log would at least help us rule out some possible causes . [6485] Heapster seems to crash loop quite frequently, and sometimes it failed the test (due to being not ready).. cc [6487] kube-proxy pod is running but not ready before the reboot test even started: This is not really a reboot test failure since the cluster was already in a bad state. We should make sure all the pods are ready after each test (e.g. \"should work after restarting kube-proxy\" ran before the failed reboot test). Another thing that would help is to increase the log level of the gke cluster. The log level of a gce cluster is 4, but I think gke is at the regular 2. Due to the low log level, it's unclear when kube-proxy started failing the readiness probe.\nFiled for (3).\nThe reboot tests basically haven't worked since 9pm last night: Interestingly, this lines up with the GKE equivalent for , but in that same window, has had no trouble. cc\nDigging into logs from the runs linked.\nAll of the gke-ci failures in are heapster crash-looping after a reboot.\nChanges to the heapster config in the last day: , . Nothing sticks out to me in those PRs.\nHeapster could be in crashloop if it cannot reach influxdb. But that is a bad design, and should fix it in today's heapster with image 18. I double checked them, it is image version 18.\nFrom , could be the num_nodes is unset, we give heapster no resource to run? If that is true, it could be point to the same version screw issue of our test infra.\nHeapster v0.18.4 will crashloop if InfluxDB isn't available. However we can update HEAD to v0.19.0, which will not crash if the backend goes down. I will send a PR to update InfluxDB version. , Dawn Chen wrote:\nGKE doesn't use InfluxDB, right?\nHeapster is the only kube-system pod that's not ready in the tests. If it relies on influxdb, shouldn't influxdb be not ready too?\nI like theory. My salt suggestion might've been horrible for unset ? I thought I did that correctly, but maybe my Python/Jinja was bad.\nYes, GKE doesn't use influxDB. My point is that if the backend storage is not ready, heapster is in crashloop. Vishnu made some changes to make heapster can run standalone, but looks like that fix is not included in heapster 18 based on his comment.\nOkay. I'm still trying to catch up. Why is heapster on GKE waiting on backend storage? And if is unset (e.g. we haven't finished running configure-), it looks like we just give the\nCatching others up: is seeing nodes that are actually hung with CPU pegged on reboot, so the heapster conversation may be a red herring.\nThere are plenty of tests that failed because kubelet didn't come back (and post status) after the reboot. There are a few that failed only because heapster was crash looping. They may be related or not...\nI spun up a GKE test cluster: and the heapster manifest looks fine, so I think I ruled out anything regarding\nI think is the winner. The reboot tests are a red herring.\nI'm actually just seeing heapster flap on a fresh cluster, no reboots.\nHeapster crashlooping is one problem. There are other cases, however, where kubelet just stops posting status all together. If heapster crashlooping can render kubelet unresponsive, that's a bug.\nis the serial console output from the node that got pegged (in ). I rebooted the node to see if it would repro, and it came back up fine. EDIT: the link to the jenkins run.\nI guess the serial console you posted is from build , the very recent one. I saw a similar one and the node never successfully come back. At this moment, I ignore those failures. The reboot failure caused by heapster crashloop is similar to what I observed 2 days ago for influxdb in gce cluster. The comment is at Can we retitle this one for (1)? Reboot test flakiness is caused by many reasons: version screw which is handled by test-infra, heapster and influxdb crashloop is handled by , kube-proxy not ready is handled by cluster team.\nOk. So we think we have the heapster flakes, versioning stuff, and kube-proxy stuff all taken care of elsewhere? I'm fine with targeting this issue at the \"node doesn't come back after reboot\" case. Is that what you were recommending?\nYes. 1) heapster in crashloop if backend is not ready is a known issue for a while, and fix is merged to heapster repositry for a while. We just need to upgrade our heapster after more testing (PR: ) 2) influxdb one requires more information, but we can handle them through 3) version screw is handled by test-infra through . already test/flake label for it 4) kube-proxy readiness issue is handled through\nwhere else should I be looking for hints on why that VM didn't come back to life until I kicked it? Or did I destroy any evidence we might have had?\nFun fact: I thought this might be related the timing of when I made the ContainerVM change on the GKE side, but I made that change for all releases 1.1.4. That picks up this test suite, too, because it's at , and it's green: So this is a divergence from master to 1.1, or between projects.\nThat suite does not run any reboot tests...\nDammit, I thought I scanned for that. Good call. We need to fix that.\nHere's the set of suites that run the reboot tests (thanks kubernetes-e2e-gce kubernetes-e2e-gce-reboot kubernetes-e2e-gke-ci kubernetes-e2e-gke-ci-reboot\nShouldn't we decided only X-reboot test suites run those reboot tests since they are disruptive?\nHm. builds 6478-6480 are all green and all after the (GKE-side) ContainerVM change, so we're looking for something else.\nThe container-vm change was merged at , which lines up with kubernetes-e2e-gke-ci-reboot 6481 (the beginning of the failures).\n6481 is the one with the new containervm change. It includes kubernetes-build/7822, which has the change.\n/ That's the OSS side change, which is irrelevant. GKE flipped at 19:45:55.\nAhhh. Fair enough.\nActually, and I just determined that the GKE-side change didn't make it out until 01:45:44, it looks like, so indeed nothing to do with ContainerVM.\nI took a more careful look at the serial output pasted, and this is interesting: We're actually in the middle of a clean shutdown, trying to unmount and failing. This same stacktrace repeats every 120s.\nI spun up an e2e cluster and with minimal testing have a node in the state of It's the same kernel hang on shutdown. ... and then I accidentally deleted it when I went to bisect, but ohwell. I'm sure I'll be able to get one back if I need.\nOkay, so I've confirmed that at the build level at , I can spin up a GKE cluster and reboot tests fail. I'm going to peel back to and see if they pass (because that matches ). ETA: For reference, I'm able to tickle this with just:\n7819 had a kernel hang on the second run as well, though slightly different signature (still aufs in the end):\nFor my next test, I'm peeling back to v1.1.4. We have no test coverage on reboots on the release branch of GKE right now, and I just want to guarantee this isn't some weird GCE hypervisor issue. And v1.1.4 is clean, at least for two runs worth.\nThis failed quickly on v1.1.5. Hrm. I'm wondering if my v1.1.4 test was a fluke, peeling back to try it there again. v1.1.5 only differs by ContainerVM (and go 1.4.3 vs 1.4.2), but all of our other timelines show that not being directly related. (Alternately, it's a lot easier on the newer kernel.) (Still no idea why the tests aren't triggered, but is.)\nwas very solid. I'm now trying on GKE test. This is a GKE unique thing, because after Friday morning at 1am, anything = will get deployed using the new ContainerVM. So I'm ret-conning a way to effectively try with the new ContainerVM, just to see if it's a contributing factor. It doesn't look like it should be. One of the other commits in the stream is a heapster version bump (from / ). If there's additional flapping or something caused by that (unique to GKE), it may be causing the hang. But that timeline doesn't match the line (relevant to , etc.)\nBad news bears: failed reboot tests. I'm having trouble rectifying this information with what we know about the timeline. But it does suggest there's a problem with ContainerVM at the very least, so there might be two problems here, so I'm rolling one back on GKE and we'll see what happens.\nI've manually confirmed that by reverting deployment of on GKE, I can run reboot tests at and a latest build. Using a special build of latest that will still trigger deployment of , reboot tests fail. Well, damn.\nGKE has been repushed and should, in theory, go back green soon. There's a lot I don't understand about this bug, and it's effectively going to stop us shipping and until we can figure it out.\nOkay, I think that reverting was not all that's necessary. We unfortunately haven't managed to collect a green on yet. That gives me some comfort, because it means that the timeline-discomfort I had previously was founded.\nHeh, I lied and right right writing that we collected our first one. Woo. I suspect will follow soon.\nHaven't we updated go to 1.5 on Friday as well? I don't know if we use the same version of go for GKE as we do in GKE, but in general we know it causes trouble. cc for Heapster thing.\nThe timeline there is off, too. The golang1.5.3 PR went in at , which is well after the trouble started.\nMorning status: Looks like we made a little greener, and didn't touch , so the patient had two viruses.\nOkay, this is my suspicion, but I can't quite piece together what's going on: For some reason, something in (? but that busted the Heapster config, so it's confusing me as to why), caused this. I'm basing this on: the timeline of the timeline of If you suppose that we injected a bug in build 7820 that caused about a 60-70% cointoss as to whether we regress (which is similar to the rate we're seeing now that has been rolled back on GKE), it's not unreasonable that was green for , and that's the only one you have to explain away on this theory. After that, it goes kind of red, and then at about 1am, comes in and wrecks completely as well. Pulling it back much further, to, say, build 7819, and you have to explain the entire range from 6477-6480 as green, which is a much lower probability. So my money is on build 7820. (7821 is revision-equivalent, so boring to talk about.)\nOh. It is Heapster. I spun up a 7819 cluster and a 7820 cluster, and the difference is that even though still had slightly busted Salt syntax, it was functional enough to create the replication controller, whereas on 7819, there's no Heapster replication controller at all. So I think the Heapster flapping is entirely the cause of the remaining issues, so we really need to get in ASAP.\nSo yesterday afternoon I was wondering why build 7820 triggered such a large regression when, say, the bump to heapster in didn't regress us at all. Build 7820 finally let the the fix in , which was actually important because GKE at HEAD wasn't deploying (until last night), it was deploying only , so the pillar wasn't configured. I'm still not exactly sure what's going on here, and I had intentionally left blank for a while during this investigation because it didn't seem like it should matter, but it's obvious that if 7820 is the culprit that it's interesting. I pushed a fix last night to GKE that adds on 1.2 clusters as well, and we've collected our first green dot. I don't know why is still flakey, though.\nEr, correction to my last comment. My change didn't make it out last night and won't be relevant until . The green is a fluke. Sad panda.\nPR introduced syntax error, PR fixed the error although it introduced another syntax issue which was finally fixed in . if you are testing anything make sure you are using version with merged. I agree with that the problem is not with Heapster accessing InfluxDB. Also the same problem takes places in autoscaling suite (described here ). We started observing such problems around Jan 19th. AFAIK once the container crashes kubelet invoke docker inspect on the failing container. How about invoking also docker logs then? It would be very useful for debugging purposes.\nYes, I understand that fixed the issue fully, but at build 7820 the cluster that's created actually spins up a cluster with a replication controller .yaml with resources. At build 7819, it doesn't. The syntax in is busted, but it's apparently not so busted. Previous to , in build 7819 (and builds previous), there was no controller .yaml at all, so there was a window where GKE wasn't spinning up Heapster at all. Hence my conclusion.\nThe syntax in is correct from Salt point of view, although the generated file has empty line in the begging and is not accepted by Api Server, so the RC is not created.\nThat's inconsistent with the fact that Heapster is starting on a 7820 (GKE) cluster.\nI have a new theory, which is that Heapster hasn't been running on GKE in 1.2 since , until .\nIt's not clear to me that upgrading heapster will solve this issue, since heapster doesn't rely on any backend on gke. Instead of asking kubelet to do that, it makes more sense for the test itself to get the logs through the apiserver. I think we should add that to the e2e test.\nI spun up a cluster from just prior to and this is the set of RCs I see: So basically, when went in, Heapster was enabled for the first time in GKE (master, not the release branches) in months and we started flipping out. This is a serious test coverage issue, obviously.\nI edited the above comment, but we would obviously catch this in the release branches because we have monitoring tied to Heapster related things. We need to be better here for CI.\nI made some investigation. The problem with crashlooping Heapster is due to: The issue was fixed in I'll create a patch release and bump the version.\nIt looks like is green. I'm hoping that , which is the first build past , will go green as well. (The cycle time on that build is waaaaay too long.)\nIt looks like we're good! is green now!\nfor node problem detector."}
{"_id":"q-en-kubernetes-f10bead500960ea3dde1ad0ae7083423c7980e31d3345e62b31b741d2afc29a2","text":"I noticed this flake by looking at the submit queue. The test is simple, but when it fails, we have no good way to debug it. Below is the test in question. When the test fails, we should print the lines we got in each case."}
{"_id":"q-en-kubernetes-1b2bcc3aa22e5d5f53df9233e5f80a56f1f4bdaeb462b3a5016092de1417360c","text":"hack/build-cross sets it without looking at the current value. And the value has to be an array, which is a bit awkward to use.\nThis could be improved indeed. Maybe something like and Haven't tested it though... But why would one want to change the value?\nI just wanted to build one, so I did the \"obvious\" thing: hack/dev-build-and-` It failed for multiple reasons :) , Lucas Käldström wrote:\nThis issue has been a major downer for me. It's much harder right now than it ought to be to articulate the few platforms for which you'd like to build each of server binaries, client binaries, test binaries, etc. Every time I've tried to do this \"the right way,\" I've spent hours looking at scripts and feeling frustrated that build options aren't documented better. Every time, I end up cutting my losses and resorting to temporary build script hacks instead, just to get the job done.\nDoesn't fix this issue exactly, but I've found that you can easily build only if you set .\nTaking a look at this today, I also thought that setting seems to be the most reasonable.\n/sig release"}
{"_id":"q-en-kubernetes-124a8f04c015bf597a839b5fc96d254a778164bb355fa6ee63ad65e704649190","text":"The delete command for a replicaset hangs forever. This then manifests as phantom RSs when I delete and recreate a namespace.\n- delete is the gift that keeps on giving\ncc\nHad you already deleted the corresponding Deployment?\nI am still seeing this problem. Deployments and replica sets are sticking.\nFrom what I can tell, calling DELETE_COLLECTION on Deployments is returning: And since the controller assumes not found errors are equivalent to already deleted, it thinks the content is deleted. When I manually inspect, I still see the deployment and replica set laying around... //cc\nholding on this while i debug some more...\nThis was a false alarm, apologies."}
{"_id":"q-en-kubernetes-56abf1c19aeb772bfdf5543a20f43dcbe7bbdea78cd050790b056a5e979d3a30","text":"http://pr- is another failure.\nAlso seen in http://pr-\nAgain. Not assigned to anyone, so I am sending to for routing\nAgain: Is there anyone working on this? I will spend some time looking into this.\nI think that the name of this issue is misleading - the issue seems to be unrelated to etcd_watcher Basically, if you look into this part of log: So it seems that, the WaitGroup that is problematic is in httptest.Server. So my feeling is that we are simply talking to a server that is already being close and this is ~roughly duplicate of We should just find where exactly the problem is and try to work-around it.\nI think this is exactly what we had before. I will send a PR with the fix today."}
{"_id":"q-en-kubernetes-7ebf7acbde7789cdda053450cb89ee471a2ab525583ef16f7638edd03a598813","text":"I'm following step by step tutorial on how to create resource quotas per namespace () looks like this: { \"apiVersion\": \"v1\", \"kind\": \"ResourceQuota\", \"metadata\": { \"name\": \"quota\", }, \"spec\": { \"hard\": { \"memory\": \"1Gi\", \"cpu\": \"20\", \"pods\": \"10\", \"services\": \"5\", \"replicationcontrollers\":\"20\", \"resourcequotas\":\"1\", }, } when i'm about to create quota it fails with: kubectl create -f --namespace=prod invalid character '}' looking for beginning of object key string Namespace \"prod\" exists: kubectl get namespace NAME LABELS STATUS AGE default &lt;noneActive 26d dev name=development Active 8d kube-system &lt;noneActive 26d prod name=production Active 8d kubectl version Client Version: version.Info{Major:\"1\", Minor:\"1\", GitVersion:\"v1.1.2\", GitCommit:\"\", GitTreeState:\"clean\"} Server Version: version.Info{Major:\"1\", Minor:\"1\", GitVersion:\"v1.1.2\", GitCommit:\"\", GitTreeState:\"clean\"}\nThose trailing commas are wrong. should look like: I'll submit a docs patch\nsweet. ta. That seem to do the trick: kubectl create -f --namespace=prod resourcequota \"quota\" created\nI'm going to reopen until the PR which fixes this merges, when it should autoclose...\nI'm following this tutorial on installing percona with kubectl, and I'm getting a similar error: When I run: kubectl create -f pmm-server- I get: error: json: line 1: invalid character ‘Â’ looking for beginning of object key string"}
{"_id":"q-en-kubernetes-8281098c4a3749f944133097833d93b74dab4d64a54e55b1ef6d2ac82651b4ee","text":"The says this about the new GUI: It does not what? Seems important!\nHmm... That's interesting. Will fix."}
{"_id":"q-en-kubernetes-eaf1a4d119ad0fb53a39512c42d412af412ec0302afd730bf3a1c25d36f547fb","text":"When creating a new cluster on AWS with the 1.2 version I always get a failure when checking docker on the nodes, but after the check loop ends everything seems to work fine and all nodes are marked as ready. The instance types for nodes is\nThat is very surprising. Are you using any particular settings (to help me try to reproduce the problem)?\nThese are the exports I make before running kube-up:\nThe problem is the instances take a very long time to start up. I had to increase the timeout to 19.\nHey how did you increase the timeout? Didn't see a setting in config- that I could override. Getting the same issue as the poster.\nThanks , that did the trick!"}
{"_id":"q-en-kubernetes-ff81b881a335ec393925d5c77ad19a1d9dde82cef6e119d6f7df449f9fa6f37d","text":"I have been having the kube-controller-manager (running in a container) crash quite frequently with this error: I cant fathom what seems to cause it. It seems to happen consistently for a while (CrashLoopBackoff occurs) and then fixes itself after an undetermined amount of time. Kubectl version: Docker version: Networking: Calico (with the calico k8s policy agent) Please do let me know if there is any other information that I can supply. I appreciate I haven't given much to go off of. I am not sure what the steps to reproduce are...\ncc\nIt looks like a bug. I'll take a look. Thanks for reporting it.\nWeirdly, the description showed up in the email github sent me but it appears absent on the github issue?\nThe description was there when I read it but disappeared now.\nWell thats very odd! If you have it on your email, could you post it on here please (and I will update the description)? I unfortunately haven't got the error message to hand now...\nI took the liberty of just editing the OP.\nthanks a lot! Not sure what happened there!\ncould you paste the returned output at \"server-ip:port/api\" and \"server-ip:port/apis\"?\nSure.\nThanks \"\" has an empty preferred version and that's causing the error you reported. Is \"\" created via the thirdparty resource api? ()\naha! Good spot! Yes, it was created via that API. This is the Yaml file I used to create it: Here is the output of\nThanks for the confirmation Your configuration is correct. The preferred version is left empty by the thirdparty resource controller, while our discovery client has an assumption that the preferred version won't be empty. I'll patch the discovery client to tolerate an empty preferred version, and will discuss with on if we should make a default preferred version for a thirdparty resource.\nThanks a lot Great job on working it out :)\nSent to fix.\nWhat would be a workaround for this before the release, how can one specify the preferredVersion?\nAFAIK, there is no workaround. There is no way to specify the preferredVersion for the thirdPartyResource.\nI have a quick workaround to fix this : 1) Change the kubernetes to reflect the following : from : thirdpartyresources=true to : thirdpartyresources=false 2) Then restart 'kubelet' service OR reboot the K8S Master server systemctl restart kubelet 3) Revert back the changes done to kube- from : thirdpartyresources=false to : thirdpartyresources=true 4) Verify whether the above change on the above manifest file has been automatically picked up by the kubernetes API ps aux | grep -i thirdpartyresources 5) Finally you may restart kubelet service systemctl restart kubelet 6) Now verify whether the 'kube-controller-manager' is in a 'Running' state kubectl get pods --all-namespaces | grep -i controller-manager Hope this fix works for you guys as well !!"}
{"_id":"q-en-kubernetes-7377f65a40865118420e0f4c4b965cad4b25ab1082e4c9d411d42ac66506315f","text":"Minor improvement in UX maybe?\nWe have semi-ambiguous syntax because 'get pod/foo' is also valid. We should have define syntax for this early on. On Apr 8, 2016 7:13 PM, \"Prashanth B\" wrote:\n+1 for the UX improvement.\nget pod/foo != get pod ns/foo ?\n.... LOL\nThis would make SO much easier to use :+1:\nI dont disagree, but it is somewhat complicated to transition now. Maybe something like 'kind/ns:name' which would be less ambiguous and strictly compatible (ns:name is not a valid name, nor is it confusing wrt kind/name). Need to define precedence of this with --namespace flag and in-file namespace yaml. I also don't see it as a huge win, personally. It would have been nice from day1 but relatively minor now, IMO. But this is a decision for UX team - if someone sent a patch implementing something like it, or at least a proposal that worked through all the details, it would be a better starting place. On Apr 10, 2016 12:33 PM, \"Lucas Käldström\" wrote:\nThis is more like P0 to me. Somewhere else I complained about having to type too much and the kind/name syntax. My vote if for kind/ns:name, too.\ncc\nWhat is the scenario you're trying to simplify? You often need to use resources from multiple namespaces on the same command line?\ncc\nns:name is not a valid name for namespaces and our core resources, but it can be a valid name for extension resources in the Kube ecosystem, like users, roles, authorizations. We'd have to add more slashes\nAttaching the namespace to the \"name\" part won't work well because then you have to chose whether is \"list all resources in namespace token\" or \"get the resource named token in the current namespace\". To allow both cases, you must attach the information as a discretely tokenized part or as part of the \"resource\" part. I think that parses unambiguously since resource can't have dots groups can't have colons and namespaces can't have slashes.\nI can assign short names to namespaces and end up with a small kubectl get line, --namespace= just feels unnecessary. I guess it would feel more natural if I could also do . This isn't a huge deal for me personally because I just bash alias my way out of typing.\nIn my case, I work with multiple namespaces, not on the same command line (I use --all-namespaces frequently, too). I guess I could create bash aliases, but I also tend to reinstall the machines I run kubectl from on a regular basis.\n--all-namespaces is so long anyway that it makes me cringe. , Rudi C wrote:\nWhat api group syntax did we settle on? I don't see it documented anywhere. cc\nWith Ubernetes, we'll also have multi-cluster scenarios coming.\nSyntax is: <resource[[.<version].<group] , Brian Grant wrote:\nI think either a short option, like or could be a solution or suffix the namespace using \" .\nwas hoping to save for indicating a for the day when we'd be able to retrieve a historical version of an object via the API. Does not alias to like openshift?\nNo it does not.\nwould be an improvement for me: it's noticeably shorter than . Does introduce ambiguities with other flags? or introduce some other issue? or is the hope that there's a nicer unambiguous solution (e.g., something similar to the original proposal)?\nI don't know of any conflicts. I'm in favor of the short flag because it gets used all the time.\nI am fine with , really appreciate if you could help with this.\nWe originally had -n in Kube but took it out while we sorted out the other config. We went and it in OpenShift because it was used all the time, but forgot to add it in in kubectl. I think it's time to bring it back to kubectl. Other folks who care about this in kubectl should v oice objections in the next day or so, otherwise I think we have consensus to add as the alias for . Can someone scan over the flags in use today in kubectl and verify that no one has introduced a ? , TonyAdo wrote:\nAlready did, just client generation. has opened to add this. I've got a minor comments unrelated to the code and tagged to give him a chance to object.\nI'm also in favor for adding as a shorthand for"}
{"_id":"q-en-kubernetes-c7a72df11498e7300fcb7761274972d0c9f40da0bd1ab208df9d0e2fd8c6758c","text":"I'd like to be able to sort by a timestamp, e.g. but that fails with an exception: I appreciate that the root of the problem is probably outside of kubernetes, yet I hope you'd agree that sorting by a timestamp is very useful functionality.\nAgree with sorting by timestamp.. Just make it clear, is of type: while only these types are allowed to be sorted:"}
{"_id":"q-en-kubernetes-21b704a68514d2b400f1a4f479b12f48c6dc4bd890b9d4b14cffa5ee4c13d7d4","text":"During\nany chance this is related to pass: https://k8s- fail: https://k8s- https://k8s-\nThis is because for some reason the submit queue rebuild didn't catch the flake before merging. Protobuf landed first, merging a new api object like petset should have failed.\nQuick fix is to rebuild the proto marshalers and push.\nAlso, hack/verify-generated- is broken, which probably is the root cause.\nLooking at it now\nThree problems: -generated-protobuf wasn't updated when we made update-generated-protobuf use docker, so it didn't fail out script was verifying the old API groups, not any of the new API groups that have been merged without protobuf serializations (although pkg/api/serialization_test should have failed, possibly the job didn't build with the latest master?), which caused the tests to fail\nSomething about the merge bot is broken - it merged without all the tests being green: . That's why this broke -\nSpawned since there is something wrong with update-generated-protobuf - will merge that to fix the failing test and then debug the generator in the morning.\nThe bot can race with pushes between when it starts testing and when it merges..."}
{"_id":"q-en-kubernetes-b0a43404a90a4502792529d6f593d8f07cbefedf62cf2856b9027033dd071c9f","text":"Hi, Deployment patching with doesn't show up in the deployment rollout history: first: after: Is it a bug? Should it show up? Thanks, Peter\ncc\nWhich kubernetes and kubectl releases? Did the patch actually trigger a rollout?\nI also run the command from CI with the latest . So the problem affects the v1.2.3 kubectl as well\nWe should also fix the way we sort revisions - 10 should be after 9, not 1.\nCan confirm this. The issue is that the patch isn't updating , so the rollout history will just copy the value from the previous version. If you started with you get that for the patches. In my case I have a different line repeated. Builds 7 and 8 were built with different . I'm not sure why the patch worked once. Command I am calling is:\nIf no one is working on patch - i can try to make one.\nI'll like to work on this issue if you haven't started yet\nActually never mind I can look at something else :)\nHi, i see that you are the author of the following comment. I think that currently every valid change cause is discarded because of version conflict. Maybe you have some information why it happens?\ncommand instead generates a patch to update then applies that.\nIt was the same for patch, but for some reason was changed.\nThe intent was two fold. First, prevent the change-cause from ever failing patch the CLI command. change-cause isn't authoritative and the patch took hold, so a non-zero return code is inappropriate. Second, if the resource version change post-patch, before the update, the change-cause is unknown, something else touched it after the fact. A command that issues a instead of an could very easily break the history since patches will attempt to reapply themselves regardless of resource version in the face of conflicts. The object returned from the patch operation is the one that we use to avoid hitting a resource conflict with ourselves. Truth be told, I don't feel very strongly about getting the history right since it can be changed anyone with write access to the object, includes secrets that should be elided (--token for instance), ignores environment variables which affect outcome, ignores the actual path used to locate the binary you're running, and references paths that are only good from one location on one machine, so if you want to switch it back to a second patch, I don't mind as long as a failure to update the object doesn't cause a non-zero return from the command.\nCould there be a race between the call and the deployment updating itself to action the changes? On 19 May 2016 9:47 pm, \"David Eads\" wrote:"}
{"_id":"q-en-kubernetes-69b2bd17365c09a24925384c93384c1277cb9057b6c5ebf013176177cc0220c1","text":"Running show failures at times: The expected queue is while got , see code here: I would give an idea that these cases have something similar, so referring and /cc\ncc\nI think so as well.\nThe debugging I to the test is obviously not correct - I'll fix that separately.\nI wonder if on very short intervals we're not getting a goroutine schedule swap, which causes this race to manifest (based on comments in the other thread).\nDoes heap insertion guarantee order if the heap value is equivalent? I don't think so.\nThis seems like it is that now() is not monotonically increasing, so we violate the add ordering condition. Since we already have a lock in most cases, we may want to either wrap now() to make it monotonic inc, or have the queue check for equality in the heap and add one nanosecond until there is no matching heap item. On Apr 29, 2016, at 7:12 PM, Marek Grabowski wrote: cc — You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub\nThanks for clarification, so it looks like we need to give some time break between doing , I'll see whether that helps.\nI believe this is a bug in rate limited queue, so we definitely need to fix it (esp. since it's happening in the real world)\nyou wrote Would you be a good person to pick an owner for this bug? Giving P1 since we think happens in real world and I don't understand severity.\nand I are probably the two who have touched it, I think as well but may be misremembering. It should be a straightforward fix to make now monotonic, I don't know when I can get to it\nMe neither. If it's P1 I most likely won't have time before v1.3.\nI think override the function call in as you did would resolve this issue, what do you think, if you agree with this, I am gonna to try that.\nIt would fix it, although I think this is broken in the core code, not just in the test. We suspect real workloads have been impacted by this.\nOpened to determine whether we actually need to fix the RLQ\nAssigning to per"}
{"_id":"q-en-kubernetes-7e9c2f49189e70f3f181de8a8d3d5cbdc6480a56bdc940348429fa13efb7c56f","text":"When i doing an elasticsearch aggs with the following statement: then, i got this: Actually, i want the pod names like this:\nI have solved this problem by adding a index template for k8s logs. Now, I can aggregation pod names of the same namespace.\nHow are you sending your logs to elasticsearch? Using the default fluentd configuration?\nFirst, I use the default fluentd configuration. Secondly, My application is written with nodejs language, using library."}
{"_id":"q-en-kubernetes-42f599f347213b163fe7149497631054b9aebf3a38dc08dc679e08b10e38a774","text":"is and does not honor the platform it is invoked on. I noticed this when I needed to install etcd on my Mac (which, in turn, was required by and transitively by the pre-commit hooks) and later failed to execute it due to an incompatible binary.\nOh wow, that explains my failures. I ended up using homebrew."}
{"_id":"q-en-kubernetes-5056a2515a29befdf589a74f13db031ed4142bbabfa8d2c37cafac2de4fccaea","text":"I started the federation-apiserver and went to and got the following output: Expected: I should have got an empty v1.ServiceList Any idea on what we are doing wrong? Have you guys seen this error?\nI think your deep copiers or conversion functions are not getting registered properly. If you look at my change in , some places are inappropriately registering with api.Scheme instead of the passed in scheme. Maybe you have that problem or a similar one, since I think you guys copied some code. , Nikhil Jindal wrote:\nDeep copy generation was slightly broken as well - you may want to double check that they are registered. On May 23, 2016, at 5:50 PM, Daniel Smith wrote: I think your deep copiers or conversion functions are not getting registered properly. If you look at my change in , some places are inappropriately registering with api.Scheme instead of the passed in scheme. Maybe you have that problem or a similar one, since I think you guys copied some code. , Nikhil Jindal wrote: — You are receiving this because you are on a team that was mentioned. Reply to this email directly or view it on GitHub\nI got the same error with latest branch, as I was swamping on service controller, so I use a previous version and worked around it.\ndeep copies were indeed missing, but adding them hasnt fixed the issue. Will debug more. is my PR to add the deep copies cc Please feel free to file issues for any such problem you face. We need to fix all these issues. Its good if we know about them early. Thanks!\nfixes this"}
{"_id":"q-en-kubernetes-c5413e82f6f4d07df951c5f23d8ac5d09ba356ec19c2b239217d4746388b00db","text":"https://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nAbhishek ought not be responsible for this any more.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention...\n[FLAKE-PING] This flaky-test issue would love to have more attention...\n[FLAKE-PING] This flaky-test issue would love to have more attention.\n[FLAKE-PING] This flaky-test issue would love to have more attention.\n[FLAKE-PING] This flaky-test issue would love to have more attention.\n[FLAKE-PING] This flaky-test issue would love to have more attention.\n[FLAKE-PING] This flaky-test issue would love to have more attention.\n[FLAKE-PING] This flaky-test issue would love to have more attention.\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nwhat is the status of this issue?\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nNo idea. This should have nothing to do with me.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nAh kk just thought you might know something given the label.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nI know it didn't get any attention last release :)\nStatus is that I'm trying to fix it... I some code to the test to help debug why the flake was occurring as I wasn't able to repro it on my local cluster.\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite} Happened on a presubmit run in .\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nhttps://k8s- Failed: [] KubeletManagedEtcHosts should test kubelet managed /etc/hosts file {Kubernetes e2e suite}\nThis seems to be related to issue"}
{"_id":"q-en-kubernetes-7679d12cc281cd209833defbf14abe814f6b5c54a99eb0fd89353cf78522b104","text":"Hello Kubernetes, The order of the columns when you get events confused me. I was thinking, Turns out it's just a visual problem where these events are sorted correctly (by ), but the first column is . I think the first column should be because that's the order the events are listed.\nThis also confused me when I first saw it - I think this ordering would be more intuitive and help users discover the aggregation feature logically, instead of first having to grapple with why events are \"out of order\"\nShould the order be always LASTSEEN,FIRSTSEEN or just when using -w? I have an ugly PR for the latter, but perhaps we should just go for the former."}
{"_id":"q-en-kubernetes-7f95224274155bb39af2e05742451e79cefc1f176d05622797c3ba56ecf46808","text":"is failing with: Examples:\nThis is the failing line of code...\nPASS: FAIL: Seems like this disappeared: Also this is flipped: we run the following\nmy leading theory is this is something in\nThe issue is that the above PR changed a bunch of to be and thus we no longer wait for getArchive() to complete before starting tests. And unfortunately the previous PR cannot be reverted\nAfter merging, kubelet-gce-e2e-ci still failed with the same issue. reopen the issue. Currently the failure blocked the merge-bot.\ndid stop the bleeding. Now kubelet-gce-e2e-ci is green: and merge-bot is unblocked.\nI think the \"go arc\" piece was a red herring. I suspect the issue is that the same PR tries to copy the binary into the archive directory, but PR stops building ginkgo, so the copy fails. That is why this line then fails..."}
{"_id":"q-en-kubernetes-137e73901e94fc53b23ce122acea4f6c63caaffb221f4c8fdcf70adeb68da5f7","text":"When replacing an existing pod, the command fails as it tries to create a pod with a name that is still used by terminating pod. How does the option works? deletes a pod (e.g. mypod) creates a new pod with the same name (mypod) What happens? As the command tries to recreate the pod, it fails as the is in state. Thus it can not be created. The problem is the does not wait for a pod to be deleted. Reproducer: run mypod --image=yapei/hello-openshift --generator=run-pod/v1 get pod mypod -o yaml replace -f --force The last command ends with:\noption introduce by\nWith set to positive value, what should be the default behaviour of the command? When set to , should the command hang for and then create the pod? Or is there a way to create a callback (inside a scheduler?) that creates the pod once the current one is deleted?\nYou could watch for the resource to be deleted and create after you see the deletion happen\nhave you ever tried to set , I think this will do a non graceful deletion. should make sure the old resource is deleted before it creates a new resource, if not, that's a bug.\nThough the help says the is to be used only with , it returns the error above.\nThen I think there are two issues we need to track: 1. option should be ignored or just returns error if users are not using option. 2. should always make sure the old resource has been deleted before create the new resource."}
{"_id":"q-en-kubernetes-5d760c5ea5685578ebb9830d3d5528d6365ffb387067e4c2605824d06bca490e","text":"As mentioned in , the federation etcd will lose its state any time the pod dies, unless we modify to add a PVC. We'll also need to create a PV and modify the turnup script to create the backing volume in the cloud provider.\nFYI I believe this is a showstopper.\nOK. implied I could use to get this provisioned on-demand. AFAICT, , and since it makes our life easier I am trying to get this to work in my test cluster, but so far no joy. Any obvious mistake?\nI needed to not specify a .\nwhew! , Matt Liggett wrote:"}
{"_id":"q-en-kubernetes-54ee9a53c145bcb7f1ea4e17c59ef3f3aaba49d2a04dbd5ca8ea969af1a94c94","text":"https://k8s- Failed: [] Container Runtime Conformance Test container runtime conformance blackbox test when running a container with a new image should not be able to pull from private registry without secret {E2eNode Suite}\nSame with , the fix is under review\nhttps://k8s- Failed: [] Container Runtime Conformance Test container runtime conformance blackbox test when running a container with a new image should not be able to pull from private registry without secret {E2eNode Suite}\nhttps://k8s- Failed: [] Container Runtime Conformance Test container runtime conformance blackbox test when running a container with a new image should not be able to pull from private registry without secret {E2eNode Suite}\nThe fix has been merged , close this one. Feel free to reopen if this happens again."}
{"_id":"q-en-kubernetes-3c64df687e12e254812ed26f14838f867cca01f1b7d5e271e2d52376c8409ae4","text":"Either include non-arch specific stuff in every tar or break that out too. ref cc\ninterested in helping with this one?\nYes, I may help\nOK, I did not actually understand what was in , apparently. Alternate proposal: [ ] update all references that need just server binaries to download the server tarball directly [ ] same for places that just need (related to ) [ ] remove all arch-specific binary artifacts from and replace with a script which downloads the client/server tarballs as necessary. The alternative option, including all of the non-arch-specific code and docs in an arch-specific starts to get very tricky, since we run into a large cross product of client x server platforms/archs."}
{"_id":"q-en-kubernetes-749738408e786a27678c468a5520c46b775b406c0396b4fc77d0584fc060b72d","text":"The code in doesn't work for me: does not pass on SIGTERM when the caller is in the same process group. Then the knocks out the process without giving it a chance to kill its child. The bit doesn't help: either the process has already been killed as above, or it gets killed now with SIGKILL; either way the process is still running. Oddly, it does work about one time in ten. I have no explanation for that. I am running from the command-line like this: , on 0-41-generic -Ubuntu SMP Fri Jun 24 11:28:43 UTC 2016 x8664 x8664 x86_64 GNU/Linux CC\nThe manual says \"This is the signal that the calling process will get when its parent dies.\"; it is applied to the process, so that will get killed if the process dies. That isn't what we are looking for:"}
{"_id":"q-en-kubernetes-df06935dc12cccd535d95a7add46acf8ed7f7b060e66bb7b117b3eccd06850f8","text":"https://k8s- Failed: [] [Feature:Example] [] Storm should create and stop Zookeeper, Nimbus and Storm worker servers {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [Feature:Example] [] Storm should create and stop Zookeeper, Nimbus and Storm worker servers {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [Feature:Example] [] Storm should create and stop Zookeeper, Nimbus and Storm worker servers {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [Feature:Example] [] Storm should create and stop Zookeeper, Nimbus and Storm worker servers {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [Feature:Example] [] Storm should create and stop Zookeeper, Nimbus and Storm worker servers {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention...\nThis has been marked with the \"needs attention\" label and needs to be addressed for the 1.4 release. Could you please close if it's fixed update on your progress to fix it make a case for it not to block the release or find a new owner who can do one of the above soon. Once you or the new owner update this issue and are actively working on it, you can remove the \"needs attention\" label.\nthe following message looks suspicious, looks a DNS issue?\nActually more like a network connectivity issue\ndo you happen to have the Dockerfile for these images? We better push them into\nI don't think this is a blocking issue. It looks a transient issue. Since the example is largely dependent on whether the image can be fetched and how long it takes to pull the image, it is very environment sensitive. I would recommend removing label. Longer term, we should move the images used in examples to gcr.\nhttps://k8s- Failed: [] [Feature:Example] [] Storm should create and stop Zookeeper, Nimbus and Storm worker servers {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nfor the last flake, based on , image pull (mattf/zookeeper:latest) started at 42:40, finished at 48:31, about 5min51sec, well passed 5 min\nstorm-nimbus, storm-worker and zookeeper are from they're all automated builds from github (quick link: and )\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] [Feature:Example] [] Storm should create and stop Zookeeper, Nimbus and Storm worker servers {Kubernetes e2e suite}"}
{"_id":"q-en-kubernetes-bfcd8c1616e8828754fd2982c8dfa9b62a9f64f48b1a3ca5310d1a2eb1f5c300","text":"I think got confused between e.g. KUBEMASTEROSDISTRIBUTION and MASTEROS_DISTRIBUTION; it definitely doesn't match GCE. PR coming in once I've tested.\ncc"}
{"_id":"q-en-kubernetes-a540a96ec019e4374a1aed3eeb2f1a3870891c79dbab54fc2c3626988ef08e83","text":"Description of problem Create a petset, when pod is evicted, the petset can't create a new pod on other node and make sure pod is running. Kubernetes version (use ): v1.4.0-alpha.2.1427+ How to reproduce it (as minimally and precisely as possible): kubele eviction-hard=\"memory.available<${value}\" a peset and wait all pod is running node MemoryPressure=true all pod status What happened: is evicted and petset can't create new pod and make sure pod is running. What you expected to happen: create new pod on other node and make sure pod is running.\nAlso when node come back with sufficient memory, petset can't make pod running again.\nFYI\ncc\nWas able to reproduce it. Pet set controller just loops with the following logs I0830 12:32:52. 9 ] Syncing PetSet default/cockroachdb with 3 pets I0830 12:32:52. 9 ] PetSet cockroachdb blocked from scaling on pet cockroachdb-0 I0830 12:32:52. 9 ] PetSet cockroachdb waiting on unhealthy pet cockroachdb-0 I0830 12:32:52. 9 ] PetSet cockroachdb waiting on unhealthy pet cockroachdb-1 I0830 12:32:52. 9 ] PetSet cockroachdb waiting on unhealthy pet cockroachdb-2\n-- do you have the petset YAML that we can evaluate? What is the quality of service tier for the pods produced by the petset? Can you include the of an evicted petset pod so we can see the final resource requirements spec?\nI am able to reproduce it with nginx pet set in e2e test. Here is yaml output of it Also it seems that i have working fix and almost finished debugging e2e test (patchset is linked above). Will be grateful for review.\nAh, this helps clear things up... the linked pr shows the problem in petset controller and not kubelet...\nDoes it make sense to set milestone 1.4 for this issue? I think the fix is ready for review"}
{"_id":"q-en-kubernetes-5ed5c1e20e14fe5a611a17f918cbdb7750120c4d2b83d5cbf5c92766c9f4890f","text":"https://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite} Previous issues for this test:\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nI'm pretty sure, those amount of failure are related to: (I don't know why yet).\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}"}
{"_id":"q-en-kubernetes-ccf323f57343c53a81e615f5f91e91830995a9424300c72d74f0847dc125d753","text":"<!-- Thanks for filing an issue! Before hitting the button, please answer these questions.--Is this a request for help? Currently, the 4th parameter name of updateServiceListener in kube-proxy is addOrDelete, However, from the func caller, we just pass for AddServiceListener, and for DeleteServiceListener, So, I think the name addOrDelete is unreasonable. I prefer to change the parameter name addOrDelete to add so that avoid making people confused. In addition, there is a typo, We should change addition to deletion! Also, there is a sentence breaking, we should change from No Service Endpoints Not Found =No Service Endpoints are found Thanks!\nFixed!"}
{"_id":"q-en-kubernetes-2e5a484b35fec960cddb956eba6f052845dad1c368a586555990c6f6b0a2bb05","text":"In practice this means that it doesn't get to update and thereby may refuse to admit pods which had been assigned to it by the scheduler. For example, if you update labels for that given node, and try to use affinity or on a pod, it won't get to run.\nThis looks like a sufficient fix for me, but I'm happy to add instead, if that'd help getting it accepted for 1.4.0: cc\ncc\nThat commit makes sense to me.\nThanks, I'll open a PR."}
{"_id":"q-en-kubernetes-235c4fd5c746c299713da1f29ab9aad59c79650aa9c03654d13b0d18ce6db4b5","text":"While following this to setup a kube-mesos integration env on my laptop, I found the default service account and default token are not created when start the cluster. As a result, I can not create a pod or deployment in this environment. Reproduce steps: the latest Mesos code, build and start 1 single master with 1 single agent. the latest Kubernetes code, use to build. the following commands to start apiserver, controller-manager and scheduler respectively: all the components are started, run command and see that there is no default token created for the default namespace. At this moment, run \"kubectl create -f \" will see the failure of creating a pod.\nThis introduced to node controller and now when we start a node controller, if is not nil, we will start this Informer controller as well. But the start of the will block main thread from starting other controllers later in the main thread. we can see that in the kube-mesos integration, we started some other controllers, including serviceaccount controller, after we started node controller. I will submit a patch later to run the in a go routine to solve the issue."}
{"_id":"q-en-kubernetes-a045972f2041bf03a456c81741b4b6a2d857d31a19a08ce9353cd2fe9dd6937b","text":"https://k8s- Failed: ThirdParty resources Simple Third Party creating/deleting thirdparty objects works [Conformance] {Kubernetes e2e suite}\ncc\nSeems like the namespace creation/deletion process is flaky. IIRC there have been some recent PRs touching this area, is that right?\nWe recently code to try to expose the real underlying error for why the ns deletion fails. This is hopefully now showing it.\nThe issue is with TPR not integrated properly with the namespace controller. Creation of a new type requires a restart of the controller-manager given current state of the code. The test must be marked flaky until the state of the art is improved (which is not happening in 1.4)."}
{"_id":"q-en-kubernetes-4ee7dcc3235d36c5ced023c560a1d39a0e56a3b4e7f45b338561a6cfa9a84c80","text":"https://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite} Previous issues for this test:\nThis is pretty much the same failure, that are happening in \"gce-enormous\" suite and we currently debugging them with The main reason of those failures is that we are terminating all watchers for a particular resource type from time to time, and some of them are actually causing a snowball effect. After more looking into it, my hypothesis is that our machinery for aggregating watchers in apiserver sometimes is just to slow. In particular, I some log in my local branch and what I'm observing is that in etcd_watcher, we are: accumulating objects in incoming channel but also accumulating objects in outgoing channel, and the accumulation there is usually higher [I will send a PR to put those logs into master - they seem to be pretty useful for debugging such issues.] So my current hypothesis is that dispatching watch events to watchers is what is too slow. We will continue debugging.\nThe logs that I seem to confirm my hypothesis - outgoing channel is generally more full then the incoming channel in etcd_watcher: vs I'm working on the PR that should at least improve the situation.\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nhttps://k8s- Failed: [] Load capacity [Feature:Performance] should be able to handle 30 pods per node {Kubernetes e2e suite}\nPlease apply the label to this issue if it does not need to be fixed before we launch 1.4. Otherwise please advise on how to proceed with respect to 1.4.\nThis was a regression - I fixed it by reverting an offending PR: This seems to be fixed now.\nI have 33 pods in my container (1 is substantive, the other 32 just run on log files). I get this all the time now. I'm running 1.5.4.\n- to be honest, I don't undertsand what your wrote. what is \"container\" in \"33 pods in my container\"? Also, if you believe there is a bug somewhere, please open a new issue instead of adding comment to more than half-year-old one."}
{"_id":"q-en-kubernetes-096053f9afbdae0d9ebcf10306786bac35334430e34aeee27d2c517b7f3c22e9","text":"https://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite} Happened on a presubmit run in . Previous issues for this test:\nfailed in https://k8s-\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite} Happened on a presubmit run in .\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nAny updates on this issue If you could you change the milestone to 1.5 and add a label for the responsible team that would be a great help as we triage flakes for the 1.5 release. Thanks!\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nThere is a problem with fetching images on GKE. It seems that pod is unable to fetch image for more than 2 minutes. It is strange, as AFAIK the image should be pre-fetched by image-puller:\n[FLAKE-PING] This flaky-test issue would love to have more attention.\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\n, all issues must be labeled either release blocker or non release blocking by end of day 18 November 2016 PST. If you are not the correct owner please identify the correct owner and add them to this thread for assignment. Thanks for your help. cc:\nWRT https://k8s- There was a problem w/removing namespace. This is a duplicate of and is not related to HPA.\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nCould you please help triage this issue as blocking or non-blocking for 1.5 (or help find someone who can). If there is a real issue here, we need to find out ASAP, or else we risk the 1.5 release date.\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nWRT https://k8s- build: kubelet logs for pod rc-light-qyowb : Kubelet is not starting pod for 2 minutes, reason unknown. The same problem is in issue Can you take a look or assign someone from kubelet team?\nI suspect that the logging level is set to in GKE tests and so most of the useful logs are not available. Let me figure out a means to bump the logging level to further debug this issue. you might still have to triage the non v1.4 release failures to address concern .\nhow can we update the environment variables for GKE test clusters? For OSS, we up the kubelet logging for test clusters\nI still don't agree with test environmental changes, which is one of the reasons we've never made this a priority. A GKE test cluster looks the same as a GKE prod cluster. How would you debug this in production? Always ask that, and you come back to test == prod.\ncc\nIn my opinion it is release blocker: sometimes pods are not started\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nThe pod startup failure you posted in is for release 1.4 branch. Why do you think it is a release blocker? We need to triage other failures on HEAD and/or 1.5 branch which could be release blockers. If the intent is to keep prod and test in sync, then let's up the logging level in prod as well. When issues happen in prod, we might need those missing log lines anyways. I suspect that we might have to have GCI rotate logs more aggressively as a side effect.\nWhy is the argument to increase the log level rather than making sure that all logs necessary for debugging production issues are at level 2 or above? Levels 3/4 shouldn't be necessary to figure out why a key control loop isn't behaving correctly.\nWhat you suggest is a good fix for the future. We need to reexamine logging levels across kubelet. However, that might not happen in the next few days. How do we make progress now?\nWould isolating logging level changes to specific test suites be acceptable? That way, most of our tests would match production setup.\nThat's what I'd suggested on the other (internal) code review.\nDo we have consensus on how to get past this as a release-blocker?\nWe are waiting on to give a go ahead from the release perspective. The kubelet failures that and I have been discussing about have not been on v1.5 AFAIK.\nSorry, I didn't notice that this problem is on 1.4 branch. If so, the issue is not release-blocker. So, I'm removing the label.\nAck. Marking as non-release-blocker\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}\nIs it appropriate to move this to the next milestone? (and remove the non-release-blocker tag as well)\nPR should be cherry-picked to 1.5 This is not a release-blocker as it affects e2e tests only.\nhttps://k8s- Failed: [] [HPA] Horizontal pod autoscaling (scale resource: CPU) [] ReplicationController light Should scale from 1 pod to 2 pods {Kubernetes e2e suite}"}
{"_id":"q-en-kubernetes-a2b232003ce728880c810de8f0dd50c5112cccba53e3d7b7c4149e9341a0037d","text":"https://k8s- Failed: BeforeSuite {Kubernetes e2e suite} Previous issues for this test:"}
{"_id":"q-en-kubernetes-0123cb6ac201bd0f2b91e7882abfd3b318b9ffdff935b996f38e63ba0a4f0805","text":"Another flake i found earlier yesterday, just linking to it for posterity, not blocking me right now. TestContainerStatus\nCould you post some error log here? I can't reproduce it with stress test\nHm, got one reproduce:"}
{"_id":"q-en-kubernetes-8635daf913727db16009000c6f67d7c9664802936521decedf5ae6216907dc0d","text":"Kubernetes version (use ): Client Version: version.Info{Major:\"1\", Minor:\"4+\", GitVersion:\"v1.4.0-beta.11\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2016-09-23T22:53:01Z\", GoVersion:\"go1.6.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"4\", GitVersion:\"v1.4.0\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2016-09-26T18:10:32Z\", GoVersion:\"go1.6.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Environment: AWS - CentOS 7.2 x8664 with cloud-init (HVM) (ami-) OS (e.g. from /etc/os-release): NAME=\"CentOS Linux\" VERSION=\"7 (Core)\" ID=\"centos\" IDLIKE=\"rhel fedora\" VERSIONID=\"7\" PRETTYNAME=\"CentOS Linux 7 (Core)\" ANSICOLOR=\"0;31\" CPENAME=\"cpe:/o:centos:centos:7\" HOMEURL=\"\" BUGREPORTURL=\"\" CENTOSMANTISBTPROJECT=\"CentOS-7\" CENTOSMANTISBTPROJECTVERSION=\"7\" REDHATSUPPORTPRODUCT=\"centos\" REDHATSUPPORTPRODUCTVERSION=\"7\" Linux ip-172-31-35-103 3.10.0-327.el7.x8664 SMP Thu Nov 19 22:10:57 UTC 2015 x8664 x8664 x86_64 GNU/Linux Install tools: Others: What happened: when selinux is enabled i was getting following error on discovery container What you expected to happen: i expected node to be able to join master, but nothing was running on 9898 port of master i.e. discovery due the error mentioned above How to reproduce it (as minimally and precisely as possible): use the api which i gave you and do normal kubeadm init Anything else do we need to know: This issue got fixed with setenforce 0\nThis issue got fixed with setenforce 0 but want to track it\ncc\nBTW the issue was on latest Centos iso too downloaded from here\ncc\nthis is to do with secrets, and kubelet log is the place to look for more info, could you please grab and paste it in a gist (it's gonna be fairly large)? Also, the kernel audit log would be handy, do you know how to get this (I forgot)?\nnot sure if this will help\nthe lines are incomplete, it's best if you redirect it to a file and upload it somewhere with a link here.\ncould you also provide the output of ?"}
{"_id":"q-en-kubernetes-a28579b9fb700fe4306bffad7a794ea4c9a1d5eb27439193bb009ba6338ebda5","text":"Kubernetes version (use ): Environment: Cloud provider or hardware configuration: Any OS (e.g. from /etc/os-release): Any Kernel (e.g. ): Any Install tools: Any What happened: If I use a template with the seperator, and if the line with the separator has trailing whitespace characters, then kubectl considers that to be the EOF instead of reading further. e.g. template What you expected to happen: It should not stop reading if there are trailing whitespace characters, it should just trim the separator line and ignore the whitespaces after the . This is how it worked until v1.3.0 kubectl version. How to reproduce it (as minimally and precisely as possible): Just use the template from above and you'll notice that it won't create the Service, but it'll create the Replication controller. Anything else do we need to know: This is on kubectl version 1.4.0\nkubectl regression. cc\nThis bug is general (can be found on get, apply, delete, ...). May be somewhere in resource builder.\nMight be , Janet Kuo wrote:\nConfirmed - is the cause. would you take a look?\nWe may need to cherrypick the fix\nYaml rules are separator must be at beginning of line, three dashes, and new line. Did we change how strict we were in that PR? On Sep 27, 2016, at 7:46 PM, Janet Kuo wrote: We may need to cherrypick the fix — You are receiving this because you are on a team that was mentioned. Reply to this email directly, view it on GitHub , or mute the thread .\nYes, and the test was changed too (originally tests allow with trailing spaces)\nAccording to spec (if my spec-alese is correct) newline + three dashes is the only requirement to terminate a document. Whitespace or prefixes are allowed afterwards, but for it to be valid YAML those must be valid at the beginning of a doc as well (as far as I can tell). So we should split and remove on any line starting with a newline and three dashes - but not consume the remaining whitespace (I believe) , Janet Kuo wrote:\nWe should probably exit non-0 and have an error message if we see something in the file that we don't expect. E.g. is neither a separator nor parseable yaml.\nWe should add a test that uses a YAML document prefix rule and verify that it correctly passes / fails when YAML goes to parse. From reading the spec it appears that anything that is invalid when n--- is present is also invalid in a single document context, and vice versa. , Phillip Wittrock wrote:\nI have rechecked the code, and I have located the root cause, since I am outing for a training these days, I will fix this right now after the training on Friday.\nI am quite sorry about the regression. :(\nshould we allow trailing whitespace characters after , if so we can trim the whitespaces when we read line by line?\nis it possible to include this fix to one of the 1.4 releases? We'd really like our users to have of k8s alpha features (Scheduled jobs is one of them) that can be configured only using 1.4 client; w/o them potentially hitting this regression.\nI will double check this regression, and open a cherry-pick PR.\nI am thinking create a cherry-pick PR for this to include it in 1.4, wdyt?\nI would like to see this in 1.4 since as you note it may break scripts. On Oct 19, 2016, at 8:19 PM, TonyAdo wrote: I am thinking create a cherry-pick PR for this to include it in 1.4, wdyt? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub , or mute the thread .\nDo we have a test case now? :) On Oct 19, 2016, at 8:19 PM, TonyAdo wrote: I am thinking create a cherry-pick PR for this to include it in 1.4, wdyt? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub , or mute the thread .\nwe already test this, you can see"}
{"_id":"q-en-kubernetes-35d3cc1f24250a067ab9790ff51c5c1866bfb8dbfe3e507bd1189b8776b497ae","text":"Just saw this warning today: Not sure if this is only limited to kubemark, or if there are other places we need to fix, too.\nSorry if this sounds stupid, but as expected, won't show this warning - what needs to be fixed here?\nwe have a bunch of scripts and Makefiles which are omitting the currently. We need to fix these by adding the missing s."}
{"_id":"q-en-kubernetes-3bf8a64b76529daf5ef9d30b1f8b10107f0cec27bd9639b72947be09713adaf4","text":"See\nCI has built and uploaded unofficial prerelease darwin binaries for 1.4 using go1.7.1: cc Official binaries will ship as soon as we cut 1.4.1.\nWe have official v1.4.1 binaries available now, too.\nHooray"}
{"_id":"q-en-kubernetes-0ffb29d4daaa35c38b9d541a819bedf43b16ac7b00fde22f3ce7ab4095070468","text":"<!-- Thanks for filing an issue! Before hitting the button, please answer these questions.--Is this a request for help? (If yes, you should use our troubleshooting guide and community support channels, see ): No What keywords did you search in Kubernetes issues before filing this one? (If you have found any duplicates, you should instead reply there.): watch leak Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG REPORT <!-- If this is a BUG REPORT, please: Fill in as much of the template below as you can. If you leave out information, we can't help you as well. If this is a FEATURE REQUEST, please: Describe in detail the feature/behavior/change you'd like to see. In both cases, be ready for followup questions, and please respond in a timely manner. If we can't reproduce a bug or think a feature already exists, we might close your issue. If we're wrong, PLEASE feel free to reopen it and explain why. --Kubernetes version (use ): 0 Environment: Cloud provider or hardware configuration: vagrant on Mac OS X OS (e.g. from /etc/os-release): CoreOS 1164.1.0 Kernel (e.g. ): Install tools: Others: What happened: We ran controller which handles our own ThirdPartyResource. When we hit \"watch of ... ended with: 401: The event in requested index is outdated and cleared (the requested history has been cleared\" error, \"forcing resync\" log is duplicated. This log message comes from . It looks like this goroutine does not exit when we hit the above mentioned error, and continued to work, and running \"resync\" more frequently. What you expected to happen: The resync goroutine must be terminated when watch is closed (i.e., ListAndWatch function returned), instead of leaking it. How to reproduce it (as minimally and precisely as possible): I just run our controller for ~15 minutes. I'm not sure why \"401: The event in requested index is outdated and cleared (the requested history has been cleared\" happens, but from the issues I read, it should be harmless. Anything else do we need to know:\nI think I see what is going on here. If exits for any reason beyond an error from the resync loop no one will tell the resync loop to exit. Furthermore, there is no \"local\" way to do this as it would be unexpected for this function to signal the Done channel as other things outside of it may be relying on it. There are some useful functions around Context () that help to create a \"local\" copy of context that can be used in this situation.\nThank you for reply. Isn't is possible to create local channel in ListAndWatch, and defer close it, and then make resync goroutine \"select\" it too?\nThis issue was labelled as component/kubectl, but actually it is not limited to kubectl. All controllers suffer from this. Only people notice this if they enable V(4) logging.\nThe really annoying thing caused by this issue is that eventually resync is done in every few seconds due to the leaked goroutine. Even if you set 30 second resync period in controller, leaked goroutines call resync too, and you end up having resync call very very frequently. If you have many resources in cache, it really performs unnecessary sync loop. This is the last thing you want.\nthis looks like a real issue. Could be related to Could you help triage? looks like a proper fix.\nI agree - this looks like a real issue. And indeed LGTM as a fix for it."}
{"_id":"q-en-kubernetes-4b2b1ba02a39aa9fbaa88beb20cf1c8940383414c25aeab377ce9dd6ee22ca14","text":"Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG <!-- If this is a BUG REPORT, please: Fill in as much of the template below as you can. If you leave out information, we can't help you as well. If this is a FEATURE REQUEST, please: Describe in detail the feature/behavior/change you'd like to see. In both cases, be ready for followup questions, and please respond in a timely manner. If we can't reproduce a bug or think a feature already exists, we might close your issue. If we're wrong, PLEASE feel free to reopen it and explain why. --Kubernetes version (use ): Master Environment: Cloud provider or hardware configuration: OpenStack OS (e.g. from /etc/os-release): n/a Kernel (e.g. ): n/a Install tools: n/a Others: What happened: now can update a LB instead of deleting it and recreating it. However the security group code is still set to try and create security group rules, and if it gets an error, delete the LB and start over. What you expected to happen: Updating a service does not delete the LBaaS LB How to reproduce it (as minimally and precisely as possible): Enable in the section of Restart a Anything else do we need to know:\nyou're absolutely right, this is a critical issue (if you enable security group management). The dangerousness of aggressive cleanups has already been discussed in detail in this thread: I hope you guys can fix this.\nYup - I have a larger refactor, that ensures this, and a few other edge cases we have found in our development. It is in massive need of a rebase though, so I will need a day or 2 of uninterrupted time to get it together\nAre you still working on this? If not, I can take a look.\nping :)\n/close\nWe just ran into this where Kubernetes recreated an LB with a new Floating IP because the OpenStack API was not available for a few seconds while the LB was updated. Could this issue be re-opened? Since this is a major bug for us, we will start working on a fix for this today.\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThere are no sig labels on this issue. Please add a sig label by either: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. See the . The in method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-74852d535c90953c7828dd3da78bb675b08619f3530bd58221027eebcf627233","text":"PetSets currently support an annotation in their PodTemplateSpec field. From the user documentation: Debugging: If one want to bring up the pets in my petset one after the other, while pausing after each step, confirming that the state is as expected, one may use this functionality. For example, a petset where each pet is a replica of a distributed database; we set the to in the petset template initially. This leads to the following progression: PetSetController creates pet-0 Manually intervene and check that it has initialized as expected. Set annotation to true (equivalent to marking that pet as healthy) PetSetController creates pet-1 Manually intervene and check that it has initialized as expected. ... Blocking the PetSet controller: We may want to ensure that the petset controller is not performing scale operations when a different user-initiated action is in progress (for example, where pets are leaving due to a kubectl drain). This may be used as a mechanism to orphan a pod from a petset controller (for a closer look), while ensuring that a replacement is not created. The use case 1 above can be realized using and is probably easier to do that way rather than use the annotation. I don't see a reason to have that case supported when we can have users scale up a PetSet gradually and manually in an identical manner. It may have been of more importance before v1.4 when the scaler & reaper were not implemented. Due to the second use-case, if it would make sense to continue to have this ability, I propose that we move it from the PodTemplateSpec to the PetSetSpec, where it will eventually reside. This seems to me, like the logical place for such a debug hook to reside. In order to run/pause the scaling operations of the PetSet controller, we could have annotations along the lines of with the annotation defaulting to when absent. /cc\nShould we just support Pause on PetSet (for part of these cases)?\nDo users have to catch the roll out at the right time in order for this to work? For example if I want to pause between nodes 2 and 3 coming up, do I need to be precise in my timing when editing the resource? How does this look from the users perspective?\nNote that Pause allows scaling in Deployments, it just blocks upgrades. That being said, I think Pause will be useful for PetSets in general. Not sure though if we can use it somehow for this specific issue.\nI like the debugging hook for PetSets. If it's optional I don't see any reason to get rid of it.\npauses petset controller if necessary and inspects the state of the system_ With the previous flow, the petset controller was paused after each pet's creation automatically, and here it would be an extra step to pause the controller it if necessary. I don't know how important it is though, but to me it seems easier to understand than a per-pod annotation.\nand I had several discussions with the lifecycle of what would be useful. There are a bunch of use cases that overlap: not a primary concern for most users but helpful (prevent changes) - very important if you are triaging failures. Technically scaling is itself a user action, so if you don't want your pet set autoscaled while triaging failures you need to take action on the HPA like deployments. Since PetSets and ReplicaSets are nothing alike w.r.t. scaling I'm not terribly concerned that pause means slightly different things if for end users it Does The Right Thing control by a third party controller - this is the babysitter controller wanting to reuse the petset controller for core stuff but still have the hook angle. Not thought through enough to design around.\nNote that Pause does not prevent scaling in Deployments, only upgrades , Clayton Coleman wrote:\nJust to summarize: The current behavior with the annotation It is set on a per-pet basis. Setting it to false on any single pod blocks all operations on the petset; scaling, creation of new pets, etc. All pets must contain the annotation, and have its value set to true in order to make progress. By default, it is set to false and makes the petset controller pause after creating one pet. The smallest change we can make for beta: Rename it to (or some better name) Make it optional (with the absence of the annotation meaning that the petset can scale) Surface the reason better when we block on scaling because of the annotation. I still think there is value in moving it to be an annotation on the PetSet resource, rather than having it be in each pod, which is confusing to users. wdyt?\nPutting it on the pod template is clever, but surprising. How many examples are dependent on it? , Anirudh Ramanathan < :\nLast I checked, all the examples had set to true. We should just get rid of it from all examples in my opinion. If we make the default (or the absence of any such annotation) to mean that the petset controller can make progress, we can safely do that. was going to do a pass over the examples after the PetSet renaming. This could be an additional step. Does that sound ok?\nIf it's not found, we assume you want to pause. You're proposing that if it's not found, we default to \"forge ahead\". SG. It should only be used in debugging/testing today. Basically if you don't know how to write your , you'd create a petset with it set, kubectl exec and write it out, scale the petset by 1, kubectl exec ..., and when you have a working cluster, put what you did in an init container and create the petset without it. This is the absolute min. You're also proposing that we move it to the petset. This makes sense overall, but I initially didn't do that because I didn't know how we would end up doing pause.\nWe could leave it there as \"alpha\" and then remove it in 1.6 once we have more time to talk about how upgrades and debugging and pause should work? , Prashanth B wrote:\nYeah the minimum work is to flip the behavior of the controller when the annotation is not found:\nAgree on flipping behavior. , Prashanth B wrote:\nFlipping SGTM. are you suggesting we leave the annotation-name as it is, and don't change it to ? If we are changing it, we should probably change the name from as well, since it doesn't make as much sense anymore. Probably ?\nor healthy rather.\nFlipping default behavior (when not present). Or did I misunderstand beeps proposal? On Oct 27, 2016, at 7:11 PM, Anirudh Ramanathan wrote: Flipping SGTM. are you suggesting we leave the annotation-name as it is, and don't change it to ? If we are changing it, we should probably change the name from initialized as well, since it doesn't make as much sense anymore. Probably unhealthy? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub , or mute the thread .\nRenaming it to beta is effectively moving it to beta, so I think given that we're still confused about how this ties in to the big picture, that might be premature. If it's either move to beta or drop, I'd probably vote for drop, it's a debug hook after all. Renaming it to something else but keeping in beta makes sense.\nsorry, keeping in alpha\nwhats going on with this tag? it's used in the docs here should i use it going forward?\nIt's alpha and may be removed at any time. On Nov 2, 2016, at 3:11 PM, Zachary Adam Kaplan wrote: whats going on with this tag? it's used in the docs here should i use it going forward? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub , or mute the thread ."}
{"_id":"q-en-kubernetes-1da508198b088dbae572d4bdb1fc451b5b409cd873c7dd20d27fd3a66f833b82","text":"Kubernetes version (use ): HEAD Environment: GCE What happened: When I set up I can see the following message: What you expected to happen: master should not be registered\ncc\nLooks like the env var isn't being respected any longer by the master scripts (but the logic for # of expected nodes did adjust correctly).\n/cc I think this may have broken in because it which unconditionally sets the field that tells a kubelet to register with the master outside of the guard that checks .\nFix in progress: . There was a problem caused by unconditional setting on the master in ."}
{"_id":"q-en-kubernetes-b27fd4ba8111ba6c27ce4b735921cc2de35b4fc130a5d3c5eb23c08df1ca0eaa","text":"From . During the firewall e2e test we need to change the service type from to to avoid leaking LB resources. But found it is not working because the field. Error I got as below: If don't delete this field, got instead another error:\nJust saw the TODO comment . Working on it."}
{"_id":"q-en-kubernetes-76ed5d73de657ebb40cede01a073f7c23ca75b99af70ac9c6718b17c4fe55f7b","text":"The logic to account for all pods being ready is incorrectly counting some pods as being controlled by a replication controller. This is causing the 1.3-1.5 upgrade tests to fail. See https://k8s- as an example. Since , the pre-flight checks fail and the rest of the tests never run. I think this was exposed by which changed some things from ReplicationControllers to Deployments. There's a pod being created by a deployment that also matches the selector of a replication controller, and is therefore being incorrectly counted by cc"}
{"_id":"q-en-kubernetes-fc8e76f63a2fff5892f2350711ca71d1b72905b8cfeeb1b70f8a6d90915c5be5","text":"Similar to this is to test the first contributor experience fails on ubuntu 16.04, docker version 1.12.3:\nGot the same error on Linux ubuntu 4.2.0-27-generic ~14.04.1-Ubuntu SMP Fri Jan 22 15:32:26 UTC 2016 x8664 x8664 x86_64 GNU/Linux Docker Client version: 1.6.2 Client API version: 1.18 Go version (client): go1.2.1 Git commit (client): OS/Arch (client): linux/amd64 Server version: 1.6.2 Server API version: 1.18 Go version (server): go1.2.1 Git commit (server): OS/Arch (server): linux/amd64\nWhat git sha of were you building from?\nI have cloned the head. I think the reason might be missing package: gccgo-4.8-aarch64-linux-gnu I'm compiling right now and update on result.\nwere you running or ? runs inside docker, so the gcc error is confusing.\nI run: sudo make release in Ubuntu VM running in VMWare workstation if that matters. Are you in kubernetes-users slack channel?\ndoes fail in the same way?\nDo you need to run , instead of just ? Can you try running the following:\nwell, if I don't do sudo docker does not work :) I need to run as root. I'll try the commands as soon as quick-release ends.\nah, gotcha. you can probably fix the docker access issue by following in the docker manual. that's probably not why it's failing, though.\nmake release still does not work :(\ncan you please run and then paste the results? ( turns up the debugging verbosity.)\nfor me: works (using your PR to fixes the gobindata dep) -local Go build- works (docker build) works if you have Docker running. does not work: on OSX did not seem to give more verbosity.\nV=4 make release did not help :(\noops, that was not the correct environment variable, sorry. I meant .\nhere we go:\nAha, I think I figured it out. Sorry for making you rebuild so many times; the issue still isn't super obvious from the logs. Basically, the bug only occurs if you don't have enough memory to do the docker builds in parallel; if that happens, you fall into this . On the parallel side of the branch, we set the platform env vars within a subshell, so each of the cross-compilation builds won't affect each other. On the non-parallel side of the branch, we don't use a subshell. Inside , there's no overrides done for . As a result, we end up incorrectly reusing the value for , , which doesn't recognize the flag being applied for the darwin build. The fix should be easy; I'll get that sent out ASAP. we probably want this fix for 1.5, since if you fall back to non-parallel crossbuilds we're doing the wrong thing.\nAck. Will keep an eye out for the fix. Please mark it as cherrypick-candidate\nThank you guys!\nI think it is almost there. Got much further in the build (4 GB of RAM). Towards the end there is a failed test that makes the build fail: This is probably a different bug, as your PR got passed the misconfigured compilation.\nFWIW, the worked on ubuntu 16.06. It did ran some tests apparently but the build succeeded.\nI just tried with and it worked fine\nsince we have a PR ready and working, can we please switch this to non-release-blocker?"}
{"_id":"q-en-kubernetes-6a09cd6936e4a7af660ca32687ff8d6b61828a4fd557f0736b3765ead399feaf","text":"On kubernetes 1.4.6 taken from releases. Using CentOS 7.2 Kubernetes worked with a 1.2 version. I just uninstalled and redo installation with 1.4.6. SELinux is disabled Everyhting seems to works (controller-manager, kubelet, and so on...) excepting kube-proxy that returns an error: \"sysfs not mounted\". API is responding (internally and externally), and I can see nodes The command that is launched by systemd is (--v=8 to get a maximum of information): But sysfs is mounted on /sys: I also tried to set /sys as writable (only to be sure, with chmod +x), and also tried to use userspace proxy mode. Note: 1.2.6 version was ok I think that the problem is that sysfs mount \"none\" as label, not \"sysfs\".\nNote: I can now confirm that the problem comes from the \"label\". I did this: Then kube-proxy successfully started.\nbefore setting . This logic was introduced in v1.3 by to workaround a known docker issue That's why \"1.2.6 version was ok\". Because sysfs is mounted as , kubeproxy couldn't find it then reported error. But why sysfs is mounted as ?\nI installed kubernetes on a \"scaleway\" (a very nice cloud provider) VPS cluster. Their VPS uses a init file that mounts fs like that. I opened an issue on their github. But anyway you may mount \"none\" for tmpfs or any ramfs, sysfs, bugfs. What I think is that the mount listing function that is used in kube-proxy code tries to list the mount \"source\" but not the \"type\". For that kind of mountpoint you can use \"none\" as source.\nthe problem is here: You should check .Type and not .Device. \"Device\" is what I named \"label\" or \"source\" because I didn't remember the name ;) Sysfs can mount a \"none\" device as type \"sysfs\", so maybe you can change. See:\nI'll do a PR\nSGTM."}
{"_id":"q-en-kubernetes-bab4044edb7a0e5617350e3d6f93b7c726f673f15e5fb436597db129dafff171","text":"The test is failing in . https://k8s- https://k8s- https://k8s- The reason is that the test expects that all pause containers () should have . However, in many tests, we are starting test containers with . Those pause containers have . Ideally, containers started in previous test should be cleaned up, so in serial suite it should not affect later test. However, it is not always true when CRI is enabled. For example, in the oom message, we could see: There are 2 problems here: Why are containers not properly cleaned up when CRI enabled? I can't even find which test leaked pause container from the log, because there are many test using pause image, and some of them start a lot of containers, such as density test, restart test. The oom score adj test should not rely on process name . However, I can't find an alternative for now, because we don't report pause container status in our api, we don't even know the container id to inspect with docker or CRI. /cc\nThis test is still flaking in cri-serial. Some examples: https://k8s- https://k8s- Need to check the logs to see if the symptoms are the same.\nIt could be that with CRI, it takes kubelet longer to clean up. Or, even worse, that kubelet is not cleaning up properly. We need more information on this on failure. Dumping pods on the node Dumping containers () on the node. Accessing docker directly is not ideal, but it's okay because the test itself only applies to docker. Alternatively, we can get the current running containers from kubelet's /runningPods endpoint. The downside is that this may lag behind the actual state a little bit (seconds). We should also make the test suite run more frequently. It runs once a day, which makes the debugging round-trip overhead too high.\nI can reproduce this process leak by running the \"Network should recover from ip leaks\" on my own node, without CRI enabled. The image used in the test was , and the docker version 1.11.2. The test creates 100 pods, then restarts the docker service by 6 times, with 20s interval in between. After the test ends, there could be multiple processes running left on the node. Those processes were reparented to pid 1, and showed no containers running. /cc\nOn the side note, the test should not be blindly checking all processes. It should, instead, take a snapshot of the system before the test and only check the new processes created by the tests.\nI filed a new issue for the docker bug since it's not unique to CRI and this test is catching it by accident: Leaving this issue to fix the test by not blindly checking all processes:\nThe docker bug is not a regression, and is tracked in . The problem of the test is not necessarily to be fixed in v1.6, at least it's not release blocker. :)"}
{"_id":"q-en-kubernetes-e710d6571fc2bb83bfa33e4a2cce734798a9edc31e924618a5805c0e3b35e1f5","text":"Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG <!-- If this is a BUG REPORT, please: Fill in as much of the template below as you can. If you leave out information, we can't help you as well. If this is a FEATURE REQUEST, please: Describe in detail the feature/behavior/change you'd like to see. In both cases, be ready for followup questions, and please respond in a timely manner. If we can't reproduce a bug or think a feature already exists, we might close your issue. If we're wrong, PLEASE feel free to reopen it and explain why. --Kubernetes version (use ): Client Version: version.Info{Major:\"1\", Minor:\"4\", GitVersion:\"v1.4.5\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2016-10-29T01:38:40Z\", GoVersion:\"go1.6.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"4\", GitVersion:\"v1.4.5\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2016-10-29T01:32:42Z\", GoVersion:\"go1.6.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Environment: Cloud provider or hardware configuration: vSphere OS (e.g. from /etc/os-release):Kernel (e.g. ):Install tools: None Others: None What happened: When I create a deployment with vSphere cloud provider dynamic provisioning, it creates a volume spec with volume path - [datastore] kubevols/volume which contains a white space in it. In this case the cloud provider queries the getDeviceNameFromMount() to return the path of the volume mounted. Since getDeviceNameFromMount() queries the filesystem to get the mount references, it returns a volume path \"[datastore]040kubevols/volume\". Later the kubelet searches for this volume path in both the actual and desired states. Th actual and desired states contains volume with path \"[datastore] kubevols/volume\". So, it couldn't find such volume path and therefore kubernetes stalls unable to make any progress further similar to one described in . Also similar issue described in What you expected to happen: All the pods in the recreated deployment comes up. How to reproduce it (as minimally and precisely as possible): See above. Anything else do we need to know: None\ncc"}
{"_id":"q-en-kubernetes-b8d2c0da68304e83135ba886da3cc34043a1f75faa72e9753c7e99fa612dae72","text":"From etcd3: May need to retry poststarthook a few times to avoid bootstrapping failures like this."}
{"_id":"q-en-kubernetes-5941efe129f89aeac5a57a181211dbed35027d19160f63506020d4b51036a413","text":"adler when brute-forced turns out to break really fast (). There are a couple of places we use it in the repo: We should probably replace it with fnv. I tested fnv and it seems to be slightly more expensive but also a lot more stable than adler:\nJust work out a patch to fix this, please take a look.\nassigning this to you since the only place left to use adler is in and it's a legacy helper.\n/assign\n[MILESTONENOTIFIER] Milestone Labels Incomplete Action required: This issue requires label changes. If the required changes are not made within 3 days, the issue will be moved out of the v1.8 milestone. kind: Must specify at most one of ['kind/bug', 'kind/feature', 'kind/cleanup']. priority: Must specify at most one of ['priority/critical-urgent', 'priority/important-soon', 'priority/important-longterm']. <detailsAdditional instructions available <a href=\"\"here</a</details\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. Prevent issues from auto-closing with an comment. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nThis was removed in"}
{"_id":"q-en-kubernetes-490ac367d4ab83a139fc1f320c659a499899ad8ab67ed3319f693dd98af4d529","text":"Our scalability tests started failing on Thu last week: https://k8s- [250 run passed, 251 failed and since then everything failed] I started looking into it and it seems that something happened to how we are handling watches (i.e. I'm seeing a huge number of watches failing with \"err too old\" errors (which I didn't see before).\nAs an example in run 258 of kubemark-scale, I'm seeing this log in the test (after 5 minutes of running test); This is extremely strange and unexpected. I looked into apiserver logs, and there is expected number of POST and PUT operations on pods (and no DELETE and PATCH operations until then). Until then, there were alltogethere ~60.000 operations on pods. However, in 5000-node cluster, the size of watchcache of pods is supposed to be 250.000, which means that at that moment we should have all the history from the beginning in watch cache. And still we are observing those errors. Something must have changed around that recently...\nAs an example, logs from apiserver from a given node: This means something is really wrong: the first watch started at 45, then after it finished, we tried starting watch from rv= and it failed with \"rv too old\" the list, returned rv= from which we started the next watch. If the watch cache size for pods would be 250.000 (50*cluster size), there there is no way that with current rv=, the rv= would be outside history window...\nI had a hypothesis that it might be the problem of breaking watch to etcd. But in that case there should be a warning \"Terminating all watchers\" log in apiserver, and there is no such log. It also seems that the size of cache is being set correctly...\nSo I tried running large kubemark locally and it seems that the problem is that the size of cache is too small. I few more logs, and here is what I'm seeing: So it seems that we are correctly initializing cachesizes, but then watch cache capacity is 100 (which explains why we are seeing all those \"rv too old\" errors. I will try to debug why the capacity is so small.\nOK - I found it. contains the fix."}
{"_id":"q-en-kubernetes-8bc52aafefd8d4f5fd7860ac78effc748d00a617bb0944080ddb6bdd9c32be6b","text":"The description of says: However the implementation will interpret the name as a regex. So for example \"...\" will match any three-letter string. This is not what I understand by \"exact match\".\nWill open an PR fix this comment"}
{"_id":"q-en-kubernetes-c1af9be05ec6d7dedbb1ae7294344c5d68ab0d6ecd719d2d691350d408b871c8","text":"If in line line is executed also for step 2.2.1 if the tool failed during first run of upgrade. I guess we should also check for the step and compare it to version."}
{"_id":"q-en-kubernetes-4b043ab61632a183920bf62a5be84e11c0702831755fb3808ab5d8790cb5d9b1","text":"At the very least it should be clear when (timestamp) the migration started and finished."}
{"_id":"q-en-kubernetes-f063045a15f2f641d3790e2b25e476dde91e4310f364256ec78b76d0b285e9b4","text":"https://k8s- do you know the statefulset owners?\ncc\nI did not see this happening until recently.\nThe controller misses that the first pod transitioned to unready and proceeds to create the second pod. The test uses a client to hit the api server and will scale up the statefulset as soon as the broken pod is observed but the controller uses caches so it may observe the scaled up statefulset quicker than the updated pod.\nI will update the test to ensure that we wait until the unreadiness is observed on StatefulSet.Status.Replicas field prior to the scaling operation\nSGTM"}
{"_id":"q-en-kubernetes-01be90805f5b079723b5db36f29de388e9c7d6fb4142d8302d92a3b827f9a2b4","text":"https://k8s- replica set controller maybe?\nThe rc seems to be updated almost a second after the failure. Maybe the test needs to retry\nAnd the real log for the failure is\n/assign flakes PRs at least once a day The can filter down to focus on a single test:\nFor when you have some time.\nReassign to you\nOutput of has been change in 1.6. That's why upgrade tests are failing. But I'm not sure why it is not failing 100% from the triage board. Do you know why? Output of 1.5 or earlier: Output in 1.6 or later:\nThe timeout is now. Is it reasonable to give it a little more time?\nThat timeout waits for the pods to run based on the API server. The rc controller is using its own cache so it may lag a bit before it actually updates the rc status. The test needs to retry for some time if the output we get back from describe does not match.\nIn other word, should also have its own retries.\nSGTM. Going to send a fix today.\nFYI, this test continues to flake PRs at least once per day:\nshould it show up in\nHm, this seems to fail in the upgrade tests. this failure seems to be related to an image name change as you already pointed out before. Mind finding what changed the test to use that image and fix the image name?\nIt is actually the key name has been changed in the key-value pair, not the value. As I pointed out in before expected key: now expected key:\nHm, this is unfortunate. These kubectl tests should really be unit tests.\nthe 24 hr test history is largely focused on failures right now, and this is highlighting flakes specifically (aka which tests are both passing and failing at the same commit the most often). Please follow the reports I'm sending to kubernetes-dev and related sigs. Also you can generate this info yourself using PS: also try out for finding failures (I'm sending out a separate email highlighting the biggest sources of failure). PPS: we are in the process of creating infrastructure to automate creating these reports, but automation never works as well as a fellow human asking for attention.\nWe still need to patch the release-1.4 and release-1.5 branch to make the upgrade tests passing. I will send PRs later this week.\nWe didn't fix\nThe change mentioned in was introduced in 1.6. The change covers describer and kubectl e2e test. After discussing with we think we are not going to backport the changes to 1.4 and 1.5, instead we are going to skip the tests in version skew cases. WDYT?\nshould fix this issue\nlooks like the PR is resolved, should we close?\nPlease wait until fix to be cherrypick into 1.4.\nSure! Which PR is this?\nIs it possible to cherrypick to 1.4 by just changing the milestone to 1.4?\nno you need to create a new PR with the target branch of 1.4\nReady to close?\nYup. /close"}
{"_id":"q-en-kubernetes-1629f917e5afb398a2ce58b7c68cfc8414afc810c0142575c793d4142fcdf0fe","text":"We need to gracefully handle the case when a Deployment specifies an absolute value of MaxUnavailable larger than its spec.replicas. I don't think we want to add validation for this, simply collapsing maxUnavailable to spec.replicas would suffice.\nAre you working on this? I am glad to take this if not.\nfeel free to work on it"}
{"_id":"q-en-kubernetes-05a35484a6fb3e3b632cb2c2bc09c0a13c7491426c4df503e3be004f93099a07","text":"When patching, in the type decoded from ToUnstructured mismatch with the one decoded form json.Unmarshal for numbers. Example: port: 80 in original object has type int64 port: 80 in original object has type float64 This is the root cause of"}
{"_id":"q-en-kubernetes-092a2c271f463c8f9716c9adec149fd83ba8b319995452bb190dfdefdb7e8e39","text":"<!-- Thanks for filing an issue! Before hitting the button, please answer these questions.--Kubernetes version (use ): master What happened: The functionality is hidden in a large function, and not explicit, testable. What you expected to happen: The logic of should read more clearly, with tests for individual filters.\nhas a PR for this .\nFew more things include removing limits in entries like\nyeah, im not sure why the thing is there, after all they are just appending to the slice and the limit wont have an effect , from what i can tell, will it? probably good idea to remove unless im missing something.\nIts part of part of PR too."}
{"_id":"q-en-kubernetes-fe8cd1417ccecb69a119143da34e7328b16728970aada998527fd4ee6dea436e","text":"Kubernetes version (use ): Environment: Cloud provider or hardware configuration: GKE OS (e.g. from /etc/os-release): container-os Kernel (e.g. ): Linux gke-wordpress-cluster-default-pool--m764 4.4.21+ SMP Fri Feb 17 15:34:45 PST 2017 x86_64 Intel(R) Xeon(R) CPU @ 2.60GHz GenuineIntel GNU/Linux Install tools: Others: What happened: ` I was able to bring up wordpress okay the first time, except GKE wasn't creating loadbalancer ips due to a quota issue, which I resolved, note at this point the mysql pod was up and had attached to it's volume. Upon deleting the wordpress deployment and creating it again I started getting the above errors. I deleted the mysql pod as well and brought it up again to see that it had the same issue. The volumes are backed by a gluster cluster on GCE. Looking at the brick logs on one of the gluster nodes I see I've tried restarting kubelet on the node. I can't find the kubelet log file on any of gke nodes. And I don't know how to get the kube-controller and apiserver logs from the master (GKE). I suspect it's a failing of the glusterfs client on the gke nodes? What you expected to happen: Deployment to mount the volumes successfully. How to reproduce it (as minimally and precisely as possible): Not sure how, but I've run into this intermittently Anything else we need to know: The volumes are backed by gluster\n/cc\nIf you delete pod and create pods soon, it is possible that you could see some errors like \"timeout expired waiting for volumes to attach/mount\" temporally. Could you wait for a few more minutes to see whether the pods could be started correctly?\nI actually tried that, deleted the pods, then waited 5 minutes and recreated them with the same error. Maybe it wasn't long enough\nCan you post output of ? It may have some other events for failure..\nI deleted the cluster this afternoon, but I checked constantly and those were the only errors, Timeout on mount and a failed to sync error\nIf you use level 4 log, it should also give the detailed messages of failed mount so that we can figure out why it failed.\nCan you email me your project name, cluster name, and zone information? I can take a look at the master logs to see if there's anything useful there.\nThis is starting to happen more frequently..\ncould you provide more details about in what situation it happened frequently? Kubelet log could give us more information. If you use GKE cluster, run \"journctlctl -u kubelet /tmp/log\" to get the log. You could also share with us the log if it is ok.\nThis seems to be the relevant part of the log I re-created that and my pods sprang back up. Weird thing is I had created that, along with the pvcs and pvs but seems it's somehow getting deleted. There's almost certainly certain I'm doing wrong but I'm not sure what. ]\nHI Jonathan, can you clarify what you had to recreate?\nThe glusterfs endpoints , Michelle Au <:\nI have a very similar problem running OpenShift 3.4 in AWS EC2 doing the same exact thing. Brand new OSE install [root master]# oc version oc v3.4.1.12 kubernetes v1.4.0+ features: Basic-Auth GSSAPI Kerberos SPNEGO Server https://172.31.40.126:8443 openshift v3.4.1.12 kubernetes v1.4.0+ POD in question is the docker-registry (kind of important for OSE to work) I've followed this to see if it would fix the issue: But no luck or change. Error syncing pod, skipping: timeout expired waiting for volumes to attach/mount for pod \"docker-registry-3-afbbt\"/\"default\". list of unattached/unmounted volumes=[registry-storage] then: Unable to mount volumes for pod \"docker-registry-3-afbbt_default(-1a45-11e7-abf8-)\": timeout expired waiting for volumes to attach/mount for pod \"docker-registry-3-afbbt\"/\"default\". list of unattached/unmounted volumes=[registry-storage] I have no idea what log(s) to look at, this was of no help when I deployed a new docker-registry: oc logs -f dc/docker-registry Any help or guidance would be greatly appreciated as this is a blocker issue for me. thanks and best wishes,\nare you using glusterfs for registry-storage volume or other type of volume? Could you share us your kubelet log if possible over slack Thanks!\nHi Jing Xu, Would you know where this log would be? Or is the name kublet<somethingso I may find it for you? I just requested to join the slack channel, as soon as I'm signed in I will share the log. Probably tomorrow as I'm on the east coast USA Thank you! Joel\nFor you are using GKE with GCI image, you can use \"journalctl -u kubelet \" to get the log. Otherwise, it should be in \" , shepp67 <:\nI also experienced this. Kubernetes version:Cloud provider or hardware configuration: OpenStack OS (e.g. from /etc/os-release): Ubuntu 16.04.1 LTS Kernel (e.g. ): Install tools: custom That kubernetes worker vm actually had volumes attached. I saw it on OpenStack dashboard. But: and actually: Besides, this is a new worker vm. I have other worker vms already running, using cinder volumes and there were no such errors. I use Helm to deploy this Pod. I have deployed it earlier and removed it. But it was 2 days ago. I found the solution here: . On that worker vm I run: Edit: or upgrade Kubernetes to = 1.6.0-beta.3 I don't know if this is the solution to issue.\nI'm seeing this same error on all my manually launched private Ubuntu based clusters (6), running 1.5.3 and 1.6.0. In my case the messages reference only stopped pods launched by CronJob Jobs. The messages indicate issues mounting every volume in the pod including PVs and node-local volumes. These pods, and future ones launched by these CronJobs, are fully successful when they run. It is only after they run to successful completion and exit that these messages show up. Glancing at the kubelet code, this appears to be coming out of some housekeeping/synch loop.\nI've identified a bug and am working on a fix for when \"Unable to mount volumes for pod...\" + \"Error syncing pod\" happen indefinitely BUT the pod is actually running just fine, as observed by . For others, if you see this message but your pod is Running, you can safely ignore it. It's caused by a small limitation of the reconstruction of state (what volumes are mounted) so there is a chance you will see this only if you restart kubelet while your pods had volumes mounted. correction: you say they even Completed, they weren't just Running fine. So yeah, it's just misleading logging and not really an indication of anything breaking\nBecause the error message is very generic, I think there might be different problems mentioned in this thread. 's issue could be fixed by 's PR are you still have any issues with mount? could you also let us know whether you still have the issue? Thanks! Jing\nHi my issue was related to which helped me sort out. I was creating the gluster endpoints without a service and turns out the kube-controller deletes all such endpoints that are not attached to a service at regular intervals. So after every so often when a pod went down and tried to come up again and mount its respective pv it failed because it couldn't connect to gluster. The fix is to create a service for the gluster endpoints.\nI also just ran into such timeout issue: After running a load test against my postgres backed web application, the database didn't respond anymore for still unknown reasons. According to kubernetes it was up and running The log showed a bunch of these messages. Not sure yet why but possibly it couldn't access the volume anymore: Short after that, the pod entered \"error\" state: The event log showed: This error disappeared after a few minutes (not sure how long exactly, but it took a while), but then I got the same error as described here: I couldn't figure out anything more on my own. Nothing in the syslog, nothing more than this in the kubelet log. Finally I rebooted the machine which fixed the issue.\ncc do you know what error of the network not ready? It might be helpful if you could share the kubelet log with us to debug. Thanks!\nAFAIK, the \"network not ready error\" is benign, kubelet will retry after this error () cc\nI hope I didn't miss some secrets in there, quite a talky log. But here it is: The database didn't respond anymore because it ran OOM and was killed, as also shown in the log. Although it appears like it was OOM killed after it became unresponsive in the log.. Either way, a bunch of other things got killed too, so granted - \"not ideal\" situation :). Yet it shouldn't drop the volume and should re-attach it.\nI think you are hitting the issue mentioned above. He has a PR to fix it. I will make sure it can merge soon. Thanks!\nGreat, thanks! Do you know roughly when this will be available in GCE?\nyou mean GKE or GCE? For GKE it will take a little bit longer. On GCE, when the PR gets merged, the fix should be available when the next release is out (I think should be soon)\nRunning on GKE. I'd rather not launch my product with this bug around. Is there a workaround for it or something I can monitor for? Or maybe it could get hot fixed on GKE? If there is no workaround, this effectively renders GKE useless for stateful stuff.\nIt's hard to say if volumes are actually being dropped/unmounted or if the FailedMount errors are wrong & misleading, so I wouldn't go so far as to say it's broken because of some volume issues. It could be something else entirely. If nothing else, the PR jing's referring to will clean up the event noise and help us debug the root cause. In the meantime, I wouldn't trust the FailedMount events to be accurate, so if your applications are unresponsive, there may be other causes worth considering\nWell, I only can say that my volumes were dropped and it didn't recover on it's own (only after reboot). Dropped a seen by kubernetes, dunno if the volumes were still attached but they weren't usable by my scheduled database. To be clear: I'm not \"complaining\" and this was just test/R&D, but it would this is the scariest kind of issue for me to happen in prod, given that it's a managed service where I can't find a workaround on my own. Rebooting fixed it in my case, but that's a big hammer..\nYou don't need to reboot the node. Delete the pod and restart it should work for this issue. Is this workaround ok for your case? , discordianfish <:\nGood to know, thanks! Will try this next time.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. Prevent issues from auto-closing with an comment. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nHi, I'm facing the same issue with k8s 1.9, installed by kops on AWS. volume created successfully on AWS and attached to the right node, but the pod is failed to start.\n/remove-lifecycle stale\ncould you please provide more details about your issue? What are the steps to reproduce it? Thanks!\nyeah can you open a new bug for EBS and cc me with details? I was hoping we fixed most of EBS problems in 1.9 - I will be more than happy to walk through and find what is causing volume to not mount. As I understand this is different from typical EBS problems where volume fails to attach. In your case, volume is being attached to right node but it is not being seen by kubelet.\nI found the issue, and it was the fact that i used c5 instances with nvme, this is why the volume never been attached as k8s looking for /dev/xvdXX and with nvme that path is different. I did notice that in 1.9 k8s should support nvme natively but it seems that it's still not stable. thanks\noh weird. You should open a bug about NVMe support then. I was thinking it would just work.\ncc Is there an issue related to nvme support? please let us know any issues you still have with 1.9. Thanks!\nThis is nvme on AWS, not related to the work is doing on gce.\nThanks Definitely related to NVMe support. Ran into this trying to mount a gp2 EBS volume on a node which was an m5 EC2 instance.\nI think I am seeing like this with , but unsure of \"why\" at the moment. It would seem works perfectly fine.\ndid you guys had a chance to open a BUG for NVMe support on EBS? If you do - please paste output of and PersistentVolume spec. We can't fix it unless we know bit more. :(\nsorry, we worked around this by reverting back to m4 class instances. It should be simple enough to reproduce however... create a k8s cluster w/ m5 EBS backed nodes and create a simple pod w/ a persistent volume. You'll notice that the pod will never launch and you should be able to debug further. If I do run into this again I'll be sure to get some more data for you.\nI too have the error, error syncing pod. kubectl describe pod kafka1-kafka-0 .... .... .... Warning FailedSync 5m (x141 over 35m) kubelet, 10.186.. Error syncing pod Warning BackOff 57s (x164 over 35m) kubelet, 10.186.. Back-off restarting failed container kubectl logs kafka1-kafka-0 -p export KAFKABROKERID=kafka1-kafka-0 export KAFKAADVERTISEDLISTENERS=PLAINTEXT://172.30.*.:9092 /etc/confluent/docker/run sh: 3: /etc/confluent/docker/run: not found"}
{"_id":"q-en-kubernetes-f59d598d5d92681fe549c4f3adb9f9b079e6c980b7927f7c2c210b3c06d36aca","text":"Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG REPORT <!-- If this is a BUG REPORT, please: Fill in as much of the template below as you can. If you leave out information, we can't help you as well. If this is a FEATURE REQUEST, please: Describe in detail the feature/behavior/change you'd like to see. In both cases, be ready for followup questions, and please respond in a timely manner. If we can't reproduce a bug or think a feature already exists, we might close your issue. If we're wrong, PLEASE feel free to reopen it and explain why. --Kubernetes version (use ): Environment: Cloud provider or hardware configuration: bare-metal OS (e.g. from /etc/os-release): Ubuntu 14.04 Kernel (e.g. ): Linux dev-k8s-controller-p1 4.4.0-53-generic ~14.04.1-Ubuntu SMP Fri Dec 2 03:43:31 UTC 2016 x8664 x8664 x86_64 GNU/Linux Install tools: chef cookbook Others: Docker 1.12.6, calico 2.1 CNI binaries: kubelet config: What happened: When trying to run pod error message is generated: So it seems that kubelet is trying to find CNI binaries in paths: /opt/loopback/bin /etc/cni/net.d /opt/calico/bin But it should look at since this is the default valur for parameter. What you expected to happen: Pod starts. How to reproduce it (as minimally and precisely as possible): Run pod: with kubelet 1.6 and CNI binaries in those path (default for calico): Anything else we need to know: After copy to and pod starts fine.\nGood, very helpful to me\nWhen set and not set , default path of CNI binaries will not take effect, the path specified by will be used instead. We should use and instead of if use CNI plugin.\nI will push a PR to deprecate the flag. It is confusing. Closing this issue.\nWas that ? It was closed without merging. I agree should be deprecated.\nI think is needed before is removed. The flag is still not deprecated, I will send a PR to deprecate it."}
{"_id":"q-en-kubernetes-0318647ed38ee346e4394847572fe3c715455e2328b33549b60619adece32e73","text":"In order to support running tests in parallel, we randomized NodePort selection. There appears to be some evidence that this is not sufficient and is causing e2e test flakes: https://k8s-\nHi, As a need contributor, I'm interesting of solving this issue. I already check the current code, and propose to retry the service creation until success or all possible ports (between FederatedSvcNodePortFirst and FederatedSvcNodePortLast) have been tried. If this approach is fine for you, I can submit a PR soon.\nHow do you plan to check if the NodePorts are available? In our presubmit test environment, tests run across many instances of federation control planes in parallel, but share the same underlying clusters. So this approach might be viable depending on how and where you plan to check for these used NodePorts.\nmy approach for getting an available port is a \"Dummy\" approach: create a list of possible available port, get one of them randomly, then try to create the Service. If Service creation failed. try to use another possible available port, until service creation success or all possible available ports have been tested.\nSounds reasonable to me. Go for it. A couple of caveats: need to create a real list, you could just use the range boundaries. careful about service creation failure. You need to check if the service shard creation failed in the underlying clusters and rollback all those shards before retrying.\nthanks for the caveats, I will do soon a PR.\nHi, what is the state of this issue? Is working in this issues with the others issues related to e2e tests?\nhas a PR to fix this issue -"}
{"_id":"q-en-kubernetes-372f31bbae35c3eba7fa05d29bef438fa01153dd17bab3d567afcb0b28d24b93","text":"We copy in BUILD files from the saved directories. The BUILD files should be preserved. This was hit in"}
{"_id":"q-en-kubernetes-cec61f8ec69a9d7d7e36e7d5da8c4f6dcc87324a534945b6990d0e34da434212","text":"The web and instructions for [Installing and Setting Up kubectl] need a slight tuneup. The homebrew instructions should reflect a few different options for Bash on macOS. bash-completion is for Bash 3.2 (macOS default), and bash-completion is for Bash 4.1+ (upgraded GNU Bash) I lost an hour of my life to this; I hope this helps others. What happened: By following the homebrew install instructions on this page, then entering and tabbing, this is the output: What you expected to happen: By following the same path as above this should be the output: Anything else we need to know: Currently the instructions assume the default Bash on a vanilla macOS install; while I understand this being most-likely scenario, the default version of Bash is shamefully old. After reviewing to options in homebrew it appears there are 2: -completion is for Bash 3.2 and -completion is for Bash 4.1+ For those of us that cannot live without all the options, this is how we do: Install: Configure Then the needs to be updated to reflect this changes: TEST a new tab (the is sourced-in) and tab twice We then get the expected result. The requested details follow: Kubernetes version (use ): Environment: personal/dev Cloud provider or hardware configuration: NA OS (e.g. from /etc/os-release): macOS Sierra Kernel (e.g. ): Install tools: homebrew [Installing and Setting Up kubectl]\nSo were you running bash 3.2 or bash 4 prior to this? It looks like the instructions are for bash-completion which should be for the bash installed by default in macOS. /assign\nI always install the latest Bash, currently 4.4.12(1)-release. Apple's is too old and I'm not going live in their cave; most people won't - that's why homebrew is a huge project. Correct, the instructions are for the stock version of Bash (presently 3.2.57(1)-release). My suggestion is to include instructions for both versions. Mostly to keep others from wasting the same time. I've already done the testing so folding in the new details should be pretty straight-forward.\nI've opened to update the kubernetes documentation, and to potentially update how homebrew works to make this clearer.\nDocs have been updated: /close"}
{"_id":"q-en-kubernetes-10a8babb28c4f6ea333338ba9c51d88b95dc0d525c3b45d618113f0ab8d21a82","text":"seen in https://k8s-\nShould be fixed by\nI believe this is fixed?\nYes, I missed this one."}
{"_id":"q-en-kubernetes-d13ccc3944e3d6c7481ff49d201d25638bf4d84e13a1a49d2028bb29d8f9531f","text":"https://k8s- Failed: TestServeExecInContainer {server}\nShould be fixed by\nhttps://k8s- Failed: TestServeExecInContainer {server}"}
{"_id":"q-en-kubernetes-f719394191e6dfa6ab5d9c8d66dff792067fac38082fdfe991b0e433d63e8030","text":"On , will fail to evaluate a filter when the member does not exist: For instance, if I were to execute Where not all in that namespace had the annotation, this would fail with the error message: This is confusing as it leads me to think more than one item was selected, so while it is not technically incorrect the error message is misleading. I would furthermore expect to more gracefully handle this case -- a missing here means that the filter should fail, not error.\nWhat do you mean by ? How is it different from ?\nAs in, don't add that item to the list of items on which the following access is made. When I do: And there is no annotation on , I just get nothing back -- no error, just nothing. Similarly, i would have expected the filtered expression to give me the for all of the pods that have an annotation with value . That, anyway is what I understand the JSONPath syntax to mean. When you write You are saying:\ncc since you authored and I'm not sure how we want to fix this.\nTaking a look now\nit looks like you need to do this (note that the annotations portion is wrapped by ): I wrote a unit test and it worked as expected. PTAL!\nI'm pretty sure that was a typo on my end but let me check -- do you have a reproducer script?\nI see an entirely different error today... weird:\nOk so you have 1 object with annotations, and 1 without. Let me look into that case.\nLet me add another annotation to\nNo difference if both have annotations:\nOk, I'll have to test using kubectl\nI'm the Origin bundled but I don't think there are any patches on JSONPath there:\nI'm able to reproduce in the unit test when 1 object has annotations and the other does not, if allowMissingKeys is true for the jsonpath parser.\nI see. Let me know if I can be of more help.\nI think I have a fix. PR coming soon.\nPossible"}
{"_id":"q-en-kubernetes-65caa9d64a4c73df5022d80330fa6a1add34a56f25ed4e6ffce49deea129d515","text":"Kubernetes version (use ): Kubernetes HEAD Environment: Cloud provider or hardware configuration: linux OS (e.g. from /etc/os-release): Kernel (e.g. ):NAME=Fedora25 What happened: apiserver panic What you expected to happen: No panic How to reproduce it (as minimally and precisely as possible): Testcase is going to be sent as a PR\nFixed by"}
{"_id":"q-en-kubernetes-aa75d975d992a909f430b0bed93e3d2043e7547cce9241b8dcea96441c405e96","text":"<!-- Thanks for filing an issue! Before hitting the button, please answer these questions.--Is this a request for help? NO Note: Please file issues for subcomponents under the appropriate repo What keywords did you search in Kubernetes issues before filing this one? (If you have found any duplicates, you should instead reply there.): kubernetes, azure file, chinacloudapi Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG REPORT <!-- If this is a BUG REPORT, please: Fill in as much of the template below as you can. If you leave out information, we can't help you as well. If this is a FEATURE REQUEST, please: Describe in detail the feature/behavior/change you'd like to see. In both cases, be ready for followup questions, and please respond in a timely manner. If we can't reproduce a bug or think a feature already exists, we might close your issue. If we're wrong, PLEASE feel free to reopen it and explain why. --Kubernetes version (use ): master Environment: Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Ubuntu 16.04 Kernel (e.g. ): Install tools: Others: What happened: We try to mount azure file to pod, and get the following error That is because azure china's storage DNS suffix is rather than What you expected to happen: Kubernetes should provide some method to let user specify storage dns suffix, so we can use Azure file volume How to reproduce it (as minimally and precisely as possible): Any Azure cluster in china can reproduce it. Anything else we need to know: kubernetes has hard coded in shows a relative issue\ncould you help confirm this?\nOne possible way is configuring DNS suffix via annotations of PV. If thinks this way okay, I'd like to create a PR to fix it.\nazure file was implemented without using resource group or azure cloud provider. The endpoint can be found at One fix is to use azure cloud provider and choose the file service endpoint from azure environment as I did in blob service . Let me know if you can help.\nI'm not familiar with azure cloud provider and volume codes. I'd like to try to dig into it to see if I can fix as the way you suggested."}
{"_id":"q-en-kubernetes-f28a055e094ad8cab6090a9353bf9b81626590240635476010468b49904b9bc1","text":"Is this a BUG REPORT or FEATURE REQUEST? BUG REPORT Kubernetes version (use ): kubernetes/master using CRI with docker What happened: Create a pod that uses a readiness probe that does an exec to command that doesnt exist See the following in logs: What you expected to happen: No error in log, expected the event to properly be thrown. How to reproduce it (as minimally and precisely as possible): See above. Anything else we need to know: Nope.\nthis is a regression.\ndebugging, will post a fix once found.\nOther reports: ,\n-- thanks for the other reports. at least now we have a reliable reproduction.\nThe problem appears to be the following: has a that correlates containers to a . It uses the type . event recorder library we use in the prober works with . If it doesn't get one of those, it expects the passed in object to implement List. As a result, we are passing a v1.ObjectReference the recorder was not expecting.\nThe bug may have been introduced in , which was there since v1.6.0.\nOpened for a fix.\nKeep this open until 1.6 is patched.\n/remove-priority P0 /priority critical-urgent (I'm not actually sure this is the right priority, just trying to remove the old priority/PN labels)\nClosing since has been merged."}
{"_id":"q-en-kubernetes-31ac89f43517e9b52bae95bbf6a3f87a02afdf8494c951688a4cda2fbf3b7019","text":"<!-- Thanks for filing an issue! Before hitting the button, please answer these questions.--What keywords did you search in Kubernetes issues before filing this one? (If you have found any duplicates, you should instead reply there.): selflink kubelet selflink default selflink manifest selflink Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG REPORT Kubernetes version (use ): admin kubectl version Client Version: version.Info{Major:\"1\", Minor:\"5\", GitVersion:\"v1.5.7\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2017-04-27T10:00:30Z\", GoVersion:\"go1.7.5\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"5\", GitVersion:\"v1.5.7\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2017-04-27T09:42:05Z\", GoVersion:\"go1.7.5\", Compiler:\"gc\", Platform:\"linux/amd64\"} Environment: Cloud provider or hardware configuration: AWS OS: Debian GNU/Linux 8 (jessie) Kernel: Linux ip-10-150-109-115 4.4.41-k8s SMP Mon Jan 9 15:34:39 UTC 2017 x86_64 GNU/Linux Install tools: kops 1.5.3 Others: What happened: When running the selfLink for pods started from manifests were incorrect. Example: What you expected to happen: To receive a correct selfLink: Example: How to reproduce it (as minimally and precisely as possible): I believe simply using a pod manifest to start a pod will cause this issue as kubelet seems to rely on defaults when starting pods from manifests. Anything else we need to know: I believe the below linked line needs to be corrected:\n/assign I reproduced it. Will fix. Thanks for reporting!\nGitHub didn't allow me to assign the following users: xiangpengzhao. Note that only can be assigned. <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. I understand the commands that are listed . </details"}
{"_id":"q-en-kubernetes-735b73e1445875f0744f7d639f99b01e93a38ecea4de40b47921bcb778960a82","text":"When a node is deleting a pod, the NodeRestriction admission plugin ensures the pod is bound to the node. It first checks the server cache for the pod. In the case where a pod was created and immediately deleted, it may not yet exist in the server cache. Before forbidding, we should do a live lookup. https://k8s- cc"}
{"_id":"q-en-kubernetes-d96d95df77a44fb7254651819de7a7da7fdea83bed17fd40f1db55b495f9afe8","text":"<!-- Thanks for filing an issue! Before hitting the button, please answer these questions.--Is this a request for help? (If yes, you should use our troubleshooting guide and community support channels, see ): Note: Please file issues for subcomponents under the appropriate repo What keywords did you search in Kubernetes issues before filing this one? (If you have found any duplicates, you should instead reply there.): Is this a BUG REPORT or FEATURE REQUEST? (choose one): <!-- If this is a BUG REPORT, please: Fill in as much of the template below as you can. If you leave out information, we can't help you as well. If this is a FEATURE REQUEST, please: Describe in detail the feature/behavior/change you'd like to see. In both cases, be ready for followup questions, and please respond in a timely manner. If we can't reproduce a bug or think a feature already exists, we might close your issue. If we're wrong, PLEASE feel free to reopen it and explain why. --Kubernetes version (use ): master Environment: Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: What happened: Currently, the YAML template for both npd and fluentd service do not have toleration, this may cause the daemonset cannot be started on a node if it is already selected to run , it is better to add toleration for both npd and fluentd so as to make sure those daemonset can run on the nodes even if it is for . All other daemonset in kubernetes also have such tolerations. What you expected to happen: Should add toleration for both npd and fluentd. How to reproduce it (as minimally and precisely as possible): Anything else we need to know:\nThere are no sig labels on this issue. Please by:<br(1) mentioning a sig: <br(2) specifying the label manually: <br<brNote: method (1) will trigger a notification to the team. You can find the team list and label list\nYou sent PRs to fix this, right? Can this be closed now?\nthe PR was not committed yet, can you help lgtm and approve it?"}
{"_id":"q-en-kubernetes-72b6f4033d47262868bfa64c947c8c44e063353890ba7dff6a2a083bcacb672d","text":"We have moved to non-global schema in a number of places in 1.6 and 1.7. It is easy to forget registering the unversioned types like , leading to encoding errors in error cases. The registration of the unversioned types should be automatic as for the types, i.e. in .\nThere are no sig labels on this issue. Please by:<br(1) mentioning a sig: <br(2) specifying the label manually: <br<brNote: method (1) will trigger a notification to the team. You can find the team list and label list\ndo we need to close out this issue? (since referenced PR has merged)\nFixed by"}
{"_id":"q-en-kubernetes-2e125aa108aaf244bcfe7ca83c7081fad24054b312c79c5cbcd86436e2bd1a1f","text":"We should do this after 1.7. xref /sig testing cc\nIs this the task that needs completed - Add support to Ubuntu e2e test-infra\nThat is different, this is about running GPU tests as part of PR pre-submit tests. Right now we skip them because they are \"[Serial]\".\nI can work on this issue in conjunction with . Is this issue just a matter of adding GPU tests to a different suite or are there more subtasks?\nAnymore progress on this?\nThanks for the reminder. I just submitted an internal request for quota to run these tests. Once that is approved, I will add a PR in test-infra to run GPU tests on all k/k PRs. It will be non-blocking initially and then become blocking later on.\nAwesome, thanks!"}
{"_id":"q-en-kubernetes-0f154968bad7b3fe4febd8724083c6d4a91f4ea6f45457b43a4887b9d4012654","text":"xref: /sig node /area test\nI am working on this.\nCan you update this with the types of tests that you are adding? Is this similar to ?\nNo, this is for running the existing GPU e2e tests on GKE which are currently running on only GCE.\nThe referenced PRs appear to have merged, closing."}
{"_id":"q-en-kubernetes-07d743f275589dc899f72f61b7bb720b2610d6e1babb36a1a07c597b266bdaf8","text":"several operations to the glusterfs mount operation that nodes are not permitted to perform: looks up arbitrary secrets based on storage class parameters. this will not succeed, as the node authorizer will only allow nodes access to secrets referenced by pods or persistent volumes used by the node attempts to create a service and endpoint using the node's API client. nodes do not have permission to create arbitrary endpoints and services. For both of those reasons, that recreate operation will fail when attempted from the node. The recreate operation should probably move to one of the storage-related controllers (provisioner or attach/detach) cc\nMarking this with 1.7 milestone until we have a better idea of whether this is blocking or not for the 1.7 release.\nCC Is the endpoint originally created by the user or by the volume plugin creating/deleting Endpoint objects? In addition to the concerns raised, having a volume create new API objects seems like a bad idea (abstraction layer violation?).\nFrom on Slack: Unless there is justification otherwise, I will roll back the PR on Monday.\ncc/\nrevert button doesn't work, have to do it manually\nRevert PR is approved and awaiting merge (). Once that merges this issue can be closed.\nOpened an E2E pr to catch this kind of behavior in the future. WIP, still exploring a method to reproduce and catch the behavior."}
{"_id":"q-en-kubernetes-bc5dc487097ca8e1afb5b49dfc6335c039b52302e29bce4de3d1449afeaae0f6","text":"<!-- Thanks for filing an issue! Before hitting the button, please answer these questions.--Is this a request for help? (If yes, you should use our troubleshooting guide and community support channels, see ): No What keywords did you search in Kubernetes issues before filing this one? (If you have found any duplicates, you should instead reply there.): local-up- Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG REPORT <!-- If this is a BUG REPORT, please: Fill in as much of the template below as you can. If you leave out information, we can't help you as well. If this is a FEATURE REQUEST, please: Describe in detail the feature/behavior/change you'd like to see. In both cases, be ready for followup questions, and please respond in a timely manner. If we can't reproduce a bug or think a feature already exists, we might close your issue. If we're wrong, PLEASE feel free to reopen it and explain why. --Kubernetes version (use ): Environment: Cloud provider or hardware configuration: KVM OS (e.g. from /etc/os-release): fedora 25 Kernel (e.g. ): uname -a Linux k8sbuild-master 4.11.3-200.fc25.x8664 SMP Thu May 25 19:03:07 UTC 2017 x8664 x8664 x8664 GNU/Linux Install tools: Others: What happened: local-up- fails to create kube-dns and StorageClass. What you expected to happen: local-up- is successfully executed. How to reproduce it (as minimally and precisely as possible): 1) Clone kubernetes from master branch 2) run local-up- $ export ENABLEHOSTPATHPROVISIONER=\"true\"; ./hack/local-up- If you already have kubernetes , try this 1) make cleanmeta 2) run local-up- $ export ENABLEHOSTPATHPROVISIONER=\"true\"; ./hack/local-up- Anything else we need to know: I think this problem is timing issue because sleeping 3 seconds before startkubedns can avoid this problem.\nThere are no sig labels on this issue. Please by:<br(1) mentioning a sig: <br(2) specifying the label manually: <br<brNote: method (1) will trigger a notification to the team. You can find the team list and label list\n/sig testing /sig storage /sig network\n/sig apps\nI met the same issue we should add a function to check if all api-versions are up"}
{"_id":"q-en-kubernetes-efc4d6f641fd1b1bcaa275636a136a6176382b556456e87b10de85fb3c9c026d","text":"Every time watch is reset, event exporter exports an entry about it to Stackdriver. To make it less irritating, this has to be done once per container run Related image PR: /cc"}
{"_id":"q-en-kubernetes-30aaf5a2ce7eb6db27689d060ddcaf85253c0ae07b0503c963889141615172e9","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: If node is unreachable, NodeController only set , , to unknown; it's better to also set NetworkUnavailable & InodePressure to unknown.\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/sig node\nIsn't a known condition if node is unreachable?\nis the status of container, e.g. CNI is not ready. If node is unreachable, it should be Unknown.\nThanks for explanation!\nwill not set NetworkUnavailable & InodePressure to unknown, refer to ."}
{"_id":"q-en-kubernetes-1dc2549ac0119567f8fdbad5a90f42a3a48bc3f10a1848651b920c883c2c8bd8","text":"There's a problem b/c pods are created to run on all nodes, but run sequentially, effectively multiplying testing time"}
{"_id":"q-en-kubernetes-b79d32e7a321057d32fcaee893eb22ae1c074b6d42568816fe8e5690a6d0ee49","text":"/kind bug What happened: A pod with an empty namespace field was mutated by a pod preset in a different namespace. What you expected to happen: Pod presets should only operate on pods in the same namespace. How to reproduce it (as minimally and precisely as possible): Submit a pod with an empty namespace field and observe pod preset mutation. Anything else we need to know?: This is a fix for an alpha non-default feature. Fix submitted in\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/sig api-machinery"}
{"_id":"q-en-kubernetes-cc9a90a08712a5d452773e568c5f455ca56f11e72f82f403f33b02eda765f73c","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: What happened: When trying to list object using field spec.SchedulerName, we are getting the below error message. Failed to list : field label not supported: spec.schedulerName What you expected to happen: To be able to fetch pods using the list and watch based on the field 'spec.schedulerName' How to reproduce it (as minimally and precisely as possible): Create a simple controller to with a list and watch method. Set the FieldSelector ( in the ListOptions structure ) in the ListFunc and the WatchFunc to 'spec.SchedulerName=custome-SchedulerName' Running this controller yields the above error. Anything else we need to know?: Description: Unable to fetch pods based on the field 'spec.schedulerName'. This was in 1.6 version, for multi scheduler support in kubernetes. We need to be able to fetch pods 1) Assigned to a particular custom scheduler. 2) Should be able to fetch the state transition of the pods scheduled by a custom scheduler using the fields selector. Whats the correct way to use 'schedulerName' fields for fetching objects? Or is there any other way for the custom scheduler to determine the above information from the kubernetes? This is an urgent issue, any quick help/suggestion will be of great help. Environment: Kubernetes version : Client Version: version.Info{Major:\"1\", Minor:\"7\", GitVersion:\"v1.7.0\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2017-06-29T23:15:59Z\", GoVersion:\"go1.8.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"7\", GitVersion:\"v1.7.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2017-07-14T01:48:01Z\", GoVersion:\"go1.8.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Install tools: kubeadm\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/sig api-machinery\n/cc\ncc\nSIG API Machinery doesn't own the contents of the API--we only need to be involved if the mechanism needs to be general.\nGiven the error message mentioned in the bug report at the top of this thread, do you have a suggestion of what we forgot to do when we the SchedulerName field to the PodSpec? i.e. to enable field selection on the schedulerName field?\n/cc me"}
{"_id":"q-en-kubernetes-d84f225075f0f84856b63395873a83aaf5523b5813df96adeb3ab8300018cf9a","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug /sig storage What happened: Offshoot of . Let's change the tests to use the programmatic API instead of gcloud CLI.\nJust confirming that this issue is referring only to the 3 gcloud CLI calls in\nyup\nPR:"}
{"_id":"q-en-kubernetes-7ea6a0d142494ff2b4ab87c8d729fbcd3c1faf2bb0e5c24469e589e79da71312","text":"Seeing flakes on master: https://k8s- https://k8s- and on unrelated PRs: https://k8s-\nI was able to reproduce this locally (although it's very hard to trigger). Initial thoughts are that it's different from ; in this case, the actual sync is instant (millseconds) but the CRD created in the test fails to appear in discovery calls over the course of repeated attempts every 5 seconds for the remainder of the test. Normally, the type shows up in discovery almost immediately after creation.\nLooks like PR might be hitting a related flaky test. I get 6 failures in 60 runs from the command I'm super new around here so thought it prudent to add the information I have found before attempting to dig deeper. I followed instructions in to build from within directory If the test failure logs are of any use to anyone just message and I'll post one of them.\nThanks. I've also been able to reproduce locally and am debugging it today. I did have to hack up the integration test setup to get apiserver logs in this context. I've also opened to explore making the integration test script more configurable with regards to logging.\nI was wrong about the discovery aspect; the CRD is coming back from discovery, but the restMapper fails to map it. I also notice we're using DiscoveryInterface.ServerPreferredResources which is not actually implemented with a cache in memCacheClient.\nokay, now I'm getting somewhere. I made a which exercises discovery and REST mapping without the GC in the picture at all, and have been able to reproduce the flake 1 time out of 40 runs so far with .\nFirst problem is pretty clear in hindsight; we reset the restmapper, then get deletable resources using the discovery client directly, then sync based on the resources from discovery. The direct discovery call doesn't use the client's internal cache, and so the restmapper might be out of sync with the subsequent discovery call, causing us to sync with resources not yet represented in the restmapper's cache. That explains the isolated flake in the test I made, but doesn't yet explain why the restmapper would continue to fail on subsequent sync calls (since the nest restmapper reset would usually bring it up to date with the resources returned from the direct discovery call).\nI'm out tomorrow, so a quick update here in case somebody else looks into it... can reproduce the issue where DeferredDiscoveryRESTMapper sometimes becomes unable to map the CRD resource, even after the type is available from discovery. I've also ruled out the memCacheClient by replacing it with a non-caching implementation like: And using that in the setup like: The issue can still be reproduced that way. Looks like more debugging of DeferredDiscoveryRESTMapper itself is required.\nFinally tracked this down. Custom resource kinds are always given a plural suffix of \"s\". If the singular name ends in \"s\", the assumes the plural form is an \"es\" suffix, which will never match the generated kind. This explains the intermittent nature of the failure (the randomly generated names often happen to match the assumptions of the rest mapper).\nis the root cause\nGood effort"}
{"_id":"q-en-kubernetes-9101ad393fb6a2a33b8bfb776eef50d2cb2916deea96a521a370ed1fc49f445a","text":"In Cinder cloud provider openstack Inside of AttachDisk function, it issues DetachDisk if it find out the the volume is currently attached to a different instance. This is very dangerous because it is very likely the volume is still mounted to that instance. The correct behavior should be fail this AttachDisk call. K8s attachdetach controller should handle unmount and detach volume safely.\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\ncc"}
{"_id":"q-en-kubernetes-58d223bcda71c0cad54d5576eb06f7eca04bacec0a41ff1f7882b5e79cede5ec","text":"Making an API request for table output and specifying the parameter causes a server error in parameter conversion, and an inaccurate error message in response: server log: E0804 09:41:14. ] apiserver received an error that is not an metav1.Status: converting (url.Values).[]string to (v1alpha1.TableOptions).v1alpha1.IncludeObjectPolicy: couldn't copy '[]string' into 'v1alpha1.IncludeObjectPolicy'; didn't understand types\nUgh, typed strings."}
{"_id":"q-en-kubernetes-d99d066ab36522f6dcfea94e20b0ee3694b05e719a1247e54f2eed3487eedcdf","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind feature What happened: Unit test error message match the conventional Go style: For example: if got, want := set.Status.Replicas, 3; got != want { t.Errorf(\"set.Status.Replicas = %v; want %v\", got, want) }\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/sig testing\n/assign\n/reopen"}
{"_id":"q-en-kubernetes-2a0e15966cdb248fd0acdf76a89014fcb0b5e0e863719e0a7c39422e7cec2653","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: What happened: e2e test panic in test What you expected to happen: no panic How to reproduce it (as minimally and precisely as possible): start \" [Feature:ManualPerformance] should be able to handle 3 pods per node\" e2e test on a 3-node cluster Anything else we need to know?: panic info : ! After some digging find why. Because the cluster size is 3 that d is 0 in sleepUpTo func which causes rand.Int63n panic happy to fix this. Environment: Kubernetes version (use ):1.7.3 Cloud provider or hardware configuration*: bare-metal OS (e.g. from /etc/os-release): ubuntu Kernel (e.g. ): 16.04 Install tools: kubeadm Others:\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/sig Testing\njust add a sanity check is fine I think\nIt seems that the count of pods is 0.\npod count is not 0, is 9(33)\nIf totalPods less than throughput, in our case, totalPods is 9, throughput is 10, the creatingTime will be 0, which makes rand.Int63n panic\nor throughput is getting from env LOADTESTTHROUGHPUT, safe to add a sanity check\nIf totalPods less than throughput, in our case, totalPods is 9, throughput is 10, the creatingTime will be 0, which makes rand.Int63n panic agree."}
{"_id":"q-en-kubernetes-fc6177451e9ae3dd109f95c34ca28c9d9a0c4aae10f2acbc42a784aa55bc42f8","text":"seems to belong to sig node PTAL and fix ASAP\n/assign as it seems like having something to do with"}
{"_id":"q-en-kubernetes-ff32de2e880660eb15ff6af8978eca25943f80e4eedc7381a29e1be2e1575046","text":"We allow users to specify in the Service object. If the user later removes this from the Service Spec, the controller will fail to sync. For example: creates an LB for Service.Spec.LoadBalancerIP = \"1.1.1.1\" with a user-owned static IP address. updates the Service to remove the requested LoadBalancerIP tries to used by the existing forwarding rule (i.e., \"1.1.1.1\"). It encounters a conflict. Controller assumes it owns the IP and by name but fails. We can fix the controller to continue syncing and using the user-owned IP for the LB, or create a new address for the LB. The former seems less disruptive to me, but the downside is that if user will not be able to let other components use this IP (i.e., 1.1.1.1) unless they explicitly delete the Service. I think it's ok but would like more opinions. thoughts?\nIf a service is updated to remove a static address, then it seems like that would be a deliberate move on the part of the user. It seems like it should probably release the address and provision a dynamic one, even if this is disruptive. How do we handle the reverse? If a user has an already existing LoadBalancer service, and they add the after the fact? Would this release the dynamic address and recreate the forwarding rule with the reserved static address?\nIt seems like the creating a new address makes sense here... General comment: given the combinations (and the bugs we have been running into), I think this is a problem that all of the potential state transitions for the load balancer have not been fully spec'd out.\nCreating a new address seems nice but the implementation is a nightmare. The service controller doesn't give you a previous Service spec, so you don't know what's changed (controller might not know either if spec was modified while offline). Given a desired service spec without a how do you decide to ditch your current IP and grab a new one? If we were to call to evaluate the situation, it's highly likely users will accidentally cause LBs to change IPs during typical usage. Imagine a user has a dynamically allocated IP for a service, but they want to promote it for safe keeping. The controller would think it needs to surrender the IP to the user's reservation. The only correct process would be to set before promoting the address; however, that will surely be a missed point in docs and result in unhappy users. I lean towards the LB continuing use of the IP. Besides the easy option of deleting & re-creating the service, they could change the to another static address then demote the address. I'm not too familiar with the ideological boundaries of controllers and spec, so let me know if this crosses the line. What if the service controller set to the created IP on first sync if the IP wasn't pre-specified? The controller could rely on loadBalancerIP to always indicate which IP to use. If the field is empty (blanked by user), then it's a signal to get a new IP.\nBy spec, I mean we should write down the careful reasoning that you just wrote in a permanent place :-)\nwould it then make sense for to be immutable on an already created service?\n+1 +1 Was thinking the same. Unlike the many annotations we support, updates to this actual field could be validated and blocked at the apiserver... Changing IP is disruptive anyway. Could we simply ask the user to re-create the Service?\nOf course, this could be an API-breaking change, and other cloud providers may actually be supporting this properly.\nI just ran across this today, but in a more likely (IMO) case. 1) create the Service, type=LoadBalancer 2) let the IP be auto-allocated 3) manually promote that IP to static and give it a meaningful name 4) observe errors from service controller trying to allocate that IP Since my Service does not spec \"loadBalancerIP\", the question is, do we: a) use the IP, but leave it as static b) use the IP, but demote it back to ephemeral c) allocate and use a new IP C is completely hostile. B is subtly hostile. A seems correct. Can it be implemented? If we find that the IP exists but the name of it is not what we expect, we should treat it as user-provided and leave it alone. This is still nasty because it logs a failure in the GCP activities log for every loop, but I don't know how to avoid that without getting a list of IPs and scanning for the IP we want.\nFiled for the looping.\n+1 as long as we can rely on the \"naming\" of the address to decide whether the controller owns the address and can promote/demote it.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. Prevent issues from auto-closing with an comment. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale Still relevant\n/assign\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale Still relevant\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nany update on this?\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/lifecycle frozen\nIf this issue has been triaged, please comment . If you aren't able to handle this issue, consider unassigning yourself and/or adding the label. I am a bot run by vllry. ‍"}
{"_id":"q-en-kubernetes-b6c7dc7df0d7bcfd58ca95e0601a8e4b9b3c643a3067c0232de981b8b87c87ec","text":"When using IP aliases on GCE, use a secondary range rather than subnetwork to reserve cluster IPs.\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals"}
{"_id":"q-en-kubernetes-a79b6bc7e527e825eca09a117ef51fcb0df687226620a6bdf023846890bd6a8e","text":"Test Cases: [ ] Egress blocking based on cidr to a set of pods [ ] Egress blocking based on namespace to a set of pods [ ] Egress blocking based on a set of pods to another set of pods\nUpdate: I'm working to try to enable these new features in a few CNI plug-ins so the tests can run locally.\nCan we please move out of 1.8 milestone?\n/label status/approved-for-milestone\n/kind cleanup\n/remove-kind bug\nThose labels are not set on the issue: <detailsIn response to : /remove-kind bug Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/kind cleanup\n/status approved-for-milestone /kind cleanup /remove-kind bug\n[MILESTONENOTIFIER] Milestone Issue Current <details open<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Escalate to the issue owners and SIG owner; move out of milestone after several unsuccessful escalation attempts. : Adding tests, refactoring, fixing old bugs. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details\nCan we please move out of 1.9 milestone? :)\nthis will be partially fixed with . I think it is safe to keep in the milestone."}
{"_id":"q-en-kubernetes-3136844f9d3895d8ddd69e5024a99bf146fde86d1eed084a149e842e256ed078","text":"bumped the vendored docker library and introduced a regression where pods with images that don't contain an explicit \"\" repo path result in failures. Results in the following error: Pod gets shows that the image has been pulled, though not by that name Changing the image to in the pod spec corrects the issue. I think the new code that is changing the behavior is somewhere in here is a new function brought in by the bump. If I checkout the commit before the bump, the problem does not occur. related:\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/sig node\ntaking a look now\nWhat docker version are you using?\nDo you have any e2e tests that are failing? We run a lot of tests with image \"busybox\"\nDidn't think about it before, but I guess 1.12 is the supported version for 1.8 eh? Let me try it on 1.12.\nHm, weird. We also have cluster e2e test coverage for 1.13. Many test are using directly from this function\nI installed a completely fresh box with docker 1.12 and could still recreate. I have no explanation :confused: from kubelet built from upstream/release-1.8\nThis is one of our e2e test node log: You could find tons of . One thing I don't understand on your side is that, logs in our test are always: Why on your node it's ? Where does that image come from? What if you delete it?\ndoes this mean that upgrades will fail if the local image names need to change?\nSeems not. Based on the following code, it's fine to pull/inspect with new reference when we previously pulled with old reference. Could not reproduce this.\nI could not reproduce with the code above either.\nFor any image which has no tag, kubelet will add a default tag , and default repo to it. So will change to and send to dockerd to pull. But dockerd knows and are the same thing, so should return right instead of not found. (as demo) Can you make sure your dockerd is the a version compatible with k8s 1.8?\nJust ran into this. If you up the kubelet's log level to 4, you'll see a \"not found\" error from For example: If we look at the code in that helper function, we have this: Here is the result of inspecting the image on my system (only showing relevant fields): You'll notice that the RepoTag it has is for , not . I am guessing the fact that we're somewhere inserting \"library\" into the image's name is what's tripping this up.\nIf it wasn't clear, I believe dockershim is what's having the issue. Once the kubelet sends the protobuf message to the dockershim socket to pull the image, it goes from to to to and then into the helper function I listed above.\nIt appears to be this: to to to That's where the \"library\" is coming from.\nPrior to the PR that bumped docker/distribution, used to have , which does not do any normalization (i.e. including \"library\").\nOK, I see why when we bumped to the newer docker/distribution, we switch to using . It's because now requires that the input be in \"canonical\" format (in this case, has and , if applicable), which you get automatically by calling ...\n[MILESTONENOTIFIER] Milestone Labels Complete Issue label settings: sig/node: Issue will be escalated to these SIGs if needed. priority/critical-urgent: Never automatically move out of a release milestone; continually escalate to contributor and SIG through all available channels. kind/bug: Fixes a bug discovered during the current release. <detailsAdditional instructions available <a href=\"\"here</aThe commands available for adding these labels are documented <a href=\"\"here</a</details\n/assign\ncould you give steps to reproduce this? We can't reproduce this and many of the node/cluster e2e tests pull images using unqualified name such as \"buxybox\"\nWhat docker distribution are you using? I could not reproduce this myself. I don't understand how does the reference show up, we never pull image with it. Is this a problem specific to a docker distribution?\nrun the e2e tests on a centos/fedora/rhel image\nIs there any signal in test-grid for centos/fedora/rhel? Would be nice to catch issues like this sooner than later.\nrun the e2e tests on a centos/fedora/rhel image there is not, and definitely should be, given that k8s releases rpms as part of each release. will be on the short list for the release retrospective.\nI feel like this is a problem in docker distribution in centos/fedora/rhel. Docker should not return when we are using to pull image. Should we fix the corresponding docker distribution instead of blocking Kubernete release?\nSince that would mean releasing a new version would regress existing users, we generally say that we'll do our best to tolerate it instead of releasing something known broken. We've done that with various versions of Docker, systemd, and iptables for years, and I'm sure we'll do our best to do that. I do agree that this looks like a bug in that docker distribution that should be fixed. I also agree that this should have been in the test grid and that the distro owner is responsible for ensuring the project has that signal.\nThanks! I agree that we should not break existing users. I'm fine with as long as we keep in mind we made this change, and clean it up in the future. I think it would be helpful if we could have test coverage in test-grid on more distros, so that we could have caught similar problems earlier.\nAgree - I think anyone who causes a release slip is required to make sure they fix the problem so it can't happen in the future. And get better all around testing. On Sep 27, 2017, at 7:40 PM, Lantao Liu <: ! I agree that we should not break existing users. I'm fine with long as we keep in mind we made this change, and clean it up in the future. I think it would be helpful if we could have test coverage on more distros, so that we could have caught similar problems earlier. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub , or mute the thread\nI have opened a bug for this\nThis has broken the CRI as well. Kubernetes should not normalize image names on behalf of the CRI runtime. It's up to implementors to decide how to handle an image name and since there's not a standard around Kube and Image names, I'd prefer not to break the CRI.\nAre you saying it's currently broken, or the fix was ok? , Antonio Murdaca <:\nThe fix is ok but the original patch caused another bug\nWhich one?\ncaused a bug fixed by\nDoes this problem fixed?"}
{"_id":"q-en-kubernetes-e40ebb6767d1e38ffdcbda9625fe881658b7fe225ed48accb10a587281467df9","text":"(SEE UPDATE at ) Currently, if you're writing an initializer controller, you're most likely using it to initialize a and deploying the controller with . Therefore you're likely using the following trick to avoid the chicken and egg problem: This bypasses the initializers for the pods and lets your controller to be deployed even though your InitializerConfiguration is configured to capture all pods. Problem occurs when you apply an update to this Deployment with the same manifest containing empty pending list. You get error: I think this is a somewhat major UX issue. It requires users to deploy the initializer controller with the but later on when they need to update the controller with , they can't use the same manifest containing the empty list. /area api /area extensibility /sig api-machinery /kind bug /assign\n/cc\nWe silently ignore some metadata updates, which is one option. Potentially we shouldn't require empty initializers pending, but simply empty initializers (which means - init directly), and only ignore that. Are there any examples of use of SMP for similar? I'm not sure this meets the bar for 1.8 but if we can find an example of something consistent already in the API I'd be willing to review.\nI don't have any concrete examples, there's an example initializer at making use the trick, I use it myself, and I suspect many more initializers to come will use this, too. I'm not sure if this would work, but does setting provide a way to skip initialization (if possible at all)? If not, we might actually need a separate field to signal skipping initialization.\n/cc\nCurrently the initializer plugin to on UPDATE, but that happens after it updating already initialized objects. We can simply change their order.\nI'd like to amend the issue, it's more serious than I thought. (thanks for bringing it to my attention If you're deploying your Initializer with and if you have an initializerconfiguration for objects, it's impossible to deploy your initializer in the first place without deleting InitializerConfiguration first: is not inherited by the created on is not honored (this field is not propagated to the created Pod). I think (2) should be fixed somehow, otherwise there will be no way of writing an initializer for Pods and deploying the initializer itself using Deployment/ReplicaSet/RC without deleting the InitializerConfiguration first. Maybe what we need is an exclusion/exemption selector (or namespace(s)) as discussed in to allow deployment of Initializers themselves easier.\nFor 1, see I don't want to pursue that though, because i) we'll need to fix workload controllers one by one, and ii) we need to carefully consider permission issues. I think we should focus on adding the namespace selector."}
{"_id":"q-en-kubernetes-b4b5959a3aaa5f30e123a555df8aee9d29b1dec6e0fefb212f2839a190ac1f9c","text":"/kind bug What happened: When building kubernetes-1.7.0 packages for openSUSE, results differ for each build in and other man pages. coming from that uses and probably sets do not have any specific order, so it would need to be sorted on output. What you expected to happen: Every build should give identical results. See https://reproducible- for why this matters. How to reproduce it (as minimally and precisely as possible): See Anything else we need to know?: Environment: Kubernetes version: 1.7.0 OS (e.g. from /etc/os-release): openSUSE Tumbleweed\n/sig release\nsomehow this bug seems to have been forgotten - can you help getting the right people to have a look?\n- can you please confirm that PR works for you?\nI tested it and it looks good.\n/assign I enjoy reading my highlights to find the issue has already been addressed. :)\nSorry this wasn't addressed sooner. It does seem like it fell into a black hole, and pinging was the right way to resolve that. :)\n/area build-release\nrelated to /reopen\nReopening this issue. <detailsIn response to : related to Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-3c9ed989ba950e66c60864531514691d49ac4eeb91c62c40bd6ad56d9c2c8cd9","text":"/kind e2e-test-failure GCE serial and GCI GCE serial started failing last evening https://k8s- https://k8s-\n[MILESTONENOTIFIER] Milestone Labels Complete Issue label settings: sig/testing: Issue will be escalated to these SIGs if needed. priority/critical-urgent: Never automatically move out of a release milestone; continually escalate to contributor and SIG through all available channels. kind/bug: Fixes a bug discovered during the current release. <detailsAdditional instructions available <a href=\"\"here</aThe commands available for adding these labels are documented <a href=\"\"here</a</details\nhttps://k8s- is still green, I would suspect something ~last Friday afternoon gets merged\nHere is what I found, after a while, the node lost their authorization to connect to API server. I believe this has the same root cause as\nI am going to revert offending pr () to see if it helps. EDIT: Looks like is in the process of reverting it:"}
{"_id":"q-en-kubernetes-785adbfc530cee5c14decf2077a251c037f7652cdfd80a2690c7ba11c85d0253","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: BUG /kind bug What happened: Attempting to run e2e test on GCE without cluster started fails with stack trace What you expected to happen: A useful error message to be printed How to reproduce it (as minimally and precisely as possible): Without a cluster started: Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration**: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\n/sig testing /assign\nThis happens because gcloud returns an error when a resource isn't found and ginkgo- sets errexit"}
{"_id":"q-en-kubernetes-0830e634f1865e538fd2ea94dff87a50d196b25f52c6a6ca814f2ee175c731aa","text":"It is possible to have multiple signals that point to the same type of resource, e.g., both SignalNodeFsAvailable and SignalAllocatableNodeFsAvailable refer to the same resource NodeFs. Change the map from map[v1.ResourceName]evictionapi.Signal to map[v1.ResourceName][]evictionapi.Signal\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/kind cleanup /priority important/soon taking a guess at labels to satisfy the milestone munger\n[MILESTONENOTIFIER] Milestone Labels Complete Issue label settings: sig/node sig/storage: Issue will be escalated to these SIGs if needed. priority/important-soon: Escalate to the issue owners and SIG owner; move out of milestone after several unsuccessful escalation attempts. kind/cleanup: Adding tests, refactoring, fixing old bugs. <detailsAdditional instructions available <a href=\"\"here</aThe commands available for adding these labels are documented <a href=\"\"here</a</details\nLooks like the PR for this is just awaiting merge"}
{"_id":"q-en-kubernetes-fc43a974ea2ea04c522d44e3808a2250bd460e5bc91555a60b00d399968fd9cf","text":"Currently, always returns with 0 exit code no matter the test fails or not. We should fix it to return non-zero exit code for test failure."}
{"_id":"q-en-kubernetes-9d96fadb6189b5bf3a00e3a1630f8f9a54b4f5aca0ead084364dd1f5a8e0cb0d","text":"When pod is evicted, pod is terminated. After this change , attach/detach controller should be able to detach the pod correctly so that new pod which uses the same volume will start up. We need to have an e2e test for this scenario.\nwhat environment is good for this e2e test? GCE or GKE or AWS? I see we have which has func, Can you comment on why is insufficient to cover this issue?\ndoes the pod need to be evicted? Or would be ok to delete the pod and then check that the PD is detached. Then create a new pod using the same PD and test that it attaches?\n[MILESTONENOTIFIER] Milestone Removed Important: This issue was missing labels required for the v1.9 milestone for more than 3 days: kind: Must specify exactly one of , or . <details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details\nping ^^\nsorry for the delay. Here we want to test the detach behavior of evicted pods. So delete a pod directly is different from it. When Pod is evicted, the containers will be killed and the pod status is terminated. But the pod api object is not deleted from API server. Before PR , such evicted pods will not trigger detach operation because the pod api object is not deleted from api server. So the purpose of the test is to make sure detach operation is triggered for such evicted pods.\nThanks Do you have a suggestion for how to force a pod to be evicted?\nYou could have the pod do some memory intensive application\nah, like my use of in the io-streaming pr :-)\n- Here's another option to cause eviction: ."}
{"_id":"q-en-kubernetes-878b3413520069a3d3965eb764a7edb4e0bab274ced2a3e5ddfa2f3f07c0e15b","text":"/kind bug What happened: When a cluster is provisioned using a containerised kubelet using the following images: Pods with PVCs are not created because kubelet cannot mount the disks. If you look at events, you can see that it's missing some binaries: This executable is . This problem can be resolved by mounting paths from the host into the rkt container, specifically and . Is this expected? mentioned in , that mounting operations are moving out from the main process into containers. Is this part of that effort? This caught me quite unexpected, and I didn't find any docs or changelog entries which indicated additional host mounts were necessary. If it's deliberate, I can close this until we perhaps document the breaking change. Is there a good place where we can document this? If it's not, I'll leave this open as a tracking issue. What you expected to happen: The hyperkube image to work without lots of host mounts. How to reproduce it (as minimally and precisely as possible): Run kubelet with rkt in one of the above listed images (see CoreOS' ). Environment: Kubernetes version (use ): Cloud provider or hardware configuration**: OpenStack OS (e.g. from /etc/os-release):Kernel (e.g. ):Install tools: tectonic-installer\n/sig storage /sig node\nSeems the base hyperkube image was changed: done in /cc\nThe pkg/util/mount/mount_linux source defaults to ext4 if a fstype is not selected. It makes sense to restore the e2fsprogs package within the hyperkube image.\ncc cc cc This seems like something that should be in the 1.8 milestone. Can someone who owns the hyperkube image comment on that?\n[MILESTONENOTIFIER] Milestone Labels Complete Issue label settings: sig/node sig/storage: Issue will be escalated to these SIGs if needed. priority/important-soon: Escalate to the issue owners and SIG owner; move out of milestone after several unsuccessful escalation attempts. kind/bug: Fixes a bug discovered during the current release. <detailsAdditional instructions available <a href=\"\"here</aThe commands available for adding these labels are documented <a href=\"\"here</a</details\nContainerization of mount is not going to happen until 1.9 at the earliest. So if hyperkube is broken, it should be fixed before ship.\n\"modprobe\" also went missing in 1.8 hyperkube images... Both kubelet and kube-proxy need it.\nAre you noticing any buggy behaviour on those components without it? Would you mind creating an issue and/or PR to restore that dep\nDone: And done: :)"}
{"_id":"q-en-kubernetes-141ad298529fc52bdd1d978ad80ddba1c5d4febe5b6ef954b2a124b684d72c89","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: What happened: my yaml: exec kubectl set resources command,just update first object: What you expected to happen: How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration**: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\n/sig cli\n/kind bug\n/assign"}
{"_id":"q-en-kubernetes-e4af3c33d68e650a678730ef7b2281c281dc93d7d3e260c24d3b6a0d9cd86129","text":"We should introduce a buffer (similar to what we have in the webhook output backend) to the log file output backend. Audit events should be queued on a relatively large buffer chanel, and an output go routine should grab events off the queue and write them to disk. If an audit event is output when the channel is full, it should be dropped (this should be counted by a prometheus metric, and an e2e test should ensure that the count is below some threshold). This behavior should be optional (configurable) /cc is tracking the regression\nNaturally os will cache , these cached data will be written to disk either by fsync or pdflush periodically. What do you think is the most time-costing procedure in audit?\nThe problem is not with the time consumed. The problem is that currently write is performed synchronously with processing the request, which slows the request down, no matter how well caching works\nThe problem is that currently write is performed synchronously with processing the request Sure synchronously write does cost time serializing. Introduce buffer mechanism can really reduce api request latency time.\nIt doesn't matter that much if it happens asynchronously Indeed\nYes, this is a good point to improve performance.\nNow I totally comprehend this. I am pleased to implement this buffer.\nI'll be happy to review that\naudit log is important sensitive information which we should try to make sure it reaches disk at all cost. Does this effort compromises that in favor of optimizing for latency ?\nI'll make sure it's configurable and if the security is critical, you can setup synchronous write in your deployment. Having said that, it's critical to be able to accommodate the case where having 100% calls written does not compensate for increased latencies. Does it make sense?\nthanks my preference would be to keep the default as no buffering when making it configurable. I didnt understand how does one setup \"synchronous write in your deployment\" and how that will affect audit logs reaching disk. My worry is mainly around a crash where the buffered audit logs dont reach the disk and are lost\nif use buffer, audit events lost is inevitable when crash. As like webhook backend, still exist this risk. But to improve apiserver qps, something should comprise. By the way, we still reserve the synchronous write mode, which is configurable.\nTo make it simple, when offered trade-off, people are more likely to give up audit logs reliability than performance. Makes sense?\ni am not against this change. i think what would be best is to make the default as no buffering and enable buffering through a parameter. Also for Enterprises, the trade off is not correct, they would prefer to NOT give up audit logs\nI understand what you're suggesting and I don't agree :) Switching to buffering will for sure increase the latency, but will only increase the probability of losing logs, which will never be zero. I'm pretty sure we need to have buffered mode by default. However, since this argument is about making a trade-off and since neither of us represents quorum, let's ask more people for their opinion What do you think?\nThe default is slightly less important here since audit logging as a whole is not enabled by default. If we do turn it on by default, it becomes more important. That said, my hunch is that we should default to the mode with lower performance impact, but I don't have strong feelings either way.\nAs long as there are options to turn it on of off (the synchronous write) I'm fine with either default. Still, I'd like to see due diligence towards ensuring low lost rate when using buffered write, and sufficient documentation around caveats of using one or the other. Audit, from my personal experience, is crucial for majority of enterprises and they won't tolerate not even single lost event.\nThat is not possible, there will always be a situation when an audit event can be lost: sudden power loss, disk/network failure, random bit flip, etc. The questions is whether 10^(-6) is low enough loss probability, or we need to get 10^(-14) by sacrificing performance (arbitrary numbers, just to illustrate my point). Note, that reduced performance can increase the number of timeouts and retries, increasing load and leading to failures and even more lost events.\nI think this is an important bug that we're not testing audit logging at scale\nMarking it as urgent, since it's blocking\nIf split into 3 prs, I must do it one by one . Now first one is up .\nSGTM Status of this issue: ETA: 1/12/2017 Risks: Complicated fix required\nhas been reviewed for almost 2 months and has not done. If I raise 3 prs to replace it, I am still worried whether all of them can be merged. They act as a whole.\nNow I'm going to devote my time to it, we'll do it\nOK.\nIn progress, reviewed first PR, waiting for applying changes ETA: Dec 01\nPer Exception discussion, this portion of the feature is being removed from the milestone. As such, changing the priority. /priority important-soon /remove-priority critical-urgent\n[MILESTONENOTIFIER] Milestone Removed From Issue Important: Code freeze is in effect and only issues with may remain in the v1.9 milestone. <details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details"}
{"_id":"q-en-kubernetes-284fb555a913f6894b005ac842118dc412cfd892c771454a17e2e489b164471e","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: Creating a service with created a load balancer without a floating IP. What you expected to happen: Load balancer should have a floating IP assigned by default. How to reproduce it (as minimally and precisely as possible): Deploy this manifest on a cluster and check the resulting load balancer. No floating IP will be assigned. Anything else we need to know?: The default behavior changed from k8s 1.7.x. It appears that an annotation that will add a floating IP, but the default should not have changed without warning. This also breaks the workflow. For example: Environment: Kubernetes version (use ): v1.8.0-rc.1+coreos.0 Cloud provider or hardware configuration**: openstack OS (e.g. from /etc/os-release): Container Linux by CoreOS 1520.4.0 (Ladybug) Kernel (e.g. ): 4.13.3-coreos Install tools: Others:\nThere are no sig labels on this issue. Please by: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. You can find the group list and label list . The in the method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals\n/sig openstack\nReiterating the mentions to trigger a notification: <detailsIn response to : Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nOops, default should be external. /assign"}
{"_id":"q-en-kubernetes-13002b8f7a203d65a315278f67f4fc6e847d33f04aed6426007d63ed6ca2eded","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: What you expected to happen: Test should not flake /sig storage /cc\n/assign\nGitHub didn't allow me to assign the following users: vkamra. Note that only can be assigned. <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\nGitHub didn't allow me to assign the following users: vkamra. Note that only can be assigned. <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n- You won't be able to assign this to me (not a kubernetes member). Am taking a look.\nvolumemetrics.go_ could/should have more logging, either calling or more calls to .\nThe test appears to fail on this metric: \"kubeletvolumestatsused_bytes\". So samplePVC != pvcName and/or sampleNS != namespace. It would be nice if the sample name and ns were also logged. Also something is weird here: the framework's is called before the local . Note in the log:\n- will add a couple of Logf() statements to log the samples - you're right that will help. Once the volume is attached, volume stats are grabbed every interval. The thing I noticed in the logs is that we issue a delete for the pod at but the volume only got attached/bound successfully at The test assumes blocks till the pod is running and PVC is attached but maybe that's not a valid assumption?\nNevermind - That log message has 2 timestamps in it - the volume was provisioned at .\nPushed The working theory being that volume stats were not collected within the wait earlier being used.\ndid you figure out if I was correct in thinking that the framework was invoked before the local ? That seems very, very unlikely but that is how it looked in the log.\n- I believe you're right that the framework is being invoked before the local one. I observe the same in a test as well, so I don't think it's specific to this call. Based on the , nested blocks should be invoked inner-outer. It's not clear, however - whether the local is nested - or at the same level as the framework block when I look at"}
{"_id":"q-en-kubernetes-565c435e3a6e8af170a56a1cf4d531f90a9eb53fea1c780715e76a7ee1908f76","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: On a fresh Ubuntu 16.04.3 system booted from the official cloud image, fails because exists. What you expected to happen: kubeadm successfully initializes the cluster How to reproduce it (as minimally and precisely as possible): a new VM from the latest the Anything else we need to know?: Contents of : Environment: Kubernetes version (use ): Cloud provider or hardware configuration: Hyper-V generation 1 virtual machine OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: none Others: none\nThis is related to the location where the kubelet persists its certificates while running in the background, waiting for config: the kubeadm 1.8.0-0 package accidentally used a location that is the kubeadm 1.8.0-1 package corrected that to use , but starting the kubelet before running causes files to be generated into the folder since expects there to be a running kubelet prior to being called, it shouldn't expect the kubelet's folder to be empty if you are scripting bootstrapping a known clean machine, there are a few possible workarounds until is released in 1.8.1 (any of the following work around this issue): verify this is the only preflight check failure, then run the init or join command with stop the kubelet service and remove prior to running the init or join command run prior to running init or join addressed as part of\ncc\nI did it worked for me!\nOn a known clean environment, that's an ok workaround for the moment, but the reason checks for existing files/folders is to avoid stomping an existing installation, and running would negate those checks.\n(If you wonder why the deployment scripts that you wrote yesterday stopped working this morning – this is why! :-)) For the time being, before running or , I'll: check if or exists if they don't exist, stop kubelet, and wipe out I'd love to know if there are foreseeable pitfalls in this approach. Thanks! <3\napologies for the churn. we traded a for this false-positive fail safe issue. working with to determine the best way to resolve this.\nNo worries, and thanks for the super-quick turnaround time. Much appreciated! (I didn't mean to sound snarky or anything; sorry if that how it came out!)\nThis guys are posting new versions without properly testing it. Can you imagine brand new centos 7 installation and then getting: \"kubeadm 1.8.0 init fails with \"/var/lib/kubelet is not empty\" how disgusting that was and what I was thinking about responsibility and professionalism here?\nEverything was done by: but now it is even getting better: The HTTP call equal to 'curl -sSL http://localhost:/healthz' failed with error: Get http://localhost:/healthz: dial tcp [::1]:: getsockopt: connection refused you should remove 1.8 and put it do \"back to the drawing board\" stadium. It is useless even to test this 1.8\nI apologize for the issue. Our top priority was to prevent data loss and issues upgrading existing installations that would disrupt already-running workloads. We are working now to resolve this issue with fresh installations.\nWish you luck.\nETA for a fix for this? should i hard code this deletion of contents of /var/lib/kubelet prior to running kubeadm init in my automation scripts? or simply add the ignore pre flight checks arguement will do the trick?\nthere are plenty of other pre-flight checks that are valuable and you don't want to skip, it's just the check for an empty that is incorrect. if you are scripting bootstrapping a known clean machine, there are a couple possible workarounds until is released in 1.8.1: run / skipping preflight checks stop the kubelet service, remove , then run /\nok just to confirm, moving forward, ideally should not be checking for contents in (i.e. having contents is totally fine), or should installing kubelet not populate contenst of ?\nCorrect. The instructions start the kubelet and let it run in a crash loop in the background, waiting for config. In that state, the kubelet is free to write to its directory containing state (), so should not require that directory to be empty in order to run\n(Reposting my earlier tip since it worked like a charm for my automated deployment scripts.) For the time being, before running or , I'll: check if or exists if they don't exist, stop kubelet (), and wipe out () This might or might not be appropriate for your use-cases, but for mine it works great (I'm automatically deploying hundreds of k8s clusters of training purposes). Stopping kubelet is necessary to avoid race conditions where it would recreate the directory before you run .\nIf i wipe out the /var/lib/kubelet/pki directory and do not restart kubelet, the kubeadm init process hangs with kubelet-check] The HTTP call equal to 'curl -sSL http://localhost:/healthz/syncloop' failed with error: Get http://localhost:/healthz/syncloop: dial tcp [::1]:: getsockopt: connection refused. [kubelet-check] It seems like the kubelet isn't running or healthy. [kubelet-check] The HTTP call equal to 'curl -sSL http://localhost:/healthz' failed with error: Get http://localhost:/healthz: dial tcp [::1]:: getsockopt: connection refused. [kubelet-check] It seems like the kubelet isn't running or healthy. [kubelet-check] The HTTP call equal to 'curl -sSL http://localhost:/healthz/syncloop' failed with error: Get http://localhost:/healthz/syncloop: dial tcp [::1]:: getsockopt: connection refused. [kubelet-check] It seems like the kubelet isn't running or healthy. [kubelet-check] The HTTP call equal to 'curl -sSL http://localhost:/healthz' failed with error: Get http://localhost:/healthz: dial tcp [::1]:: getsockopt: connection refused. Unfortunately, an error has occurred: timed out waiting for the condition This error is likely caused by that: The kubelet is not running The kubelet is unhealthy due to a misconfiguration of the node in some way (required cgroups disabled) There is no internet connection; so the kubelet can't pull the following control plane images: You can troubleshoot this for example with the following commands if you're on a systemd-powered system: 'systemctl status kubelet' 'journalctl -xeu kubelet' When i check the status of kubelet, it's restarting and never starts because So, i am guessing my only option is to skip preflight checks, let me try that out..\nWish you luck. This is not the only problem: kubectl cluster-info Kubernetes master is running at http://localhost:8080 [root ~]# kubectl apply -f W1004 06:49:56. 2422 ] Failed to download OpenAPI (Get http://localhost:8080/swagger- dial tcp [::1]:8080: getsockopt: connection refused), falling back to swagger The connection to the server localhost:8080 was refused - did you specify the right host or port? Why did you put this 1.8 version when it is obviously not properly tested. Or tested at all?\nNot to forget: systemctl status firewalld firewalld.service - firewalld - dynamic firewall daemon Loaded: loaded (/usr/lib/systemd/system/firewalld.service; disabled; vendor preset: enabled) Active: inactive (dead) Docs: man:firewalld(1)\nkubeadm does not run kubernetes unsecured. following the setup instructions in the output of configures to speak to the initialized server:\nOk. This worked. Thank you. Node also by running kubeadm join.... Now the problem is dashboard. I have followed: but still don't understand where to start kubectl proxy and max.I god when I try to access dashboard is (https://192.168.4.10:6443/): { \"kind\": \"Status\", \"apiVersion\": \"v1\", \"metadata\": { }, \"status\": \"Failure\", \"message\": \"forbidden: User \"system:anonymous\" cannot get path \"/\"\", \"reason\": \"Forbidden\", \"details\": { }, \"code\": 403 } I am total newbee and anything which is not clear into documentation I can't resolve. I am using this for testing purposes and stil after 10days can't get dashboard nice and clean and then explore full power of kubernetes. Right now it is unusable even for testing.\nHi All, Need help here current i'm installing Kubernetes version: v1.8.0 using the command \"kubeadm reset && kubeadm init\" it went hanged state. systemctl status kubelet log showing that Oct 04 20:15:02 cluster-1 kubelet[]: W1004 20:15:02. ] Unable to update cni config: No networks found in /etc/cni/net.d Oct 04 20:15:02 cluster-1 kubelet[]: E1004 20:15:02. ] Container runtime network not ready: NetworkReady=false reason:Net...nitialized Oct 04 20:15:02 cluster-1 kubelet[]: E1004 20:15:02. ] Failed to get system container stats for \"\" Oct 04 20:15:02 cluster-1 kubelet[]: E1004 20:15:02. ] Failed to get system container stats for \"\" Oct 04 20:15:07 cluster-1 kubelet[]: W1004 20:15:07. ] Unable to update cni config: No networks found in /etc/cni/net.d Oct 04 20:15:07 cluster-1 kubelet[]: E1004 20:15:07. ] Container runtime network not ready: NetworkReady=false reason:Net...nitialized Oct 04 20:15:12 cluster-1 kubelet[]: W1004 20:15:12. ] Unable to update cni config: No networks found in /etc/cni/net.d Oct 04 20:15:12 cluster-1 kubelet[]: E1004 20:15:12. ] Container runtime network not ready: NetworkReady=false reason:Net...nitialized Oct 04 20:15:12 cluster-1 kubelet[]: E1004 20:15:12. ] Failed to get system container stats for \"\" Oct 04 20:15:12 cluster-1 kubelet[]: E1004 20:15:12. ] Failed to get system container stats for \"\" Warning: kubelet.service changed on disk. Run 'systemctl daemon-reload' to reload units. After that i'm tried to weave it failed with error kubectl apply -f \"$(kubectl version tr -d 'n')\" The connection to the server localhost:8080 was refused - did you specify the right host or port? W1004 18:53:33. ] Failed to download OpenAPI (Get http://localhost:8080/swagger- dial tcp [::1]:8080: getsockopt: connection refused), falling back to swagger The connection to the server localhost:8080 was refused - did you specify the right host or port? Please anyone can help me on this\nbe sure to follow the instructions to add the admin kubeconfig file so kubectl can communicate with the initialized API server:\nFor anyone following this thread, if you encounter issues other than the message, please report them in a separate issue (after searching for existing reports), to ensure they are triaged and routed appropriately and resolved as quickly as possible. Thanks.\nthank you for your quick response now i'm getting kubectl apply -f \"$(kubectl version tr -d 'n')\" Unable to connect to the server: Forbidden W1004 20:34:18. ] Failed to download OpenAPI (Get https://192.168.1.3:6443/swagger- Forbidden), falling back to swagger Unable to connect to the server: Forbidden\nyou seem to be hitting a totally different problem than the one related to ; can you please open a different issue? Thank you so much!\nI also the same issue'var/lib/kubelet is not empty' after I used the command \"kubeadm reset && kubeadm init\" it want to hanged state.(if want u can see my above threads).to move to another step I'm trying to install the weave it was failed saying that kubectl apply -f \"$(kubectl version tr -d 'n')\" The connection to the server localhost:8080 was refused - did you specify the right host or port? W1004 18:53:33. ] Failed to download OpenAPI (Get http://localhost:8080/swagger- dial tcp [::1]:8080: getsockopt: connection refused), falling back to swagger The connection to the server localhost:8080 was refused - did you specify the right host or port? Can you please help me to solve the issue Thank in advance\npick to 1.8 branch is reopening to track issue until resolved in 1.8.x\nif you get it means that your kube config file is missing. By default, tries to connect to . But when you deploy a cluster with , the API will be listening on port 6443. So you should see an error relative to port 6443, nor port 8080. If you see an error related to port 8080, it means that you haven't copied the configuration file () generated by . I hope that makes sense. My piece of advice would be to destroy the machines that you are currently using and restart from scratch (assuming you are using VMs), making sure that you follow each step carefully. Very often it solved the problem for me, because I had forgotten a step, or deviated from the instructions. Especially if something fails, it is often easier to restart from scratch, until you understand fully the technology and each underlying command. Good luck!\nHello hemaprasad, The problem is that when people know it is hard for them to set their mind to explain things to people who are new in a same area. Explanations given by jpetazzo and liggitt are ok but they are stepping over some steps which they assume but we as newbee don't. Now, to have kubernetes operational we all should: -stop the kubelet service, remove /var/lib/kubelet/pki, then run kubeadm init (without starting kubelet service first) Same goes for nodes: -stop the kubelet service, remove /var/lib/kubelet/pki, then run kubeadm join (with a token saved from kubeadm init) Now... how to make dashboard accessible from my workstation which is out of kubernetes cluster? Suppose for this I should go to some other thread?\nyour explanations are correct! As for your dashboard question, I have no idea, but I would recommend to open another github issue for that (since the title of this issue is \"kubeadm 1.8.0 init fails with \"/var/lib/kubelet is not empty\"\"). Thank you!\nThank you for fast reply. Also I am trying these innocent commands from documentations: scp root . ./kubectl --kubeconfig get nodes but I am getting: Unable to connect to the server: x509: certificate signed by unknown authority (possibly because of \"crypto/rsa: verification error\" while trying to verify candidate authority certificate \"kubernetes\") My guess is that is also the same reason why dashboard over proxy is not working. Some help?\nIt looks like you are copying from a remote host. If it is the case, you also need to edit to point to that remote host (instead of ). The tutorials and documentations imply that people will run directly from the master.\nNice this is... so why into same documentation exist this: Controlling your cluster from machines other than the master In order to get a kubectl on some other computer (e.g. laptop) to talk to your cluster, you need to copy the administrator kubeconfig file from your master to your workstation like this... Never mind, it is obviously long way to go till this function even in test environment :(\nHello all. Finally dashboard starts working from master node. Please, do NOT use 1GB-2GB on master node, it would behave erratically and services which you expect to start normally... won't. Use at least 4GB and then also be patient. http://localhost:8001/ui/ will respond with strange formed URL so please use copy-paste: http://localhost:8001/api/v1/namespaces/kube-system/services/https:kubernetes-dashboard:/proxy/ That's for today. Hope this will help somebody.\n[MILESTONENOTIFIER] Milestone Issue Needs Approval Action required: This issue must have the label applied by a SIG maintainer. <details<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Never automatically move out of a release milestone; continually escalate to contributor and SIG through all available channels. : Fixes a bug discovered during the current release. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details\nI've tried you're workaround But I'm then on kubeadm init I'm hitting this:\nthis is fixed in v1.8.1\nAfter installing 1.8.1 kubelet won't start:\nNot only what you mentioned dkirrane, there is a problem with: They never wrote ANY detail how to add a new Node in a Cluster when your initial Token (which is valid for ... unknown) expired! Not a word. Only if you want to test-and-try yourself. Lovely beta troubleshooting job this is. Documentation should be much, much better. Especially for somebody not working on Kubernetes dev. team.\nIf you try with this you will see that there is no way to use IP address mentioned: (Text is from: ) \"To implement this automation, you must know the IP address that the master will have after it is started..... kubeadm can generate a token for you: kubeadm token generate Start both the master node and the worker nodes concurrently with this token.\" Where to use mentioned IP address? How to make any useful Node cluster joining after Token has expired?\nHello, solved a Token problem. Documentation really needs to be updated considering newbee mind. Otherwise it is source of frustration. Currently nodes are ok and operational. If this above helps you... ok.\nThis issue is fixed in the latest version of kubernetes. Just reset the kubeadm and remove the /var/lib/etcd as kubeadm is expecting it to be empty. Then start the kubeadm init. It should work fine."}
{"_id":"q-en-kubernetes-dcd38f96bb0b0d9330540253d08f14b2aaabd10f649fad7a7075c60a5f24ba31","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: CRD: Existing object: New object: says: If this is a problem with the new object definition rather than a bug with , I'd expect a better error message. (The new object looks fine to me, though; it should be identical to the old object.) What you expected to happen: I expected the object to be updated without an error. How to reproduce it (as minimally and precisely as possible): See description. Anything else we need to know?: Environment: Kubernetes version (use ):Cloud provider or hardware configuration**: GCE OS (e.g. from /etc/os-release): CentOS Linux 7.4 (1708) Kernel (e.g. ): Linux master-head-1 4.13.4- SMP Wed Sep 27 13:32:23 EDT 2017 x8664 x8664 x8664 GNU/Linux Install tools: Custom scripts Others:\n/sig api-machinery\nAre you able to reproduce this everytime? I am able to use apply normally with your example.\nI can reliably reproduce it if the annotation exists: If I use to remove the annotation, then it works the first time but fails thereafter:\nThis is happening for other CRDs as well (Prometheus Operator ServiceMonitors). If I apply a ServiceMonitor CRD 3 times: If I remove the annotation or replace the object () then I can work around the issue.\nSeems any CRD is affected; the CRD used by our are also not working: This has completely broken our CI/CD workflow... unfortunately we'd already completed the TPR-CRD migration for the 1.8 upgrade before noticing this problem, so we can't really downgrade. How safe is as an alternative to ? It still generates an update event for watchers, not a delete/recreate, right?\nIt looks like something going wrong during the patching. After using changes from , I get this: /cc /area third-party-resource\nBut uses json merge patch and not strategic merge patch for custom resources:\nI was able to reproduce this in v1.8.0 with the original manifests at the top. As found, it only happens on the 3rd execution: It also seems to only happen when the ought to be a no-op. If I change something in the IPPool spec, works once more, then starts failing again. When the is a no-op, the PATCH request body generated by looks like this: When the is NOT a no-op, the PATCH request body looks like this:\nThere's a typo in the code generating the error messages for both and : They should be calling instead of . When I fix that, the error message becomes:\nI can reproduce this without , just by doing a no-op twice:\nThe same issue affects v1.7.8 (tested on GKE)\nSame issue with v1.8.1\nThis just fix the error message: But it doesn't fix the issue\nStill investigating the issue but I noticed that when we do a no-op patch for native resources, we get a \"not patched\" response. However, for non-native resources, it still tries to patch it (and bumps the resourceVersion too).\nI don't konwn whether this is a issue, I can't do strategic merge patch to CRD. When I run , --type=merge must be passed.\nWe should be able to apply strategic merge patch to CRDs. However, we can't apply strategic merge to CRs because we're missing the metadata about how to handle each field for the strategic merge patch. Unfortunately, this error is occuring even with . :(\n/assign\ndo you know if the fix fir this will make it to 1.8.3\nhopefully yes, this should be merged soon.\nthanks! this has been a blocker for us to upgrade to 1.8\ncan you confirm this is completely fixed? The PR writes \"partial fix\".\nNow we can't reproduce this issue with the method descripted above. I still need some I extra time to research it. Maybe we can close this one. If I can reproduce the issue in another way, I can open another issue with new situation.\nIf it's not fixed, let's keep it open.\nFun, a fork closing issues.\nFrom Then I think we can officially close this.\n(the fix is merged into release-1.8, but after v1.8.4-beta.0)\nwhat stable release of k8s will this be in? 1.8.4 or 1.8.5?\nIs there an update for this?\nlooks like it was merged here: but did not make it into 1.8.4, might make it into 1.8.5"}
{"_id":"q-en-kubernetes-34d17b13918e1176128a125a4446a9ee51e6076dbf365b28187a9878bb1d98f0","text":"This issue is meant to track planned work items for pre-allocated hugepages across releases. This issue captures the plan of record discussed in the wg-resource-mgmt meetings. [x] HugePages should be restricted via ResourceQuota [Owner: , fixed ] Assess ability to graduate feature to beta. [x] HugePages should require the user to request cpu or memory [Owner: fixed ] [x] HugePages should not be tied to QoS [Owner: fixed ] [ ] HugePages promoted to beta [Owner: ][ ] HugePages isolated at container and not just pod boundary [Owner: TBD] [ ] HugePages should be restricted via LimitRange [Owner: TBD] cc\n/cc\nIf no one is doing it, I'd like to work on the ResourceQuota part.\nwould this be discussed in the meeting recently? i would like to try on the \"via LimitRange\" but seems more details need to be settled.\n/cc\nRegarding to the ResourceQuota issue, all I could think is to add support for resource HugePages in podEvaluator of quota, so that it could be aware of the hugepage declared in resourceQuota, and that requested in podSpec. It is thought to be a small change, The only thing that needs attention is the way to address resource HugePages, as it's name is not fixed, like CPU or memory, but in a specific format. Correct me if I miss something :)\n-- the quota change is relatively small, if you want to pick up the work, i would be happy to shephard it.\n-- you as owner, thanks for volunteering! let me know when you have the pr up\nare you ready for the LimitRange part? :)\nyes, I've worked on this for some time and will commit the pr ASAP.\n/cc\nhi, I am confused by the tests I've made, which seems the \"HugePages should be restricted via LimitRange\" has already been done. I created the LimitRange as below: apiVersion: v1 kind: LimitRange metadata: name: mytest spec: limits: default: hugepages-2Mi: 40Mi defaultRequest: hugepages-2Mi: 40Mi type: Container and the pods created by the following yaml: apiVersion: v1 kind: Pod metadata: generateName: hugepages-volume- spec: containers: image: tomcatjava:v6.0 command: name: example and the output of the pod in yaml is like: spec: containers: command: image: tomcatjava:v6.0 imagePullPolicy: IfNotPresent name: example resources: limits: hugepages-2Mi: 40Mi memory: 250Mi requests: hugepages-2Mi: 40Mi memory: 125Mi Is this the LimitRange part already been done? Otherwise anything wrong with my understanding? Also, I found an subtle error in the error info when creating pods restricted by the hugepage , would you please take a look?\n-- thanks for the feedback, i have been distracted looking at other topics, but hope to return to this later this week.\nI have filed an issue related to hugepages, would you please check when you are free? link:\nyou can't re-open an issue/PR unless you authored it or you are assigned to it. <detailsIn response to : /reopen </details\nhi, I've worked on the HugePages isolated at container and not just pod boundary issue and would like to confirm some details. My first though was to implement this issue via docker, just like what cpu and memory did. But docker seems not support hugepage limit in present. Which means we have to get it done by coding in kubernetes or libcontainer, is that right? Please let me know if there are any other instructions that may help :)\n-- to support isolation at container boundary, we would need support from the container runtime. that would require support in the CRI to expose the resource, and support in individual container runtimes (contianerd, cri-o, docker, rkt, etc.) to actually make use of it. in practice though, i have not seen a major requirement to support container boundary isolation. do you have use cases in mind that would require it?\nI have no any other use cases as well, for now all the achievements fit well for our requirements. BTW, does the still need to move on? I found it seems already being supported by default.\nI don't think the intent was for to close this /reopen\nyou can't re-open an issue/PR unless you authored it or you are assigned to it. <detailsIn response to : I don't think the intent was for to close this /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-73a311e0fb78fe06f854f9b67bb3be0d69fae466b5afe5af981f39439c3cf6d4","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: Pods with volumeMount gitRepo volumes in 1.8.0 see the following error in kubelet logs (also in ) It is important to not that the git repo clone is done via ssh. I think what is happening is that hyperkube image now used to have command in it and now it does not have it. So the git clone command that kubelet runs fails with the above stacktrace since it cannot find . What you expected to happen: gitRepo is volume mounted. How to reproduce it (as minimally and precisely as possible): Reproducible in kubernetes 1.8.0+ hyperkube images. Anything else we need to know?: This probably is due to missing ssh binary in the new 1.8.0 hyperkube images. The commit might be an issue. Will comment further if I find otherwise. Environment: Kubernetes version (use ): =1.8.0 Cloud provider or hardware configuration**: Not Application. OS (e.g. from /etc/os-release): hyperkube base image Kernel (e.g. ): Install tools: Others:\nHow did you installed Kubernetes may I ask?\nhow does this help?\nNormally, it is suggest to install kubelet as a daemon on host instead of using hyperkube container. cc you may be interested in this issue.\nNeed your help in identifying the fix here. I can send something for review. Suggestions are welcome.\nThe problem is that ssh isn't installed inside of the container. I don't know if it should be either, as the hyperkube image grows so large. If you install the openssh-client, it should work\nYes I did identify that as the problem if you see the description I mentioned. I pointed to a git commit that might have caused this. I am not sure what package installation has changed in the hyperkube container. \"ssh\" used to be available and now it is not. For the gitrepo feature to work it should be available.\nAh okay. If it was there before, we might want to add it back. SGTM.\nYeah, I switched the hyperkube image base from to our with some additional dependencies back. Apparently due to lack of volume + hyperkube test coverage, I've missed a number of dependencies needed. cc\nwe could use dropbear as a replacement for just the ssh client?\nI would prefer to use the openssh client over the dropbear client. It has wider usage and a stronger security track record IMO. The dropbear client also has somewhat different usage and configuration, so arguably using it is still a regression over the previous state. Saving a couple MB isn't worth using worse utilities, however much fun \"dockerfile-golfing\" is.\nto fix\ncherrypick to release-1.8:\nThis should be fixed in v1.8.3."}
{"_id":"q-en-kubernetes-b42bb21fa6ea11da2e50c80e3ebc579d3657ae0bbb3b36ccf0c72a7b30fe8c11","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: What happened: I'm running a Redis pod with a PV provisioned as an Azure Managed Disk. It's got an exec readiness probe that calls in the container. At some point it receives a SIGTERM. The kubelet then tries to sync the pod and restart the container, but it panics in (looking over the code, that's the only place where the error from is ignored) As a result, the pod hangs in a state, since the readiness probe cannot be ran in a non existing container. What you expected to happen: The container would successfully restart. How to reproduce it (as minimally and precisely as possible): It's the 2nd time it happens but I haven't yet figured out the conditions to reproduce. Anything else we need to know?: Environment: Kubernetes version (use ):Cloud provider or hardware configuration**: Azure w/ Managed Disks OS (e.g. from /etc/os-release):Kernel (e.g. ):Install tools:Others:\nReiterating the mentions to trigger a notification: <detailsIn response to : Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nHi could you also share the storageclass, pvc & pod yaml files? I would like to repro, thx\nI may insert a nil check statment in this func: But I need to reproduce this bug anyway, so that I could find the root cause of this issue, thx.\nYes, an error check is what I'm hoping for :) I've used the chart. The PV/PVC were created manually, they're set to an empty storage class. I should mention that I've also noticed other pods being terminated/recreated at the time this one failed, however I didn't trace the cause of this yet.\nthanks for the reporting, I may have found the root cause, just need a few more debuging time, thx.\nGitHub didn't allow me to assign the following users: xiazhang. Note that only can be assigned. <detailsIn response to : /assign xiazhang Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign andyzhangx\n- Thanks for the fix!"}
{"_id":"q-en-kubernetes-0d5fbe44a2b03c6b67116f7bffb16750ef5c7051d39d0ca104216a0475b3df8d","text":"Our 5k-node cluster jobs recently started failing even before the e2e tests start. The reason is the step which checks that all kube-system pods are running is timing out: cc\nFrom the scheduler logs I'm seeing that the pods are being scheduled quite slowly, getting worse with time: 2-3 pods/min looks quite bad. cc\nAlso finding such traces: Seems like most of the time is being spent in priority calculation.\nFrom the pod spec, I'm seeing soft inter-pod anti-affinity at node-level topology: I'm almost sure this is the one causing the mischief. We've been having performance issues with anti-affinity traditionally :p\nOk... PR is the cause. It introduced anti-affinity to kube-dns. I think we should revert this change given known issues with anti-affinity. - Opinions?\nFYI - it is taking 80m until the last pod is scheduled when I'm running 5k-node cluster locally.\nThat's quite unfortunate. It looks like we should revert the change until the anti-affinity change is scalable enough to be used more broadly, unless the scheduler folks have a target fix for this use case.\ncc\nWe are very well aware of the performance issues in our affinity/anti-affinity functions, but I am not aware of a concrete solution at this point. This is partly due to the flexible design of affinity/anti-affinity concept. My humble opinion at this point is to drop some of the flexibility as they are rarely used, in order to streamline the algorithm and improve performance. Anyway, I think the proper solution to this problem needs more time than your requirements to address the issue in hand. So, you may want to change your configuration and drop the anti-affinity rule.\nwould it make sense to add a note to the anti-affinity user documentation to hint at performance problems at large scale, until this is fixed?\nRe-opening this issue as this problem has struck again. The following runs of our 5k-node scalability jobs failed to bring up the cluster on time: https://k8s- https://k8s- https://k8s- https://k8s- And the reason is kube-dns pods pending to be scheduled for a long time:\nThese failures started happening about 4 days ago and I'm pretty sure it's this one: Add self anti-affinity to kube-dns pods Sadly, we'll have to revert it again for now :( cc - Seems like the improvements to anti-affinity performance made under weren't enough?\nWe need to find other improvements for the predicate. If we cannot find major improvement opportunities, we may have to change its design to be more scalable at the cost of the reducing its flexibility.\n- my personal opinion is that we need to change the design anyway. Having it work both ways (i.e. existing pods everywhere impacting placing of our pod) will always be painful (especially for things like cluster autoscaler)\nI agree. The symetrical application of anti-affinity is problematic. I also think we need to change it.\nGreat. - do you agree with that too?\n, I'm wondering after some affinity/anti-affinity optimization in scheduler, do you think if it's fine to test coreDNS with anti-affinity policy in large cluster with 5000 nodes?\nWe probably should try to see where we are - but this requires testing before enabling that.\nI agree with - Could we first try checking with a benchmark integration test () as a proxy for real cluster test?\nI'm wondering how many real nodes are required for testing?\nSince the integration tests fake out the nodes, I think it's safe to crank it up to 5000 (that's the current largest integration test benchmark size) I'm also hoping to refactor our benchmarks soon, so let me know if you have any thoughts about them."}
{"_id":"q-en-kubernetes-410ef366e2b72f2d21696681942578239c8f98dffebaca626b2dfe032753546e","text":"BUG REPORT: /kind bug /sig aws What happened: Adding this annotation to an existing service: did not trigger an update on AWS ELB (no new tags). What you expected to happen: Expected to see those 2 new tags under the existing ELB: How to reproduce it (as minimally and precisely as possible): a service without annotation the annotation tags on AWS Anything else we need to know?: Creating a new service WITH the annotation will create the expected tags on AWS. Environment: Kubernetes version (use ): 1.7.9 Cloud provider or hardware configuration**: AWS OS (e.g. from /etc/os-release): Debian GNU/Linux 8 (jessie) Kernel (e.g. ): Linux ip-xxxxx 4.4.78-k8s SMP Fri Jul 28 01:28:39 UTC 2017 x86_64 GNU/Linux Install tools: Others: Using Kops\n+1\n+1\n+1\n+1\n+1\nI've done some work on a fix for this, will try and get a PR submitted later today."}
{"_id":"q-en-kubernetes-cf260a4414bee12fed0df1f9bbd824860b6046cff0bea685f4f14bd09d4b29cd","text":"On AWS, mutable API calls such as AttachVolume or DeleteVolume have separate API quota from API calls that do not modify the resource. Currently k8s makes a high number of AttachVolume and DeleteVolume API calls even though k8s can check if the operation is going to succeed.\n/sig aws"}
{"_id":"q-en-kubernetes-5c27b3b02c304571a16f876186f2122703d7dbcacd59c19977a1f2e30cde3b19","text":"We should make the CRI log parsing code in kubelet to be a library. So that container runtimes and external tools don't need to copy the whole code ( vendor it instead ). Referring\n/assign\n/sig node\nI can help on it.\n/assign\nGitHub didn't allow me to assign the following users: yanxuean. Note that only can be assigned. <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThanks. I have a branch ready for this. Will open a PR later and ping you when it is out.\nok"}
{"_id":"q-en-kubernetes-2354d38d89f6dd8b381ae31424a08aa7093eac46e22478abd675b16478095578","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT: What happened: When a statefulset object is created, a lot of controllerrevisions are created at the same time. the controllerrevisions have differenct hash value but the same Data.Raw. How to fix it: two controllerrevisions are seen as the same , when [ControllerRevisionHashLabel] , Controllerrevision.Data.Raw , Controllerrevision.Data.Object are strictly the same // this line shoud be to fix the bug like below: Anything else we need to know?: Environment: Kubernetes version (use ): 1.7.8 Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\nI would help fix it\n/kind bug /sig apps\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten /remove-lifecycle stale\nStatefulSet uses the following code to construct its history object (controllerrevision): However, collisionCount shouldn't be blindly increased when there's name collision. Because controller uses a cache store, the name collision could be due to a controllerrevision created but not yet observed by the controller. The right way to handle it is to first check if the controllerrevision with the same name has the same content as the one we're going to create, like how DaemonSet controller handles it:"}
{"_id":"q-en-kubernetes-f4435f48cba5c6b9ba6240af42e1efb7e38ee023f272e6d6f2f2182e6ebddb41","text":"/kind bug /sig cli What happened: Tried to list the age of pods in the namespace in an otherwise empty cluster $ kubectl --namespace kube-system plugin aging The Magnificent Aging Plugin. No pods What you expected to happen: That the pods in the namespace would have been listed How to reproduce it (as minimally and precisely as possible): the and from into . the command: Anything else we need to know?: No Environment: Kubernetes version (use ): v1.8.2 Cloud provider or hardware configuration: minikube & GKE OS (e.g. from /etc/os-release): Mac OSX El Capitan Kernel (e.g. ): Darwin Kernel Version 15.6.0 Install tools: Google Cloud SDK\nhi for cli issue, could you please file it in this repo next time ? thanks !!\nWill the k8s github bots be able to resolve references in commit messages made in kubernetes/kubernetes PRs be able to find the corresponding issue ticket in that other project? Semi-OT but related: I know there's an overall intention to refactoring k8s proper into many distinct projects, so it would be helpful if either the contributor guide contained a mapping of each and every module making up a k8s release with its corresponding github project and protocol governing the component. Or if each directory (module) in kubernetes/kubernetes contains its own explaining the contribution protocol. Right now documentation and information about ownership, contribution and protocols of various project modules are a sprawl and a bit of a mess, creating a significant barrier to entry (contribution). What's the proper place and form to discuss these sort of things?\nthanks, I think those are some good suggestions, as far as I know you could send an email to our mailling-list(google-groups) or create an new issue then cc"}
{"_id":"q-en-kubernetes-03c92a40809bb9bebafe7d5451e4686cb2d18a7692b03032be3fd36ff7b15566","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: See in : https://k8s- /cc Fixed in\n/sig aws\nAdd /priority critical-urgent since marked it as a release blocker.\nPing! Release team here. This issue doesn't have any associated PRs which are passing tests. As a reminder, 1.9 Code Slush is on the 20th, and Code Freeze is on the 22nd. Do you think you'll have a working fix by then? If you plan for this fix to go into 1.9, the issue needs to be approved, and approved for the milestone.\nthe associated PR is . It needs another look from .\nPeter, i just nudged that along.\nyeah I see your and . Thanks Davanum!\n[MILESTONENOTIFIER] Milestone Issue Needs Approval Action required: This issue must have the label applied by a SIG maintainer. <details<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Never automatically move issue out of a release milestone; continually escalate to contributor and SIG through all available channels. : Fixes a bug discovered during the current release. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details"}
{"_id":"q-en-kubernetes-9088f0ead78b2fff04dd418b6ce000d4b309048b0554dcd7a11b323e0a89de4d","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --BUG REPORT /kind bug What happened: I tried to submit a Service Replace () to the kube-apiserver and this results the kube-apiserver container crashing and restarting. PUT localhost:8001/api/v1/namespaces/<namespace/services/<servicenameWhat you expected to happen: I expected the server not to crash and to get validation errors if the request was invalid. How to reproduce it (as minimally and precisely as possible): Create a service, get the resourceVersion and clusterIP from the created resource GET localhost:8001/api/v1/namespaces/<namespace/services/<servicenameSubmit a replace command to PUT localhost:8001/api/v1/namespaces/<namespace/services/<servicenamewith body like If the same request is submitted with the nodePort fields in the ports array then the request succeeds. I get that I may be sending an invalid request without a node port (though shouldn't the merge deal with that?) but it shouldn't take down the API server. Anything else we need to know?: Environment: Kubernetes version (use ): Client Version: version.Info{Major:\"1\", Minor:\"6\", GitVersion:\"v1.6.3\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2017-05-10T23:29:08Z\", GoVersion:\"go1.8.1\", Compiler:\"gc\", Platform:\"darwin/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"8\", GitVersion:\"v1.8.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2017-10-11T23:16:41Z\", GoVersion:\"go1.8.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Cloud provider or hardware configuration: Bare metal cluster OS (e.g. from /etc/os-release): Ubuntu 16.04.3 LTS (Xenial Xerus) Kernel (e.g. ): 4.4.0-98-generic Install tools: Others:\nthanks for reporting this! Seems like this is similar to and should be fixed by . can you PTAL that PR to see if it's good to merge?\n/sig api-machinery\nDuplicate of"}
{"_id":"q-en-kubernetes-c6165eb60d2fa6134ed892e253816858f4aa3ad29aca9e33f3237a292c078a05","text":"When one of the aggregated API servers returns a different definition of an object than what the other servers return, the merged openapi schema may contain multiple identical copies of the object definition (they differ only in the name). Is this a BUG REPORT or FEATURE REQUEST?:BUG /kind bug What happened: After changing the definition of the RawExtension type in one of the aggregated API servers, the aggregator returned the following openapi schema (note that v2, v3 and _v4 are identical; don't mind the actual new definition of RawExtension - it's incorrect): What you expected to happen: The merged schema should only contain two different RawExtension object definitions instead of 4. Environment: Kubernetes version (use ):\n/sig api-machinery\n/assign"}
{"_id":"q-en-kubernetes-9a1238de1c4b0813e72eceedd9b4cbeef1b3d12bf0b7bc6fd41e77d11828f5e0","text":"https://k8s- https://k8s- cc: /kind bug /priority failing-test\nThis is at least twice in a week we've had cross break due to a public method signature change not being brought to all platforms () We should do:\nI'm lost - the PR in question adds an argument and fixes the call-site - what is broken?\nAhh, I see.\nHow did that PR get through, if it is so obviously broken on Cross?\nCross build doesn't run 100% of the time as a pre-submit (because it takes so long), and it can't block. It's something sig-testing is tracking, but there isn't a quick fix yet.\nLooks like was skipped for that PR: \"pull-kubernetes-cross Skipped\" not sure why yet.\nCross runs when a file matching: changes, (which needs to be fixed, opening a test-infra issue to expand the regex), but also some files could have build tags without the file name matching () Additionally cross is not blocking (), the underlying bug should be fixed now though so once we get the fix in to k/k we can make it blocking when it does run.\nI find has already sent a PR. Thanks!"}
{"_id":"q-en-kubernetes-346125e2da6893c97c11617d4b86f3d67d99d614b263d3a1dde1fc0bab91d262","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: When kube-apiserver's livenessProbe suggests that pod is unhealthy, kubelet tries to kill its container with 30 seconds grace period. The container is killed ungracefully with SIGKILL when grace period passes. This happens each time. What you expected to happen: I expected that SIGTERM will be passed to kube-apiserver which will try to gracefully stop before 30 seconds pass (at least in some cases). How to reproduce it (as minimally and precisely as possible): Inject kube-apiserver failure (e.g. ) Watch kubelet logs for container's death You will see that container is killed after 30 seconds each time Anything else we need to know?: I believe the problem is that in kube-apiserver.manifest uses '/bin/sh -c CMD' as a PID1 process which doesn't pass SIGTERM signals to child process. See Environment: Kubernetes version (use ): v1.7.11\n/sig node\nLooks like without the pipe command will have to be removed which will make the logs hard to track. There is indeed some which makes kube-apiserver run in background and trapping SIGTERM then send a SIGTERM to apiserver's PID.\nI think this is the case for all our components, not just kube-apiserver. - FYI\nHow about using 'exec'-based solution suggested here: In our case it means prepending 'exec': I did changed this in my cluster and the process seems to meet requirements: fd/1 and fd/2 are redirected to /var/log/kube- container is killed instantaneously If everyone is fine with such a solution, I can prepare PR adding 'exec ' in front of all kubernetes/cluster/saltbase/salt/kube-apiserver/*.manifest using '/bin/sh -c CMD' pattern.\nNot sure if understand what you mean. Before there was no logs, so /bin/sh -c wasn't needed. -based solution seems to have all required properties. What is your concern about using it?\nUmm. just ignore my previous comment. I just tried prepending and it works fine for me."}
{"_id":"q-en-kubernetes-4acb7ecd05e9c199186dfb06debf6d91c0b60ed3707740e0351c00fb351b6646","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: What happened: Now, benchmark test make docker as the only runtime. What you expected to happen: benchmark test should be generic to all CRI container runtimes Related to How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\n/area test /area node-e2e\n/sig node"}
{"_id":"q-en-kubernetes-1d61960bcc4b874029c504d8fec26e7edd4b95c6ded5d0f98e72ee2715d02acd","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: Priority admission controller of mirror pods. This is an error and should be removed. What you expected to happen: Integer value of Priority should be populated by the admission controller. How to reproduce it (as minimally and precisely as possible): Create a static pod with PriorityClassName set. The created mirror pod will not get the integer value of priority. /sig scheduling xref\nwhat is the kubelet expected to do with a low priority static pod if it gets evicted?\nalso doesn't this mean a user can claim their pod is a mirror pod via an annotation and get it admitted with any priority they want?\nknows better about kubelet evictions, but in my opinion a static low priority pod should be evicted similar to other pods. Whether it makes sense for static pods to be low priority or not is a relevant question. I would say it doesn't make sense, but one might find a reason to have static low priority pod. Not really. This check effectively bypasses integer resolution of priority causing the priority of all those mirror pods to be considered zero. This if statement is wrong and should be removed.\nIt also bypasses the check that prevents the pod creator from setting an integer priority themselves. In any case, agree removing the bypass is good\nStatic pods that fail for any reason (eviction is just one) are not re-run. To avoid evictions entirely as a static pod, users can use the critical pod annotation.\n, it seems we just log error if creating mirror pod failed; refer to . Does it make sense?\nI am not sure I understand the question entirely. The mirror pod exists only to show the status of the static pod. If the mirror pod doesn't exist, the static pod should still function correctly. Since that is in , it will be retried periodically.\nThe scheduler dependent on mirror pods for resource usage :). If creating mirror pod failed (e.g. wrong priority class name), there'll be ping-pong between scheduler and kubelet: scheduler send pod to node, kubelet reject it. But we can-t just return error, kubelet can start in standalone mode.\nHowever critical pod annotation is ignored if pod priorities are in use.\nThis is not the case. See in your issue.\nHowever critical pod annotation is ignored if pod priorities are in use. the intent of pod priorities was to replace the critical pod annotation with something more principled. is that not how it was implemented?\nPriority replaces most of the functionality that the critical pod annotation provides. However, the critical pod annotation is still useful for static pods, as it can prevent kubelet evictions. But for all other use-cases, I agree that priority is a much better way to manage which workloads are run, and which remain pending. I am not sure if the scheduler translates a \"critical\" pod to a \"system-priority\" pod for scheduling purposes, but the kubelet still prevents the eviction of static critical pods as an escape hatch for important static pods which cannot tolerate kubelet evictions.\nWhy would that not be modeled as a system priority? I expected critical pod annotations to not be present once the priority feature was completed.\nI agree with this. do you think we should not evict static pods which are system priority, so that even users of static pods can move from using the critical pod annotation to use pod priority?\nI think Kubelet should not evict static system pods, but even with the current logic of eviction, if these Pods do not use more resources than requested, kubelet never evicts them as their priority is higher than other pods.\nI think as a best practice users that write static pod manifests should apply a system priority to prevent eviction, but I do not think the kubelet should do anything magical about that source."}
{"_id":"q-en-kubernetes-9df8e5ba2e9f919d9bec92d550979c4e420a21d16e91da208714cd0aae742f47","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /assign What happened: The tests are failing again because the scheduler messages have changed yet again. I'm going to remove the check for the exact messages because this is at least the 3rd time this has happened. Instead depend on unit/integration tests to check for specific failures. Also there's no need for these tests to run across all types of local volumes because it's just testing the scheduler part."}
{"_id":"q-en-kubernetes-a3d500050110fa46acb05ec5be5def924145d5dad82bb32c3fe8d00d1dd85a4d","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: I tried to use pvs provisioned by the cephfs storageclass build from PV and pvc are OK but it failed when mounting. here is the output of the ceph-fuse mount command: I try the ceph-fuse mounting outside kubernetes environment, and found that I have to pass the --id argument to make ceph-fuse work correctly: so I add the --id option in file on line 401: and cephfs volumes can mount corretly with ceph-fuse then. Anything else we need to know?: My kubernetes cluster version is v1.6.7, running on arm64 servers, with ceph version 10.1.2, the kernel version is 4.4.58 . I backported the latest version to mount cephfs volumes with ceph-fuse.\n/sig storage\nDoes this problem still exist in the latest k8s version?\nyes, I've tested on kubernetes v1.10.0-alpha.3 with ceph version 10.2.10 and the problem still exists\nOK, let me see it ,and feedback soon. =)\nHi, one question is that do you use 'admin' as user or other user to mount cephfs by fuse ?\nwhen no \"--id\" specified, the default user is admin, which will result in bug when user is not admin.\n/assign"}
{"_id":"q-en-kubernetes-b9858e0baaee4f0ad9651e6053a4173caa66ca467e51b54e6d1c9d78936f9530","text":"/kind test-flake https://k8s-\n/kind flake\ncc\nSaw this on\nSorry, I haven't had time to look into this. Let's see if can get to this earlier than I do.\nI'm seeing this a lot in the rush to code freeze.\nif this is actually a 1.10 bug, can you add status/approved-for-milestone to it? Thanks.\nSomething suspicious from the , but probably it's not related to the failure:\nThe garbage collector's behavior looks right, it does delete the dependent configmap: The tested start at 16:44:27., and failed with \"test timed out after 5m0s\" at 16:44:50.. I'll check if other failed cases have the same causes.\nThe \"Kubernetes Aggregated Failures\" shows there is no \"TestCRDDeletionCascading\" failure in the last tests (). Maybe I used the wrong tool? cc\ncc because triage does not support pull-kubernetes-unit? I cannot find in as well\ngave a new failure sample: https://k8s- (And the triage dashboard can't find it). Anyway, the test showed similar behavior. The tested started at I0226 20:57:02.. The gc behaved correctly: And the 5min timeout occurred 23s after the test began. I'll check what's the 5min timeout for.\nThe 5 min timeout is for all the tests in the test/integration/garbagecollector package. The tests are run sequentially, so is the last one to run, and unfortunately uses up the 5 min quota. The two logs I looked into showed the test itself was healthy. We just need to increase the timeout limit for this package.\nfound a lot of failure samples here: https://k8s- The first 11 of them failed because the 5m timeout.\nIndicates this is still present and severe.\nI was trying to get recent failures, it seems we recently moved intergration tests out of the unit test suites, https://k8s-, where there is no failure yet. I'll try again tomorrow.\n(we rename the job this afternoon, it was named kubernetes-unit but only runs integration test)\nhttps://k8s- that looks all green so far\nCan we also check run time of the tests in that package over the past couple weeks? It seems like this just spiked. Did our tests slow down, or did we add tests to that package?\nas far as I can tell the runtime of the test didn't spike ( shows the test running in just over 300s and also included the original bump of the timeout from 300s to 600s), but the timeout was accidentally decreased from 600s back to 300s as an unintended side effect of\nah, ok\nSince it has passed the last 26 times, is this safe to reclose?\n/close"}
{"_id":"q-en-kubernetes-fba186c8bcf8192fecd39603df79086034a841b1628e75fdd3294455f8cd9fd2","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: Bug Report /kind bug What happened: This is related to I am trying to integrate Cloud Foundry UAA () with kubernetes using the OIDC flow and the JWT validation is failing on the kubernetes side with an error of emailverified is required. emailverified is not a required claim in the OIDC core spec From a security point of view emailverified is useful when Self Service is enabled which is true of internet identity providers User Provisioning workflow in almost all cases is admin driven and not self service and emailverified becomes an obsolete claim. With things as they stand today, this flow is not conducive to Enterprise Identity Providers. What you expected to happen: Authentication should succeeed How to reproduce it (as minimally and precisely as possible): Pass a JWT which is missing email_verified claim using the OIDC flow into kubernetes Anything else we need to know?: Environment: Kubernetes version (use ): Client Version: version.Info{Major:\"1\", Minor:\"9\", GitVersion:\"v1.9.0\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2017-12-16T03:15:38Z\", GoVersion:\"go1.9.2\", Compiler:\"gc\", Platform:\"darwin/amd64\"} Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\nReiterating the mentions to trigger a notification: <detailsIn response to : Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nIt's not marked optional either, which was the original motivation for taking the safe route and assuming false if it wasn't provided. This would be a feature for us to detect the presence of the claim and act differently if it isn't there. Do any other providers do this? Almost all that I've worked with do support this claim.\nHi , Enterprise providers like ADFS , Okta don't support this as part of the user schema. There is the concept of whether a use is active/enabled which is not the same as email verified always. If a user is not active, a token should not be minted in the first place. I am trying to understand what additional security does email_verified provided in the context of entire identity providers where there is no self self service use creation. Thanks, Sree\nOkta supports , but it looks like ADFS doesn't: We don't know when the provider is self-service or not. Not assuming makes the client logic uniform and less error prone instead of trying to reason about the intent of the provider. There's some discussion when we enabled this validation, including some UAA notes. , specifically\nMore than 80% of the enterprise customers of Cloud Foundry are using either AD/ADFS . This becomes a breaking change. UAA can add a email_verified claim. However its going to be of little value if the underlying AD doesn't propagate that. My two cents is this check doesn't make sense in an enterprise set up\nhas agreed to take this (can't assign her because she's not part of the org) /sig auth /kind feature /milestone v1.11 /priority important-longterm /status approved-for-milestone\n[MILESTONENOTIFIER] Milestone Issue: Up-to-date for process <details open<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Escalate to the issue owners; move out of the milestone after 1 attempt. : New functionality. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details"}
{"_id":"q-en-kubernetes-f2af98395fc32c90692404f5842866454ad062b03995abe94159620d2df7dd0a","text":"Is this a BUG REPORT or FEATURE REQUEST?: What happened: docs recommend using '--leader-elect' to have HA setup for kube-scheduler and kube-controller-manager. However, these two components approach differently what is default value for the flag. . . Seems that setting the value to true in kube-scheduler code was lost in refactoring scheduler API config in . The code before refactoring had the value . What you expected to happen: kube-scheduler set true as a default value for '--leader-elect' Environment:\nnot sure I picked right SIG\n/sig api-machinery\n/assign\n/sig scheduling"}
{"_id":"q-en-kubernetes-2f1261686a24e9a9b8cb41f2d352702bfa4eaf05fdd39b4b9e9d7a090314c871","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: What happened: When useInstanceMetadata is set to true for vmss nodes, their providerID is wrong. What you expected to happen: Their providerID should be in format azure:///subscriptions/&lt;subscription-id/resourceGroups/&lt;rg-name/providers/Microsoft.Compute/virtualMachineScaleSets/&lt;scaleset-name/virtualMachines/&lt;instance-id How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\n/sig azure\nadding priority important-soon so that this issue can stay in 1.10. Please update if incorrect. /priority important-soon"}
{"_id":"q-en-kubernetes-5db6b38c9e0a6f738b58da969e759df11708e8928302c7f7de882de90bc62043","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: When creating an IPv6 cluster, the --pod-network-cidr needs to be a /66, it cannot be for a larger network (smaller value) What you expected to happen: Should be able to use values smaller then /66, /64 or a /48 How to reproduce it (as minimally and precisely as possible): Create a cluster with Kubeadm, and set the podSubnet as below networking: podSubnet: \"fd00:1:2:3::/64\" Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: /area ipv6 /sig network\n/assign rpothier"}
{"_id":"q-en-kubernetes-d4bfe54ff37fb83a884623cd544dfe42ec65baa73bf3910601f0a26632482c49","text":"In following with history we will be curating action items here, but it's still a WIP, atm: All non-bug-fixes for this release cycle should be merged, but there is still an ongoing list of bugs that folks may execute against for the remainder of the freeze and we will evaluating on a case-by-case basis. /cc\nThank you for creating this issue!\n/cc\nAdding required tags. Please use this issue to supply a frequent summary update of the issues you are tracking in the kubeadm repo and their status. Especially, please call out and link to any bugs that are blockers for 1.10. /kind feature /priority critical-urgent\ncan we have a general status update on the kubeadm issues? The issues themselves don't generally have status updates. This is the last week before release, where are you with the kubeadm changes?\nWe're all green except for the 1.9-1.10 upgrade tests which appear to have other issues that are not kubeadm specific. Final default update PR is which will auto-close this issue.\n^ I don't know why we have this policy but it's ridiculous...\nare there any 1.9-1.10 upgrade issues which aren't already represented by issues against kubernetes/kubernetes? that is, are there any in the kubadm repo only?\nNot that I see atm, I still need to triage the inbound issues.\n/status in-progress\n/status in-progress\n[MILESTONENOTIFIER] Milestone Issue: Up-to-date for process Note: This issue is marked as , and must be updated every 1 day during code freeze. Example update: <details open<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Never automatically move issue out of a release milestone; continually escalate to contributor and SIG through all available channels. : New functionality. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details"}
{"_id":"q-en-kubernetes-9a2b07cf0c73dadc896291ace41839fae52c7d65b3c953ac91d92e03ffc8bbe1","text":"This test is failing in both gce serial suite: http://k8s- /sig network /priority failing-test /priority critical-urgent /kind bug /status approved-for-milestone cc /assign PTAL\nxref\n/milestone v1.10\nthis also fails in http://k8s-\nFailing to connect to Services at all. I can't even ping pod IPs:\nSo I can ping pods same-node and cross-node right up until it kills the apiserver. Rebooting fixes a given node. The apiserver demonstrably comes back - I can query it for info...\n- I could use help - have to shift attention soon.\nFWIW, I am testing on a Calico-enabled node\nI'll take a look at this as well.\nWeird, now even after a reboot, this condition is sticky. I have a cluster in this state, if you want to poke.\nugh, now it is resolved.\nI just repro'ed without the e2e - have a cluster, proved ping works, killed apiserver ping fails.\nInteresting - docker shows the pods not running, but apiserver says they are alive.\nUpdate: docker sees no running pods. API server thinks kubelet is running them. kubelet is getting \"\" from apiserver, even after kubelet restart.\nUpdate and recap: Bring up a cluster Run some pods Prove that pods are running and services work pkill -9 apiserver Wait. Watch kubelets \"lose\" pods that were assigned to them Observe that pods are not running and services do not work Observe that apiserver says the pods are running pkill -9 kubelet Observe that it is still broken kubectl delete one pod Observe that kubelet is fixed for that one pod, but not others Q: Why is kubelet not getting a full listing when it restarts?\nWhat exactly does it mean for it to be fixed for a deleted pod?\nKubelet started running the pod that was deleted - it got the update for that one.\nThe cluster I was testing was a week or 2 old. I built a cluster from HEAD and I can not repro - said the same. Is this test running against head of master?\nI also ran the test manually against my new cluster and it passed\nFWIW I was also unable to reproduce this yesterday on a cluster from HEAD.\nDid some investigations and had some findings for this specific failure: Traced down a bit on the history, the first occurrence was https://k8s- From the , seems suspicious because it modified . Also found reported a relevant symptom: The fix () for that was merged a day ago and may explained why we can't repro from HEAD anymore.\nnice, is this issue reproducible without the fix ()?\nSorry should have had confirmed this, bringing up a cluster now.\nQuick update: I was able to repro the same failure () from HEAD with reverted. Though this test is now flaky on some other errors: https://k8s- https://k8s- They all seem to failed while communicating with apiserver.\nThanks ! So fixed the issue and I am curious about what introduced the issue. Do you happen to know when did this test start to fail? Just trying to find out what went wrong in between. We have some other possibly related issues so finding out the root cause of this one would be super helpful.\n/assign\nThe first occurrence (for that exact failure) I noticed was https://k8s-, which happened on 2018-02-15 09:49 PST.\nYou all rock! I see a couple greens on it now, but still mostly red?\nThe test is now flaky mostly on creating an exec pod after restarting apiserver. Some sample failures: To elaborate a bit more, a snippet of the test procedure is: ssh to master and (sending a SIGTERM) Keep polling on /healthz on apiserver until 200 is returned. Create an exec pod for later steps (and failed on various erros --- EOF, connection refused, apiserver terminating). According to above test failures, my suspicion is that the test is assuming the apiserver is restarted too soon. It is possible that apiserver captured SIGTERM and did a graceful termination, so that polling on /healthz may succeed even in the termination phase. I'm not surprised trying to create a pod via a terminating apiserver will hit those errors. Could anyone confirme the SIGTERM theory?\nSpoke with folks, sounds like after SIGTERM apiserver does wait a bit for remaining requests to finish. I will take a stab to refine the test. cc\nSent .\nTest is now passing on gce but failing on gke. Apparently we don't have a way to trigger apiserver restart on gke at this moment. Sending a PR to skip this test on gke.\n[MILESTONENOTIFIER] Milestone Issue Needs Attention Action required: During code freeze, issues in the milestone should be in progress. If this issue is not being actively worked on, please remove it from the milestone. If it is being worked on, please add the label so it can be tracked with other in-flight issues. Note: This issue is marked as , and must be updated every 1 day during code freeze. Example update: <details<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Never automatically move issue out of a release milestone; continually escalate to contributor and SIG through all available channels. : Fixes a bug discovered during the current release. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details\nCommented too quick. On gke we do a same-version master upgrade. Will send another fix :/"}
{"_id":"q-en-kubernetes-f3c67bfc2950087449c2463907f069dc46972ea9be99ce2576cbe8e050232c18","text":"/kind bug /sig storage What happened: When mounting iSCSI PV kubelet doesn't give multipathd enough time to create multipath device and fallbacks to mounting single SCSI device What you expected to happen: it uses multipath device to mount How to reproduce it (as minimally and precisely as possible): create ~5 pods on a single node, each with dynamically provisioned PV (we use Trident for that), observe how some mounts are /dev/sd and not /dev/mapper/ Environment: Kubelet 1.9.3 Baremetal + NetApp ONTAP ove iSCSI + Trident dynamic volume provisioner CoreOS (kernel 4.14.24, multipath-tools 0.6.4)\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nI can reproduce this issue fairly often (although not reliably). The problem is in in the AttachDisk() function. The basic problem is that the method that finds the multipath device looks at each device path one time, and if none of them have an associated multipath device, it just falls back to formatting and mounting the first device path. There is no wait loop for the multipath device to appear. It would be easy to add a wait loop, but that would slow down the non-multipath use case. A possible optimization would be to only wait when 2 or more paths are known, but that still opens up the possibility of incorrect behavior when attaching while all but one paths are down, or when only one path is discovered (for whatever reason).\nAfter thinking about this over lunch, I think a possible fix would be: add a 10 second wait loop for the multipath device if and only if the PV are 2 or more target portals specified. This puts control in the hands of the provisioner without modifying the API at all, and not breaking backwards compatibility. The one corner case I can see is what would happen the PV has 2 portals, but the system is only able to log in to one. This would occur when multipath was correctly configured on the storage controller and in the provisioner, but one of the paths was down (for whatever reason) at attach time. I'm inclined to make the attachment FAIL if the multipath device can't be found within 10 seconds, which would be a change to current behavior, but this can easily be worked around by the cluster administrator setting up multipathd to create devices for single path disks. Provisioners would want to document the recommended way to do that (i.e. set find_multipaths = no, and configure the blacklist appropriately).\nis vaguely described as \"target portal ips for high availability\", without describing what the \"high availability\" means. Should we assume that it means multipath? Then Kubernetes can always wait for multipath device to appear when more that one is specified. Would it break anything? Does anyone run multiple portals for the same volume without multipath? Is it something we should support? any opinion?\nyes, the iscsi portals were made possible by for HA/mpio use case (see ). The lack of concrete description makes it less obvious though.\nwas it working perfectly with older kube versions and the issue started to pop up recently or from ? Why I am asking is, multipath support was in kube () and atleast I havent seen this issue reported till this report or very recently. Gluster team were also using multipathing feature for quite some time where we always got mpath device mounted reliably, but noticed this issue very recently or with latest OCP builds. iic, There are good number of changes happened in between especially on device discovery/attach ( ex: ) disk paths. I am not concretely saying its a regression or a behavioral change or side effect of some other patch, but trying to figure out if thats the case. any thoughts here ? my understanding was/is that with more target portal IPs means ( atleast) in ISCSI context, but we could have mention as well in the description. Suppose the backup portal IPs are filled by provisioner/admin but initiator havent enabled ( dm_mulitpath is not loaded or deamon service is not running) software in the kubelet/node, what will happen ? Is it going to be infinite wait for failure ? my understanding is that, if multipath is enabled, even if it failed ( for some reason) for backup paths, a mpath device will be created for single path in the system. If thats the case above shouldnt be a concern and device mapper layer will take care of the path discovery later when path becomes stable/available. I dont think so.\nwe started with 1.9.x on kubelets, so can't tell much about mounting behaviour of previous versions. it is going to fail, but it is fine, nodes should be correctly configured. Lets say you specify fsType, but given FS module is not loaded by kernel, or now mkfs.<FSbinary is available to the kubelet, mount is going to fail and thats OK IMHO. Kubelet shouldn't try to fallback to anything less than was asked, mounting via single portal when multiple portals are specified is similar to trying \"default FS\" if required fsType can't be used. Yes, default configuration for multipathd is option (note, that default in Linux distributions might be different), which creates multipath device for every non-blacklisted block device.\nI'm fairly confident that adding a wait loop for the multipath device in the case when there are 2 or more portals in the PV spec won't cause any regressions. The thing that gives me pause is that there are cases when there's a single portal in the PV spec, but the device still supports multipath and the existing sendtargets command is the mechanism for finding those other paths. I'm wondering if it's possible to detect that case (reliably) and use a wait loop for the multipath device in those cases too.\nis the default on RHEL 7, see\nDid you mean Multiple Connection Per Session (MC/S)? I don't think this is supported in open-iscsi. Open-iSCSI only supports MPIO (multiple portals)\nI don't understand the move by RedHat to default to findmultipaths = no. It basically guarantees bad behavior in cloud-oriented use cases when there's a temporary path failure. My reading of the linked document is that the main motivation for changing the default is to avoid the need to setup a correct blacklist. It seems to me that correct behavior is more important that avoiding some work up front when configuring multipath. No, I'm referring to the . If a storage backend has multiple target portals, but the dynamic provisioner only puts one portal in the PV spec, then as long as that one portal is accessible at attach time, open-iscsi will discover the other portals and log into them. The only weakness to this approach is what happens when that one path is inaccessible at attach time due to a network switch/network port/NIC failure, while the other ports are still available.\n, AFAIK multipathd (binary) in RHEL 7 defaults to , but configuration utility they ship writes config with , that utility needs to be called explicitly , IMHO that should be direct scsi mount unconditionally, without multipath involved\nThe thing that gives me pause is that there are cases when there's a single portal in the PV spec I would be inclined to agree, except that someone the sendtargets logic to the iscsi code. This code exists for no reason other than to allow finding multiple path given a single path. In fact the existing NetApp dynamic provisioner (Trident) relies on that behavior currently to get multipath support (this is something I've fixed and will upstream shortly). If we decide that PVs with exactly one portal should not support multipath, then we should remove that sendtargets logic from the iscsi module in kubelet."}
{"_id":"q-en-kubernetes-1957c684b46f7dba3b59a7efc638192de1585f81f68da93f440c101bf5bc4bbc","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: Download the kube release source artifact from and do a build. All the kube binaries have a extraneous comma in the version: $ /usr/bin/kube-proxy --version Kubernetes v1.9.3, What you expected to happen: kube binaries built from git archive source tgz to show the correct version without extra comma. this was introduced in git attribute export-subst on creates a string format like ''HEAD -release-1.9, tag: v1.9.3, origin/release-1.9' . so, the regex capture should ignore ','\n/sig release\nThe issue is that our regex didn't handle the case where the tag isn't the last part of the string filled in by : One the 1.9.3 source tarball linked, this looks like This should be a pretty easy fix - I don't think commas are allowed in semantic version strings. /assign\n/kind bug /priority important-soon\nThis PR likely affects v1.10 as well. I have a fix out:\nConfirmed: this affects v1.10.0-beta.0 (trying ): Though applying my fix from reveals a different bug: The issue then is that there are multiple tags matching that commit when it was built: who could guess that hacks would break?\ndisregard the second part of my comment above (at least for now); v1.10.0-beta.1 has a more reasonable string: (but it still demonstrates the original bug this issue mentions)\nwe need a sig lead to status/approved-for-milestone here\n/status approved-for-milestone sure, good to be fixed :-)\naaaaaand priority increased and re- to milestone....\n/priority critical-urgent\nMaster branch is fixed. Creating cherry-pick PRs now: 1.9: 1.8: 1.7:\n[MILESTONENOTIFIER] Milestone Issue Labels Incomplete Action required: This issue requires label changes. priority: Must specify exactly one of , or . <details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details\nCherry pick PRs are in. Next patch release of each branch should be fixed. /close"}
{"_id":"q-en-kubernetes-4b95e73f98faca5940a16ee2fccddf37ae93437213e073578ff34d76b732271f","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: Service loadbalancer source range don't work in ipvs proxy mode. Which result in gci-gce-ipvs CI failed. http://k8s- How to reproduce it (as minimally and precisely as possible): In gce environment. a service with backend. service type to service util service external IP is specified. the service from pod inside and pod out of the . Environment: Kubernetes version (use ): master Cloud provider or hardware configuration: gce OS (e.g. from /etc/os-release): ubuntu Kernel (e.g. ): Install tools: cluster/kube- Others:\n/kind bug /sig network /label ipvs\n/area ipvs\nThe test seems to be flaky now. Can you take a look please?\nI don't know why, but I think this flaky have the similiar reason with other normal service-test cases, like Those flaky are pretty annoying, problem is, I have run those cases in my own environment many times, never fail... : ( But still, I'll do my best to find the reason."}
{"_id":"q-en-kubernetes-1e6e7a922459aa1ae762d90ee2c0c561e09f7928de6f5ad4ad5cc72934b7dcd1","text":"New tests for subpath consistently fail in ci-kubernetes-e2e-gce-multizone: https://k8s- /kind bug /sig storage\nIs this a blocker for 1.10?\nI would expect so, yes. /milestone v1.10\nI'm on it /assign\nThe problem is likely due to inline GCE PD volumes not having enough zone information. Considering changing volumes to PVs instead.\nfix was in progress in to change to a PV\n/unassign\n[MILESTONENOTIFIER] Milestone Issue: Up-to-date for process Note: This issue is marked as , and must be updated every 1 day during code freeze. Example update: <details open<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Never automatically move issue out of a release milestone; continually escalate to contributor and SIG through all available channels. : Fixes a bug discovered during the current release. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details"}
{"_id":"q-en-kubernetes-897c473b11be2622aeec9f7a71c36a9c53396d522c6fb073dc9f488488c3739c","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: e2e test \"gci-gce\" is running in 2 versions: with kube-dns as DNS Service : https://k8s- with CoreDNS as DNS Service : https://k8s- In CoreDNS use case the test is failing What you expected to happen: This e2e tests should not depend on what DNS service is running as default in K8s. CoreDNS is currently BETA, will move to GA soon (after v1.10 though) See : How to reproduce it (as minimally and precisely as possible): look at logs of dashboard : https://k8s- go run -v --test --test_args='--[sig-autoscaling]sDNSshorizontalsautoscalings[DisabledForLargeClusters]skube-dns-autoscalersshouldsscaleskube-dnsspodssinsbothsnonfaultysandsfaultysscenarios$' Anything else we need to know?: Environment: Kubernetes version (use ): branch master Cloud provider or hardware configuration: GCE Install tools: deployment intra-test Others: see https://k8s-\n/sig autoscaling\n/unsig cli\nLooks like the test is searching specifically for a deployment labeled \"kube-dns\". There are several references to \"kube-dns\" scattered around it, actually.\nRight now the tool \"dns-horizontal-autoscaller\" seems dedicated for kube-dns. This tools is deployed as \"add-on\" when enabled by this ENV variable : I guess we should have the definition of another dns-horizontal-autoscaller, dedicated to coredns (unless we generate the yaml file). and the choice of deployment is based on the ENV variable . see\nCurrently, the translation of the deployment of \"dns-horizontal-autoscaller\" for CoreDNS is in progress by It is expected that the corresponding e2e test will work as is (as the artifacts are selected by labels, which are the same for CroeDNS and kube-dns)."}
{"_id":"q-en-kubernetes-c3f9dbcfbf0abfa183188a971933f09b1bd1d9b55d8d4f63807c7685fadf987b","text":"/sig gcp /sig cluster-life-cycle /kind flake /milestone v1.10 cc https://k8s- starting last night /assign can one of you verify?\n/kind bug /priority important-soon\n/priority critical-urgent /priority failing-test /status approved-for-milestone\n[MILESTONENOTIFIER] Milestone Issue Needs Attention Action required: During code freeze, issues in the milestone should be in progress. If this issue is not being actively worked on, please remove it from the milestone. If it is being worked on, please add the label so it can be tracked with other in-flight issues. Note: This issue is marked as , and must be updated every 1 day during code freeze. Example update: <details<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Never automatically move issue out of a release milestone; continually escalate to contributor and SIG through all available channels. : Fixes a bug discovered during the current release. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details\n/assign according to , gonna move those jobs to us-central1-c instead"}
{"_id":"q-en-kubernetes-8e791342a72fb228aa97c5b91a2cfb5922e30c40e783a5d69356e7bc9daaf03d","text":"/kind bug What happened: We had a pod with persistent storage. We drained the node. The node VM was deleted. The PV was not attached to the node that the Pod migrated too. The kube-controller-manager logs contained the following. What you expected to happen: The PV can be attached to the node it was migrated to. How to reproduce it (as minimally and precisely as possible): Create a 2 node cluster on vSphere Deploy a pod with persistent storage attached Drain the node Delete the node VM immediately after draining Anything else we need to know?: The issue seems to be in the call to getVSphereInstance. This will return an error if the Node doesn't exist. Precluding the logic on line 804-813 from being executed. A restart of the kube-controller-manager resolved the issue. The following e2e test is not triggered in vSphere environments - Which might expose the issue. Environment: Kubernetes version (use ): 1.9.5 Cloud provider or hardware configuration: vSphere OS (e.g. from /etc/os-release): ubuntu Kernel (e.g. ): Install tools: CFCR Others: /sig storage /sig vmware\nMore logs:\n/assign"}
{"_id":"q-en-kubernetes-3e77b8825eecef15f3da8f836140d3907ef15c25fa9ca3794d81a4d982094993","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . --Is this a BUG REPORT or FEATURE REQUEST?: What happened: Pods stuck in terminating state on Windows node after deleting it What you expected to happen: Pods get terminated and should disappear from kubectl How to reproduce it (as minimally and precisely as possible): Run a pod with windows container Delete it Pods are still terminating Anything else we need to know?: Could you pls help to resolve this issue. K8s pods stuck as Terminating after getting deleted on windows node with windows container(IIS). Docker ps -a not showing this container, so it is completely deleted, but in kubectl it is marked as Terminating. For networking we using OVN. kubectl logs: kubelet logs: Environment: Kubernetes version (use ):Cloud provider or hardware configuration: Hyper-V VMs OS (e.g. from /etc/os-release): Windows Node: Master Node: NAME=\"Ubuntu\" VERSION=\"16.04.4 LTS (Xenial Xerus)\" ID=ubuntu IDLIKE=debian PRETTYNAME=\"Ubuntu 16.04.4 LTS\" VERSIONID=\"16.04\" HOMEURL=\"\" SUPPORTURL=\"\" BUGREPORTURL=\"\" VERSIONCODENAME=xenial UBUNTUCODENAME=xenial Kernel (e.g. ): Linux esd-kbrnts-lnx-1 4.13.0-37-generic ~16.04.1-Ubuntu SMP Wed Mar 7 16:03:28 UTC 2018 x8664 x8664 x8664 GNU/Linux Install tools: OVN 2.7 on windows OVN 2.7.2 on linux Others: Docker Version on Windows node: Server Version: 17.06.2-ee-6\n/sig windows\nIt needs sometime for kubelet to remove the pod. Could you wait a while and check whether the Pod removed?\nHi actually I have waited for 1 day, and it is not deleting. Can delete it only using --grace-period=0 --force. But I don't want to use this manual approach). Could you pls give some advice on this? Thanks in advance!\nis a reference to this issue. Checking kubelet logs will always help to troubleshoot such problem.\nActually, I have attached kubelet log in the ticket and think the main reason is Also as there any option for kubelet to automatically force delete Terminating pods after some period? And --containerized is also not solving my issue. Thanks!\nKubelet won't delete the pod from apiserver. Since it is already not exist on the node, you could force delete it by kubectl.\nIt is clear, so how we can resolve this error: ? I think this is the core item of our issue. Thanks.\nI'm not using OVN. Have you met this problem when using OVN?\nadding Bob to comment on this. i have seen this issue before, but it has been quite a while.\nHi, Guys will be very grateful if you will provide some solution for it. We have tried to add to deployment spec but it doesn't help. Thanks.\nThe error message of is actually a warning message. It shouldn't block the removal of pods. Could you recheck the following things: the node, check whether the pod is still existing, e.g. by the pod exists, check if there are other error or warning message in kubelet logs; or else, the pod could be force removed safely from apiserver\nHi, I had posted this info in the ticket description. not showing this container, so it is completely deleted, but in kubectl it is displayed as Terminating.\ni think you may get more support and faster responses in our slack channel for sig-windows.\nSo it could be safely force removed from API. And agreed with slack is a more faster way to such troubleshooting issues.\nThe issue seems to be related to . The only platform specific implementation for PodContainerManager is the one for . There is a default implementation (), that always returns true in the Exists call. Disabling the CgroupsPerQOS option on the Windows nodes resolves the issue.\nWindows doesn't support cgroups. You should always disable it.\nit is clear enough that Windows does not support cgroups). Unfortunately, it's not specified in the documentation. So to disable CgroupsPerQOS you should need to add this flags to kubelet: Maybe it will help someone... Thank you all for discussion!"}
{"_id":"q-en-kubernetes-336255f6e649bc5e9a36c586d29d9f11ed1db30d4ded4439288854d114501c77","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: In the kubelet, when using the kubelet creates static pods but doesn't update the status accordingly in the PodList (kubelet ). You can find that kind of Pods with the following annotation: The status will stay like that: What you expected to happen: The status is updated. How to reproduce it (as minimally and precisely as possible): Use a static pod. Anything else we need to know?: I tried to fix it in but the test-grid continuously failed at the reboot phase (see ). I had to revert it (, ). I'm creating this issue to track this bug. The problem in the e2e tests are probably from . This is an old output of the failure: https://k8s-\nI don't think this is a bug. only exposes what kubelet sees from the sources, and doesn't include statuses known only to kubelet. Since for static pods, their statuses are not written back to the source, they do not show up in . The fact that it has the condition is unintentional and a bug. It has just been fixed recently.\nI confirm that this behaviour is still happening. Having pods with status where they are actually . The kubernetes API shows them as when Kubelet endpoint as .\n/sig node\nI agree with that this is working as intended.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nI understand that the source of truth is currently the API server and that details can be retrieved there. We have however found this to be a scaling challenge. In our use case, we are a monitoring application that runs as a DaemonSet. Monitoring systems such as ours need to continuously refresh the PodList because Kubernetes operators want a real time feedback on application health via metrics, logs and traces. Having each node that is monitoring query the Kubernetes apiserver to get each of the Pod statuses results in high load on the API server. This is exacerbated and becomes difficult to scale as clusters grow to a large number of nodes. We have however found that having each node query its own Kubelet API locally is quite scaleable and allows this work to be distributed across the cluster. Ideally we would be able to use this API for static pods as well since the state of the static Pods is stored inside the kubelet. Other benefits of using the kublet API are that: It provides the link to the container runtime in the section, which is helpful for metadata collection and association. Is easily managed the RBAC system through webhooks If there is another approach or API that you’d suggest we’re happy to rethink how we collect this data.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nI also meet this issue.\nReporting the status as \"Pending\" might be consistent, but it's incorrect. If the pod is created statically and is health checking OK, it's not \"Pending\". The Kubelet API is almost very useful but the status flag seems to make it much less so since it's not possible to tell from the API endpoint if a static Pod is actually up successfully or not.\nHi team, I'm reaching out to see if we can revisit this issue. We need this to be able to monitor pods locally without having agents reach out to the apiserver. Currently the pod is in a wrong state (Pending when it's actually Running), and we're missing all status information (pod IP, container statuses, etc.). We would be happy and willing to update to solve the issue if we get the green light that this will be accepted. cc\nI think as long as we fix the test failures the change caused, we can re-introduce it. Feel free to assign me on a PR which accomplishes this, and make sure the reboot test passes.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale"}
{"_id":"q-en-kubernetes-47f4fe1153b9c8db2f96f1dbd3b510d20b25e68b2d642a04fba8f7548b259f44","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: says: What you expected to happen: Don't suggest that -- as it doesn't actually do that when used. How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: (This was reported in but that issue was closed) Environment: Kubernetes version (use ): (This is indeed head from today) Cloud provider or hardware configuration: macOS OS (e.g. from /etc/os-release): Kernel (e.g. ):Install tools:Others:\n/sig cli\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale"}
{"_id":"q-en-kubernetes-ae2354e5f90f5ada93795ba7ce49d7414be41b906f7bd0cd4821d722fa0df6ba","text":"https://k8s- Please work on finding a fix for the flake. /priority failing-test /priority important-soon /sig api-machinery /kind flake\n/assign\nI think this was fixed by\nwe had a failure on 3/28 (after the above fix merged). https://k8s- Looking at the logs this seems to be a different error than the one fixed by . Can you confirm plz?\nHmm that is different, and I will look into the reason.\nThere hasn't been a failure since 3/28, So while the test might have flakes it is definitely less flaky than 12 failures per week now\nYes the latest runs look good. Please close the issue\n/close"}
{"_id":"q-en-kubernetes-9b1fa47804b1f7e830fa325e71a35010f3c16e79f3c6e5b6bb423d5ac4aea52d","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: Many service related test cases in test case are flaky. Especially related cases. CI Link: http://k8s- What you expected to happen: IPVS should sync rules within . How to reproduce it (as minimally and precisely as possible): a cluster with ipvs proxy mode, and specify . a service with backends. check the delay from create service to sync ipvs rules. (This issue is probabilistic, about 50% possibility to happened.)\n/sig network /area ipvs\n/assign"}
{"_id":"q-en-kubernetes-178631faa4b3e89ae90e213ef25c42d7a4deb3801f14f70b17d85af2c5cd7474","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: Deploying kube-proxy with mode ipvs on a system with the required ipvs support compiled into the kernel rather than as modules causes the kube-proxy deployment to fall back to using iptables. tryIPVSProxy is only checking for the presence of kernel modules and not if the running kernel has the required support compiled in. What you expected to happen: tryIPVSProxy should succeed if either the required modules are found or support is compiled into the running kernel. How to reproduce it (as minimally and precisely as possible): Boot a system using a kernel with ipvs support compiled in rather than as a module (such as the kernel used by and a host without iptables installed Add --config=/etc/kubernetes/kubelet- to the kubelet arguments using the following kubelet- contents: Install using kubeadm with the following config: Check the output of the kube-proxy pod: Anything else we need to know?: From , it should be possible to check for the existence of modules.builtin and if present use that in addition to using modprobe and the contents of /proc/modules. Environment: Kubernetes version (use ): v1.10.2 Cloud provider or hardware configuration: local kvm host OS (e.g. from /etc/os-release): custom linuxkit build Kernel (e.g. ): Linux linuxkit- 4.14.40-linuxkit SMP Thu May 10 10:06:57 UTC 2018 x86_64 Linux Install tools: kubeadm\n/sig network\n/cc\ncc\ncc please take a look."}
{"_id":"q-en-kubernetes-d943f6711cd68b765c622b0185eababba15b96010c64af1a6f9d1d7374169a9c","text":"both in PR and CI runs see /assign\n/assign\nInteresting fact: is not flaky, only . The test codepath for these two are almost identical. I wonder if this is a real bug not a flaky test. still investigating.\nstill seeing flakes like https://k8s- the loop that ensures the watch cache is primed goes version-by-version, and creates and observes the delete event for that version. it does not ensure the delete event is drained from the watch caches for the other versions.\na better approach would be probably be to:\nwhy the poll? Because the watch is not reliable? Otherwise, we have that structure now.\nMoves the version loop inside the poll Today we loop over versions and poll per version, which can leave unconsumed delete events in prior versions\nCannot follow. Now we have the following without a poll:\nAt minimum, I would move the version loop inside the poll so that when we exit we know the deletion events from isWatchCachePrimed are consumed by the watch caches for all versions. If we wanted, we could move the create/delete up out of the isWatchCachePrimed function or pass versions into it to avoid creating/deleting an object per version\nOk, so we were talking about different code. good. To understand: we do a call per version, each creating and deleting an object, and waiting for the delete event with a watch call. a following watch will not see the events of the calls of itself (because they use the same versioned watch cache) and of any version in front of it in the version slice (is this true?), but it might see events of calls of versions later in the version slice. So with your outer poll and inner version loop we would make sure that we drain every version's cache for following rv=0 watch requests. Sounds plausible.\nWhere do we have the official watch axioms?\nI think I know what is the problem here. We make sure the watch is primed for . then we move on to and make sure the watch is primed by creating another object in . The watch cache is drained for but the watch cache for is not and that is the problem here. will send a fix.\nyep, same thing I described above. The solution looks good.\nI didn't see your comment when I commented. Your description is accurate."}
{"_id":"q-en-kubernetes-093df870a7b5c49df22f3deeaa7adc10504bc34c455e3bb09727f921a17ea670","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If this may be security issue, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: During k8s cluster launch, is checked and if any etcd binary is found regardless of current OS Distribution, expected etcd binary is not downloaded and etcd soft link is not updated to point desired etcd binary, so cluster creation fails. What you expected to happen: Along with expected etcd version, OS Distribution should also be checked for successfully and efficiently installing etcd during setting up local cluster. How to reproduce it (as minimally and precisely as possible): Checkout k8s codbase on MacOS and launch cluster. working k8s codebase under virtual machine to launch cluster in VM. (Cluster creation would fail, prompting error that etcd needs to be installed and update in {KUBEROOT}/thirdparty/etcd-v{ETCDVERSION}-linux-* Anything else we need to know?: It's an optimization on managing thirdparty etcd binary for local cluster setup. Environment: Kubernetes version (use ):Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\ndo you have a patch that you are working on? we don't really support local-up-cluster on macos yet. /sig testing\nis fixing this issue. Though I use Mac as my workstation but sometimes need local-cluster to test codebase so I mapped (softlinked) my GOPATH in virtual machine keeping separate bin directory. Everything is running perfectly fine after making changes mentioned in this PR. So thought to submit small optimization on etcd installation logic.\n/close"}
{"_id":"q-en-kubernetes-cab42aeb58a419434b628a8ba1af6fcf4b705a5482ca0d7362b0ced98b6e521a","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: What happened: In order to migrate to a new version of kubernetes on GKE. I cordoned(Kubectl cordon <node) all nodes in the old node pool(1.8.9) in order to drain the node pool(pods on the node) with an intention to cause the pods to run on a new version of the node pool(1.10.2). As soon as I cordoned all the nodes in the old node pool, it caused the downtime of ingress(nginx). The nginx controller pods where running fine on the cordoned node pool. But the cordon caused the nodes to be removed from the target pool of ingress load balancer. Below are the documents that suggest the above same method that I followed as best practice for zero downtime upgrade. 1). 2). What you expected to happen: I expected \"kubectl cordon <node\" not to have the node/s removed from the target pool of the loadbalancer because the nginx controller pods where running absolutely fine on these nodes. How to reproduce it (as minimally and precisely as possible): 1).Cordon all the nodes in an old node pool that runs nginx controller. Anything else we need to know?: Environment: Kubernetes version (use ): 1.10.2 Cloud provider or hardware configuration: GKE OS (e.g. from /etc/os-release): Container OS Kernel (e.g. ): 4.14.22+ Install tools: Others: Cordon causes to remove the nodes from the target pool of the ingress load balancer. The health checks fail on the nodes of the target pool, which should not be the case. Because the nginx controller pods run fine on these nodes.\nTechnically, this isn't a Kubernetes bug as this is working as intended; however, it does beg the question why we remove unschedulable nodes if they're marked as ready since we perform the to determine LB backend nodes. is it scheduable? is it not a master? is it not marked for exclusion? is it marked as ready? If any of those are a no, then it'll be removed from the pool on the next sync. Perhaps someone from can shed some light. /sig network\ncc This problem extends beyond GCP/ingress controller and to anyone using pools to upgrade nodes. A typical workflow I see is you'd cordon a node pool so as you drain each node the pods don't go to another node in the pool that'll shortly be drained as well. The current logic does not support that and complicates cluster administration because cordoning does not respect PDB's.\n/remove-sig apps /remove-sig gcp (since the service controller falls under sig networking)\nThose labels are not set on the issue: <detailsIn response to : /remove-sig apps /remove-sig gcp Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nRef I thought a general practice for draining the old pool is to move over one node at a time in a rolling fashion. Shouldn't there always be certain backend pods being able to receive & serve traffic? cc\nWRT draining, it is recommended to do it one node at a time ; however, in real world scenarios doing it one node at a time can cause a lot of avoidable container churn if you don't cordon off the nodes you at least plan to drain so that pods aren't being scheduled on a node you'll be draining after the next one. According to the comments in the service controller it seems the unschedulable part exists because it was how we differentiated masters in the past; however, we have labels for that now. We also have the service exclusion label (can't find when/if that'll be promoted though as i don't think there's a ticket on it).\nI'm having problems with this issue as well. The fact that nodes are removed from target pools when cordoned means that I sometimes get brief interruptions of the traffic to a service when I cordon a node. I'm guessing the node isn't gracefully removed from the load balancer which results in some traffic being cut off trying to route through the recently cordoned node. The same problem occasionally happens when deleting nodes / node pools or when the cluster autoscaler deletes a node. We are running GKE ver. 1.10.2-gke.1. Our best defence against these issue have so far been to limit node changes as much as possible and not use any auto scaling.\nI've posted a question regarding this issue in SO some weeks before it was created. The suggested action was to use node taints () instead of cordon, that way nodes will still be marked as .\nThank you for the link, it Certainly helped. Perhaps, if thats the case(as mentioned in the accepted answer of the above link), then to cordon all the nodes of old node pools should not be a recommended action, at least in the GKE upgrade docs.\nonly says: And I can't find what more it does in the documentation. So I would expect that it has exactly the same behaviour as a taint. Why is removing running pods from the balancer working as intended? If it isn't a bug then the documentation is certainly unclear in this regard.\nThis just caused a major outage for us too. I agree with the previous posts, while this may be working as the code intended, it does not follow the patterns and documentation k8s provides. When I cordon a node it should not impact the traffic going to the node or any pods on it. I should be able to gradually remove traffic by draining nodes/pods and respecting PDBs.\nthis issue touches on the fact we are removing masters as well and inadvertently removing cordoned nodes. The logic in the service controller mentions we only filter out unschedulable nodes since that's how masters were setup.\nWe wanted to replace all of the kubernetes cluster nodes to different node instance types, so we started off by cordoning all of the nodes, but it surprisingly caused an outage for us. As pointed out, the only documentation around cordoning is , which I took to mean that new pods can not be scheduled on those nodes. It was surprising behavior to find out that, not only does it not allow new pods to be scheduled on those nodes, but that it also causes the service controller to remove unschedulable nodes from load balancing pools. I understand that it's working as intended, but if that was also documented as part of the cordon operation, I would have been able to avoid an outage.\nI finally caught up to this bug. Fun one. I agree that this is working-as-intended and that it stinks. We can do better. As a short-term workaround, you can taint the nodes before cordoning one-by-one. That will stop new work from arriving, but will leave old work there. If we agree on an appropriately named taint, we can have nginx tolerate it by default. Am I missing any reason why this would not work? If that flies, we should fix docs ( those are GCP docs). I am happy to help with that. Longer term maybe we can do something smarter. I am not sure what that would be, just now. Ideas?\nIs there any reason we can't just leave the nodes in the load balancer pool when the node is cordoned? And when the pods are terminated the nodes can be removed? I may be missing some context here?\nThe presumption is that a node that is unschedulable is going away, or needs some other repair. Using it as a target for load-balancers means there could be sessions routed THROUGH that node. Even as the pods are drained. Right up until the node suddenly dies, causing an outage of a different flavor. What should really happen is that a node which becomes unschedulable starts draining connections. Once all the pods are moved off, the node can be removed from the pool entirely. I don't think any cloud providers, even GCP, handle this just right today. David Radcliffe <:\nWait, if you brought up a new nodepool, all those new nodes should be in the TargetPool. Cordoning the old nodepool should only terminate the existing connections, new connections should be fine. Something doesn't add up...\nWas this using\nYes\nRight, well, that explains it. So even draining won't do what you want - you need new connections to be allowed until you have capacity in the new pool. The good news is that you can do this entirely yourself, with taints. We just need nginx to tolerate the taint, which you can also specify yourself. The only real alternative I see would be, as you said, to leave nodes in the LB pool until the node dies. Given that we never specified this, it is probably different on every provider, which would make it hard to change (especially if the provider already provides clean drains while unschedulable).\nIIUC, we don't want new pods to land on the tainted/cordoned nodes. In that case, we can place \"NoSchedule\" taint on the nodes. We don't need any tolerations on nginx. This will prevent new instances of nginx from landing on the tainted nodes, but the existing ones will keep running there.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nThis just surprised us also. I understand the technical reasons behind it but the command name and associated documentation is misleading. To me it isn't intuitive that the command does not remove compute (pods) but does remove some networking. I would expect cordon to keep new pods from getting scheduled to it and keep it from getting to new services/ELBs/etc. The command would then cause pods to be removed and also take the node out of networking paths. Almost need a new state for a node of that is clearly defined.\nMy main issue is with the docs, and also GKE docs. They should explain this pitfall. After I got aware of the issue I understand why it works as it does. A simple workaround is to taint the nodes you are going to rotate first.\nJust chiming in; We as a GKE customer have today been bitten by this issue in production. The documentation of (in the context of GKE) is not extensive enough so we could be aware of this issue. This should be addressed. Luckily the outage was not severe because we were using the node-by-node approach (and accepted the container churn this would cause). We explicitly took this approach because our previous nodepool upgrade was a total outage also caused by in a different way. (for those interested in our total outage, not related to this issue): Due to the fact that also results in the re-calculating the amount of schedulable nodes and thus scaled the amount of pods in our cluster (in our case: from 110 back to 2) resulting for us in major DNS outage internally. Be warned that for a nodepool upgrade can have a lot of unwanted & unexpected side effects.\nIt looks like this happens only to AWS and GCP but not on Azure? Can anyone confirm this? I just tested in both AWS and Azure, and while I see logs like in AWS Kubernetes, I do not see any logs like that in Azure. Is it intended? If so isn't behavior different for each cloud provider? i.e.) cordoning a node in AWS Kubernetes prevents loadbalancer from routing traffic to the affected node but cordoning a node in Azure Kubernetes will still let the loadbalancer route traffic to the affected node. Is it intentional?\nBesides running a test like the above, how would I determine what the behavior is on AWS? I haven't found anything in the node administration docs or the Service that would imply that would shift traffic away from a Pod. In fact, the doc for explicitly states, emphasis mine: If I hadn't come across this Issue due to reports from GKE, I suppose it would only have been a matter of time until we had a high-impact production outage. There's a disconnect between Kubernetes API and the LoadBalancer as to whether Pods running on a 'd node are Ready. From all the docs I've seen, I would expect that either a) would evict Pods over to a Node before setting the instance as , or b) LoadBalancer would not equate with\nHi All, What approaches people using too workaround this? From reading the comments I was hoping to follow the custom taint approach. However, I've noticed executes a cordon making it difficult to drain the node safely with this approach.\nI was lucky in that I experienced this issue while cordoning our testing environment, so we didn't experience a production outage. I think what I'm going to try is the following... it's going to be time-consuming but I think it's the safest bet. Double the number of nodes in the server group to bring online the new nodes first Place a custom taint on each \"legacy\" node As you said, executes a first, so that's a no-go. This is the time-consuming part... for each \"legacy\" node I am going to each pod, and let my PDBs take care of maintaining minimum number of pods. I'm hopeful to try this approach this afternoon, I'll comment back here if I do!\nSame as myself! Pity neither of us have a story for Since my comment this morning I've been trying a few approaches, I've ended up with the following: Create a new node pool of equal size in GKE with the upgraded kube version Set the current node pool to not be able to grow beyond its current size - if your using regional nodes this can be annoying, let it bring up new nodes anyways and just have them taint as NoSchedule straight away Taint all old node pool nodes with NoSchedule Go through each old node pool node executing a taint of NoExecute which will cause the node to drain and go onto available nodes i.e. your newly provisioned node pool Ensure your nginx ingress is deployed with tolerances:\nHi All, I'm at a bit of a loss on this one. I've started using custom taints to drain the nodes without getting them removed from the load balancer. My application can hold a HTTP connection open for up to 600 seconds. I'm using nginx as a daemonset, have it configured with a keepalive timeout of 620s, and have . I drain the application first using a taint on each node. I have nginx on each node tolerate this taint and remove it separately at a later point using a different taint. I'm waiting 660s between drains of each node for each taint to ensure nothing is dropped - this all works fine as expected. However, once I kill the node pool or do a kubectl drain after the node is empty of all my stuff plus nginx ingress I start seeing connection issues. I'm a bit lost where to go at debugging them, best guess is something to do with kube-proxy. Can anyone shed some light on this? is there anything on the Google Layer 4 load balancer I'm missing here? Is there some way to correctly drain kube-proxy (or does that even make sense?)\nThis doesn't address the network issues you're seeing, but for those trying to \"roll their own cordon\" using taints: Using a NoExecute taint to \"drain\" the node won't be orderly and won't respect any PDBs. You want to replicate the logic from which tries to do evictions of every pod in a loop. (I realize this is a burdensome idea, but it is also the safest idea.)\nTo add on to ,: One approach that does not use node taints to force migrate, is to change for the requirements of your app. A change in node selector (e.g. based on instance type), along with a PDB (e.g. 66% / out of 3 replicas availability) should migrate you over fine. This worked out better for me than to drain or add taints to nodes to force push apps out. After the migration completes, you should be able to the nodes, and delete the node pool. FWIW, I too saw connection issues during load testing of a migration, and at least on my setup it was due to using an AWS NLB for the NGINX Ingress Controller. Switching to a classic ELB instead worked, as the NGINX LB Service seems to hit some sort of by my accounts when using NLBs. self-plug: I happened to recently cover a similar topic in a that you may find relevant. Hope it helps\nFor better or worse, for in my case pods that are behind an ingress resource are using node affinity to preferably not schedule them onto the same node. Originally this was done to ensure traffic is balanced across the different nginx ingresses. It has had a positive affect in this case too. Very interesting! Thank you for sharing, I'm not sure that swapping out the Layer4 GCE load balancer is an option here, but you have me thinking. I'm wondering if the TCP keepalive between the LB and nginx is potentially keeping connections open. , and suggest that there may be a 600 second TCP keepalive going on. Will retest doing a sleep of 620s after nginx removal and before node drain.\nHas anyone come up with a strategy to safely move workloads between two pools? Reading this bug, it seems there's no way to reliably do this without causing an outage if you follow the recommendation of + . In my case I'm not using an nginx ingress but services with type . From what I can see, adding to a node causes the Service Controller to consider the node not ready and thus removes it from a target group. This would mean that between and there is no way to add a taint to the node that will leave it in the LB until it's drained. Using doesn't work in this case since the thing I'm draining might just come back and I'm not sure if the service controller will still consider the node ready in that case, but I believe it does. The one strategy I can think of is using labels on the node and an anti-affinity on the pod. You'd then add the label to a node that would push the pod away and then delete the pod. As long as the anti affinity uses this should be safe to do in preparation of a migration. Once the pods have been deleted you could then the nodes.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/lifecycle frozen /remove-lifecycle stale /remove-lifecycle rotten\n- As I mentioned in I set up a script that first taints all nodes in the old node pool and then cordon+taint one by one node. If your ingress services are redundant this should work well.\nIs there a plan for cordon behavior to improve on GCP? We have relied on this behavior for years on our AWS clusters and it was very surprising to find out.\nThis is a bug with service load balancers - not something we should document. We defined a label in that cluster admins can use - Unschedulable has nothing to do with SLB inclusion. which is beta and on by default in 1.19.\nDisagree. Unschedulable means \"no new workloads\". It has nothing to do with existing workloads, that is why we have drain (but we should have something in kube that communicates nodes are draining). I have opened to remove this check - we have an alternate approach now and the old logic is wrong (the comment in the code even says \"this is for masters!\" and that's wrong now too as part of the kep).\nA node status condition?\nPossibly, or a taint, or a label. Something concrete.\n/milestone v1.19"}
{"_id":"q-en-kubernetes-757ffe38af6bfdd71e6f1e75d3560a3f5382542c1abe2fb0fcd35a36bf5c5dd1","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: after granting sp access to azure ACR , pull image from ACR would fail, and after wait about 15-30min, pull image would success. Root cause is that needs to be refreshed when doing , otherwise it will always return empty registry list. Related issue: What you expected to happen: How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): v1.7 - v1.11 Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: /sig azure /assign I will implement that function.\nThanks -- can we add a milestone label?\n+1. We should fix this within v1.12 release and cherry pick to old releases. /priority critical-urgent /milestone v1.12\n/status approved-for-milestone\n[MILESTONENOTIFIER] Milestone Issue: Up-to-date for process <details open<summaryIssue Labels</summary- : Issue will be escalated to these SIGs if needed. : Never automatically move issue out of a release milestone; continually escalate to contributor and SIG through all available channels. : Fixes a bug discovered during the current release. </details<details<summaryHelp</summary<ul<li<a href=\"\"Additional instructions</a</li<li<a href=\"\"Commands for setting labels</a</li</ul</details\nI'm not sure if this is related, but I'm seeing nodes (after they've been working fine for a few days) randomly start to fail pulling images with the error . Edit: I see this as the culprit: , after which all image pulls start to fail.\nthis is not related, original issue is that after granting sp access to azure acr, pull image still fails with . After my repro and investigation(really spent a lot of time...), I found the issue lies here, would return empty even after granting sp access to azure acr, and after around 15min, it returns correct loginServer list: Looks like it's a SDK or acr list with specific sp issue, workaround is wait around 15min or more time, it could recover by itself. Paste my investigation logs here, I grant sp access to acr in around , you could see it returns empty registry list first and then after a few more minutes, it returns the correct list with loginServer:\nBTW, in v1.12, acr list() uses new API, I will check whether in v1.12 would have same issue. Acutally I have already implemented LazyProvide func(check here: ), while it won't fix the original issue.\nUpdate: After a few more tries, I found it's related to this code snippet, before , we need to generate a new ServicePrincipalToken every time: I will move this code snippet into Provide() func, and send out a PR! The failure recover time is within 2min now."}
{"_id":"q-en-kubernetes-fc7c8e4dca431b05eb10616ff2a250bffbca5716787380e263be4e54f6be7096","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind feature What happened: I need to know to which Pod(s) a PVC is being mounted to, without having to go through each Pod one by one checking the Volume mounts and the PersistentVolumeClaim. The kubectl describe pvc and get pvc, only tell you to which Persistent Volume the PVC is bound to, but not if is being used by a Pod(s) What you expected to happen: Add a new field in the describe or get that lists the pods that are using the PVC as a volume mount How to reproduce it (as minimally and precisely as possible): Have a PV/PVC bound and create a pod using a volume mount Anything else we need to know?: Not sure if this is only for Azure Disk, which is the Storage Class I am using. Environment: Kubernetes version (use ): 7 Cloud provider or hardware configuration: Azure OS (e.g. from /etc/os-release): Ubuntu 16.04 Kernel (e.g. ): 4 Install tools: Azure Kubernetes Services Others: -n/a\n/sig storage\n/sig cli\nI think the kubectl describe can just construct this on the fly. Another angle is should we add this information to PVC object itself\nThat will be great, as right now the only way to correlate a PVC to a Pod is via a \"describe\" on the Pod.\nI can work on a PR for this!\nGitHub didn't allow me to assign the following users: clandry94. Note that only and repo collaborators can be assigned. <detailsIn response to : I can work on this! Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n+1 for the feature. cc\nHow to find \"what pods are using particular PVC through command line\"\n+1 for the feature.\nthis feature will be in Kubernetes 1.12\nok kubernetes 1.12 released, how to? which command?\nCan someone point to the documentation or paste command to get the pods which have the PVC in use\nsee the associated PR to see an example usage of this The line in should show it.\nWorkaround for the meantime:\nThanks , this workaround is great for those people who still use old kubernetes version. However it will meet an error if the pod have no volumes, so I improves it by adding a to check whether the volumes exists or not.\npython script to list pods with pvc in namespace(s): Is a part of a cli to set pods with velero/restic annotations. Hope it helps someone :)\nFor posterity... There is a Used By: field in the output of describe.\nclient-go didn't give this api... ,PodPVCIndex can solve the problem\nHi thanks worked for me ;) but I want to make it reverse my workarounds with adding after didn't worked do you have any ideas\n$ kubectl describe pvc -A Used\" Name: ocs-deviceset-lso-volumeset-0-data-1b45dn Namespace: openshift-storage StorageClass: lso-volumeset Used By: rook-ceph-osd-1--8fwpd Also useful ^\nthe reason it didn't work is because each return does, in point of fact, have a key defined for claimName. That key is just set to null value. So, what you'd actually want to do is test for nulls. As nice as answer is, it's a PITA to parse unstructured human-readable command output, as opposed to command output in a markup language that well known shell parsers like jq and yq can work with. Towards that end, because I had use of this, I came up with the following: That simple not null test at the end plucks out all your nulls. Obviously invert the last conditional if you want all the nulls, but you'll get all sorts of temp things that way that you probably won't care about."}
{"_id":"q-en-kubernetes-392a747797b3e1c2752bc049b84086988d167ceb650865e5c35034d88d820b6c","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug /sig network /area kube-proxy /area ipvs /assign I'll send a fixing pull request tomorrow. What happened: On centos 7.2 with ipset 6.29, kube-proxy in ipvs mode failed to create ipset sets. After some dig, it turns out to be problem of ipset on centos/rh. See I tried ipset 6.29 on ubuntu, it's OK. But when I tried ipset 6.29 on centos, I got errors: What you expected to happen: Since the problematic ipset versions have very large installation base, we could remove ipset commet, or remove ipset commet with problematic ipset version. The problematic ipset also says it support comment by --help. The problem could be detected by a test run of \"ipset create foo hash:ip comment\". How to reproduce it (as minimally and precisely as possible): Create some service in kubernetes, run kube-proxy with ipvs mode on Centos 7.x node. Anything else we need to know?: Environment: Kubernetes version (use ): commit (actually all commits after ) Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): CentOS Linux release 7.2.1511 (Core) Kernel (e.g. ): Linux VM34centos 3.10.0-514.26.2.el7.x8664 SMP Tue Jul 4 15:04:05 UTC 2017 x8664 x8664 x8664 GNU/Linux Install tools: Others: ipset v6.29, protocol version: 6\ncould you test it before commit , that's ok?\n, yes, I did. It's OK. But I think the we don't need to revert the iptables comments.\nif ipset package not installed, will set module or match-set option in rule be weird and now seems not."}
{"_id":"q-en-kubernetes-1d1abdd74cdb3621d2817b67140cd372d542d8b5422c1bfbe8473adbaa86f3f5","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: What happened: After merged, the leader election is calling too many api calls. Previously it calls with a 2s interval by default. Now it calls with 0 interval. Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\n/sig api-machinery\n/assign\n/kind bug"}
{"_id":"q-en-kubernetes-7b1b5aac033196141508f40a7a5ea6449e76d4ca03418f562f45bd84bbb3caae","text":"What happened: Data race in tests in an unrelated PR Logs https://k8s- /kind bug /area platform/vsphere /area cloudprovider /sig cloud-provider\n/kind flake\nAnother time it flaked in an unrelated PR https://k8s-\nbringing this to your attention as a vsphere dev.\n/assign\nLooks like the race was in Feb 2017 here: But that code path wasn't covered by go test until vcsim support was here: I'll work on a fix."}
{"_id":"q-en-kubernetes-6d88c3f57c9a7fe083bc7fa6547d93018ed161eee205c0f2188edca1d2af8f07","text":"Is this a BUG REPORT or FEATURE REQUEST?: What happened: The output of my kubeadm: When I pull the image with docker pull, I get such output: But other images are successfully pulled as we can see from previous log. Why?\n/sig cluster-lifecycle cc\nUgh can you fix this please? I missed this in the PR that just got merged, should be without the arch suffix as this is a manifest list\nGetting on with this ASAP.\nfyi ,"}
{"_id":"q-en-kubernetes-1c2f27cc7995dc0195a6d562a190312b043b529fecef2bd3bd3317c48a4bd434","text":"Implement proposal for Azure We should implement interface in Azure cloud provider. returns true if the instance is in stopped state to detach volumes. But if instance is deleted we should not return error. Related features: Related proposal: Related PRs: /kind feature /sig azure /cc\n- can you assign me on this one ? I already have a PR in flight.\nThat's great! /assign just loop in for this RP review, thanks.\nGitHub didn't allow me to assign the following users: yastij. Note that only and repo collaborators can be assigned. <detailsIn response to : That's great! /assign just loop in for this RP review, thanks. Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\no, you are not a k8s member yet, anyway, you could send out the PR for this feature, related to this issue, thanks.\nThanks, looking forward to your PR."}
{"_id":"q-en-kubernetes-21d79d4b86405dc56238b9a9ac0bd543c859fd67cfeabcc17cfe386742e18a8d","text":"/kind bug What happened: When creating a node port application (in this case the example guestbook) on a host server that is dual stacked and has AAAA and A DNS entries, the node port on the v6 address is open to kube-proxy and hangs. What you expected to happen: The v6 address would refuse the connection and the client would fall back onto the v4 address which would connect correctly. How to reproduce it (as minimally and precisely as possible): kubeadm init with the above, and join in some workers. Then install the guestbook application Obtain the node port with and telnet to one of the servers on that port from a dual stacked client. The client will connect and refuse to respond to 'GET /' telnet directly to the IPv4 address of the server on the port and all works as expected. Anything else we need to know?: kube-proxy is noted as listening on the port. Environment: Kubernetes version (use ): ubuntu kubectl version Client Version: version.Info{Major:\"1\", Minor:\"11\", GitVersion:\"v1.11.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2018-07-17T18:53:20Z\", GoVersion:\"go1.10.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"11\", GitVersion:\"v1.11.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2018-07-17T18:43:26Z\", GoVersion:\"go1.10.3\", Compiler:\"gc\", Platform:\"linux/amd64\"} Cloud provider or hardware configuration: Brightbox OS (e.g. from /etc/os-release): Ubuntu 18.04.1 LTS Kernel (e.g. ): Linux srv-jidtj 4.15.0-29-generic -Ubuntu SMP Tue Jul 17 15:39:52 UTC 2018 x8664 x8664 x86_64 GNU/Linux Install tools: kubeadm 1.11 Others:\n/sig network\nLooks like kube-proxy in ipvs mode listens on the nodeport ports to ensure they're not in use by something else. It just binds and listens, it doesn't serve requests. It binds on both ipv4 and ipv6 but never expects to receive connections because the iptables/ipvs rules intercept them, but on ipv4 clusters there is nothing stopping ipv6 connections coming in. The comment there does admit that only a bind is necessary to do this, but doing a full listen just made e2e tests easier. /cc\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-testing, kubernetes/test-infra and/or . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/reopen This now causes conformance failure on e2e tests due to the promotion of the Nodeport test in\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nStill an issue on 1.16 when running in standard IPv4 mode\nAs I understand this thread, I have the same problem. I thought that it is a node configuration or some kubernetes flag that I didn't set but it seems it ain't. I ask already on SO: Environment Kubernetes version (use kubectl version): Client Version: version.Info{Major:\"1\", Minor:\"16\", GitVersion:\"v1.16.0\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2019-09-18T14:36:53Z\", GoVersion:\"go1.12.9\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"16\", GitVersion:\"v1.16.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2019-10-02T16:51:36Z\", GoVersion:\"go1.12.10\", Compiler:\"gc\", Platform:\"linux/amd64\"} Cloud provider or hardware configuration: On premise, on virtual machines using hyper-v OS (e.g. from /etc/os-release): NAME=\"Ubuntu\" VERSION=\"18.04.3 LTS (Bionic Beaver)\" Kernel (e.g. uname -a): Linux kubernetes 5.0.0-29-generic # 31~18.04.1-Ubuntu SMP Thu Sep 12 18:29:21 UTC 2019 x8664 x8664 x86_64 GNU/Linux Install tools: kubeadm version: &version.Info{Major:\"1\", Minor:\"16\", GitVersion:\"v1.16.0\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2019-09-18T14:34:01Z\", GoVersion:\"go1.12.9\", Compiler:\"gc\", Platform:\"linux/amd64\"}\n/lifecycle frozen\n/assign bowei\nIf this issue has been triaged, please comment . If you aren't able to handle this issue, consider unassigning yourself and/or adding the label. I am a bot run by vllry. ‍\nLooking through the comments, this looks like a bug that could have a targeted fix. /remove-triage unresolved /good-first-issue /help-wanted\nThis request has been marked as suitable for new contributors. Please ensure the request meets the requirements listed . If this request no longer meets these requirements, the label can be removed by commenting with the command. <detailsIn response to : Looking through the comments, this looks like a bug that could have a targeted fix. /good-first-issue /help-wanted </details\nI will attempt to solve this issue.\nany success or progress? If you left this issue I would like to give it a go as a first time contributor. Thanks, Yuval\nI'm still working on this as my first issue, and planning to produce a solution by 20th.\nI wasn't able to tackle the issue and will not continue trying, feel free to take it.\n/assign yuvalsteuer\nStill, open -- anyone actively working on this? I would like to attempt if not. :) /assign rkc007\nOK, again still open, doesn't look like anyone is working on this. I'd like to try it. The fix that occurs to me is simply to only listen to IPv6 addresses if the cluster is IPv6. If the e2e tests fail in this case (from looking at I don't think they will) then maybe they need to be updated. Does that make sense? What I'm not sure about is if we should do nothing with IPv6 ports or bind (to claim the port) but not listen.\n/assign michaelbeaumont"}
{"_id":"q-en-kubernetes-57b5e16394867179672bc71a3976bca45150df747198e4d516bcbb5c497411bc","text":"/kind bug Hello, actually this is a bug about not about k8s. I use to observe request headers and found if the request have multiple same headers echoserver will give a wrong result: result: Expected result:\n/sig testing\nany chance you know some lua?\nI never used lua before, but I'm willing to try to fix this issue."}
{"_id":"q-en-kubernetes-2d583728f7b7db25a152cbc0daa919f3cf4a41b0a2c72387b0cf92c2183ad4d1","text":"Is this a BUG REPORT or FEATURE REQUEST?: What happened: Azure will not update the DNS label of the public IP, when update the service only changing the annotation of the DNS label from empty. annotations: -annotations: \"to-update\" What you expected to happen: It should update the dns label of the assigned public IP How to reproduce it (as minimally and precisely as possible): It probably caused when applying function 'getPublicIPAddress' in This function returns a bool 'existsPip' to judge the public ip existence. When only the annotations changed, the existsPip still returns true. Then, the reference in line 553 and line 446 step into this bug. pip, existsPip, err := az.getPublicIPAddress(pipResourceGroup, pipName) if err != nil { return false, err } if !existsPip { return true, nil } Firstly, in line 554, it will make function 'isFrontendIPChanged' return false, and then will not run 'ensurePublicIP' in line 691. Secondly, even if 'ensurePublicIP' in line 691 runs, it will detect the public ip exists and make no update to the DNS label in line 446_. Environment: Kubernetes version (use ): Cloud provider or hardware configuration: Azure\n/sig azure\n/assign\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\n-- what's the status on this?\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-testing, kubernetes/test-infra and/or . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/open I will take a look at this\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-testing, kubernetes/test-infra and/or . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nStill not fixed. /reopen /kind bug /priority important-soon /milestone v1.16\nReopened this issue. <detailsIn response to : Still not fixed. /kind bug /priority important-soon /milestone v1.16 Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig cloud-provider\ni'm working on it.\n/assign"}
{"_id":"q-en-kubernetes-a38798d209f0702d464d1d4ca1e4d0e5108c803367931f11e6980bbace35c0ee","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: PR() fixed pull image error from a cross-subscription azure container registry by service principal, while for MSI, this issue is not fixed. Need to find a way how to fix this issue for MSI, related code is here: Here the difficulty is could not list all ACR since it's based on one subscriptionID: So if we could not get in a different subscription, how could we set the . What you expected to happen: How to reproduce it (as minimally and precisely as possible): error msg would be like following when pull a repository from an ACR in another subscription: Anything else we need to know?: Environment: Kubernetes version (use ): v.9 - v1.12 Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: /sig azure /assign\nBy further investigation, I think we should implement LazyProvider for azure acrProvider: Related code:\nUpdate: It's hard to fix this since in the beginning we cannot list ACR in another subscription, will always list acr under current subscription, and without target loginServer address of ACR, we cannot get : And in func, we cannot get target repo, it's not passed on\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\n-- is this still an issue? If not, let's it out.\nit's still an issue, and currently it should be low priority, let's keep it\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-testing, kubernetes/test-infra and/or . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/remove-lifecycle rotten /reopen\nReopened this issue. <detailsIn response to : /remove-lifecycle rotten /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-3000984cb75e45bc5b032966c893f1be3072f7c8abd7255b07d061a11a831fe1","text":"/kind bug /sig release Such as:\nCool issue. /assign\nI see the code in was pushed by you. I don't understand what Intercept() does. Do you think there is an Infinite recursive call?\nChaos client isn’t used anywhere, and we need to remove it soon.\nOk, I can help remove it.\nBut there are also one place using it. How to deal with it? Delete it directly? Or something we need to do?\nHas never been anything other than experimental and was not used in practice. We can remove it and it’s flag from kubelet as it was experimental. are we deprecating experimental flags for one release or can we just remove?"}
{"_id":"q-en-kubernetes-844205b3dd953936ebb1454fae42d0bf5c5ef743dd58fd900c2a13921a65ebfc","text":"We fixed invalid device path in bunch of places via - but call still could use invalid device path. That check can sometimes use invalid path and hence can cause UnmountDevice operation to fails even though device has already been unmounted. /sig storage /sig openstack /assign\nActually this does not affect version 1.10 onwards. Not sure if 1.9 is still supported. I think I will close this for now. /close\nI just noticed that 1.9 is still supported and hence it is probably worth fixing there for now.\nHow come the check can \"sometimes\" use invalid device path?\nSo on Cinder for example - call would return something like as attach device path. But if you see the linked PR(), will do a and use that device path for global mount. In general it is always a good idea to use device path returned by as single source of truth. This applies to something like EBS with NVMe too where device is discovered via call (but nvme disks aren't directly affected by this bug because returned device path on EBS is bogus).\nSo on EBS with NVMe for example - check will always fail (but still return success) because it will use something like as device path which does not actually exist on the node.\nYou said the issue only happens \"sometimes\". Can you clarify why it only happens occasionally?\ndoesn't the linked PR make it clear? detection based on the drive serial number (aka It's cinder ID) on the kubelet itself. As for why - Cinder sometimes returns wrong device path on call, I don't know. But it happens and we have worked around it.\nFixed by"}
{"_id":"q-en-kubernetes-437d6c360673c1e8f00e42bdafdadaee2e596ba27a49b7d01b3a77e242c81c64","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: What happened: CSI Plugin calls to get node ID information (see ). If that call is not implemented by driver, registration with kubelet will fail. What you expected to happen: Per CSI Spec, the deprecated call should be supported until CSI 1.0 is released (see ). Therefore if a by Kubernetes fails (because it is not implemented by the driver), we should fall back and try call instead. How to reproduce it (as minimally and precisely as possible): Use any CSI driver that implements but not -- says OpenSDS driver is currently hitting this, and may be used for repro. Anything else we need to know?: When CSI moves to 1.0 we can drop this fall back logic. If a fix is available within , we can include it, but 1.12 release should NOT block on this. /sig storage Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others:\nplease verify this is the issue you hit. Specifically see if you can find in your kubelet logs.\nCC who volunteered to take a look\nI don't see \"error during CSI NodeGetInfo\" in my kubelet logs. I don't see glog.Infof(log(\"Register new plugin with name: %s at endpoint: %s\", pluginName, endpoint)) so it seems that RegisterPlugin is not called. I do see a message: \"Plugin Watcher Start at /var/lib/kubelet/plugins\". I don't know why RegisterPlugin is not called. I'll continue to check.\nHere's an update. I implemented NodeGetInfo. I followed documentation about the Kubelet Plugin Watcher: https://kubernetes- to get my plugin registered. After that I don’t see the “driver name csi-opensdsplugin not found in the list of registred CSI drivers” any more. I see that now RegisterPlugin gets called and I see the following message in I0914 14:44:50. ] Register new plugin with name: csi-opensdsplugin at endpoint: /var/lib/kubelet/plugins/csi- So this confirmed that missing NodeGetInfo is the problem. I still have problems but I think that’s a config issue on my side. I the following as cluster role rule for attacher and node: apiGroups: [\"\"] resources: [\"csidrivers\"] verbs: [\"get\", \"list\", \"watch\", \"update\"] However I’m still getting the following errors in the logs: I0914 16:12:31. ] Listing and watching v1alpha1.CSIDriver from E0914 16:12:31. ] Failed to list v1alpha1.CSIDriver: is forbidden: User \"system:serviceaccount:kube-system:attachdetach-controller\" cannot list resource \"csidrivers\" in API group \"\" at the cluster scope kube-controller- I0914 16:11:51. ] Listing and watching v1alpha1.CSIDriver from E0914 16:11:51. ] Failed to list v1alpha1.CSIDriver: is forbidden: User \"system:node:127.0.0.1\" cannot list resource \"csidrivers\" in API group \"\" at the cluster scope\nDid more investigation on this. Here's an update. I set the following feature gate to true on my local cluster: alpha: CSICRDAutoInstall CSISkipAttach CSIPodInfo I got some new errors in and kube-controller- I0916 07:22:06. ] Listing and watching v1alpha1.CSIDriver from E0916 07:22:06. ] Failed to list v1alpha1.CSIDriver: the server could not find the requested resource (get ) I0916 07:25:24. ] Listing and watching v1alpha1.CSIDriver from E0916 07:25:24. ] Failed to list v1alpha1.CSIDriver: the server could not find the requested resource (get ) In addition, I found the following in kube-controller- E0916 07:22:01. ] failed to create CSIDrivers CRD: &v1beta1.CustomResourceDefinition{TypeMeta:v1.TypeMeta{Kind:\"\", APIVersion:\"\"}, ObjectMeta:v1.ObjectMeta{Name:\"\", GenerateName:\"\", Namespace:\"\", SelfLink:\"\", UID:\"\", ResourceVersion:\"\", Generation:0, CreationTimestamp:v1.Time{Time:time.Time{wall:0x0, ext:0, loc:(time.Location)(nil)}}, DeletionTimestamp:(v1.Time)(nil), DeletionGracePeriodSeconds:(int64)(nil), Labels:map[string]string(nil), Annotations:map[string]string(nil), OwnerReferences:[]v1.OwnerReference(nil), Initializers:(v1.Initializers)(nil), Finalizers:[]string(nil), ClusterName:\"\"}, Spec:v1beta1.CustomResourceDefinitionSpec{Group:\"\", Version:\"\", Names:v1beta1.CustomResourceDefinitionNames{Plural:\"\", Singular:\"\", ShortNames:[]string(nil), Kind:\"\", ListKind:\"\", Categories:[]string(nil)}, Scope:\"\", Validation:(v1beta1.CustomResourceValidation)(nil), Subresources:(v1beta1.CustomResourceSubresources)(nil), Versions:[]v1beta1.CustomResourceDefinitionVersion(nil), AdditionalPrinterColumns:[]v1beta1.CustomResourceColumnDefinition(nil)}, Status:v1beta1.CustomResourceDefinitionStatus{Conditions:[]v1beta1.CustomResourceDefinitionCondition(nil), AcceptedNames:v1beta1.CustomResourceDefinitionNames{Plural:\"\", Singular:\"\", ShortNames:[]string(nil), Kind:\"\", ListKind:\"\", Categories:[]string(nil)}, StoredVersions:[]string(nil)}}, err: &errors.StatusError{ErrStatus:v1.Status{TypeMeta:v1.TypeMeta{Kind:\"\", APIVersion:\"\"}, ListMeta:v1.ListMeta{SelfLink:\"\", ResourceVersion:\"\", Continue:\"\"}, Status:\"Failure\", Message:\" is forbidden: User \"system:serviceaccount:kube-system:attachdetach-controller\" cannot create resource \"customresourcedefinitions\" in API group \"\" at the cluster scope\", Reason:\"Forbidden\", Details:(*v1.StatusDetails)(0xc4212af560), Code:403}} So I made this change to add RBAC rule for customresourcedefinitions: After that, my CSI plugin is functional. I can create/attach/mount volumes successfully now.\n/cc\n(As we discussed offline) the latest code in master no longer has CSICRDAutoInstall, CSISkipAttach, CSIPodInfo flags. New CSI features are now flag gated under CSINodeInfo and CSIDriverRegistry flags (both alpha and disabled by default). KubeletPluginsWatcher must be enabled as well, but that's in beta now so it's enabled by default. Since this PR merged: , CRDs must be installed manually, and you can find an example of it here: (note they have to be created with --validation=false kubectl flag for the time being). This PR has not been cherry-picked to 1.12.0-rc.0 yet.\nThanks Tested latest code. If NodeGetInfo is implemented, we just need to make necessary changes when deploying driver-registrar sidecar container so KubeletPluginsWatcher can work. NodeGetInfo is needed for kubelet to register CSI plugins. Also tested with and without \"CSINodeInfo=true,CSIDriverRegistry=true\" and manually installed CRDs using when they are enabled.\nI have the same problem when deploying my CSI plugin based on CSI v0.2.0 on Kubernetes v.1.12.0. Daemonset pods of CSI plugin cannot launch successfully. Kubelet logs:\nConsidering Kubernetes v1.12 is already out, and we planned to deprecate the old call in CSI 1.0/Kubernetes 1.13 anyway, we'll pull of the bandaid and deprecate it now. If you are running Kubernetes 1.13, make sure your CSI driver implements the .\nlooks like you might be running in to a different issue: Kubernetes v1.13 also enables Kubelet device plugin registration by default. Before upgrading to v1.13, ensure the CSI sidecar container for your CSI driver is configured to handle plugin registration (set the parameter on to expose a new unix domain socket to handle Kubelet Plugin Registration). From\nHi Because goes Beta and is enabled by default in Kubernetes v1.12. After disabling this feater gate, I can run CSI plugin based on CSI spec v0.2.0 in Kubernetes v1.12.2. When Kubelet enable , Kubelet would call gRPC call to register CSI plugin. gRPC call was first proposed at CSI spec v0.3.0. Therefore, user should disable feature gate when running CSI plugin based on CSI spec v0.2.0 in Kubernetes v1.12.\nI have implement the NodeGetInfo call. But it still has this issue. My kubernetes version is 1.14.3."}
{"_id":"q-en-kubernetes-44de15891075df631b07f50c082635d98a7de9f9bd002018dc655f215347ea12","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: After running (on master branch), a new file named left as a surprise. and the file is not matched by git ignore so it's tracked by git which is really disturbing. I tried to find out where it's created but failed. Any one has a context w/ it? Environment: OS (e.g. from /etc/os-release): Darwin Kernel Version 17.6.0: Tue May 8 15:22:16 PDT 2018; root:xnu-4570.61.1~1/RELEASEX8664 x86_64\n/sig instrumentation\nthe only piece of code in local-up-cluster that touches the file is\ncan't reproduce this. Two questions: Can you see this everytime? What is the SHA of the commit on master when you tried this? Can check with that too.\nfinally i realize that the cause is command is built receiving different flags between Linux and OSX. on Linux, you'll find that for linux sed the is explained as: '--in-place[=SUFFIX]' This option specifies that files are to be edited in-place. GNU 'sed' does this by creating a temporary file and sending output to this file rather than to the standard output.(1). so the is optional. but for OSX, it is: Edit files in-place, saving backups with the specified extension. If a zero-length extension is given, no backup will be saved. It is not recommended to give a zero- length extension when in-place editing files, as you risk corruption or partial content in situations where disk space is exhausted, etc. the difference is that on OSX, flag is requiring a variable , so when we're using on OSX in the script: the which is supposed to be received as a flag will be a suffix for the backup file. then the magic happened.will raise a fix for this. /assign\nlinking the difference between and"}
{"_id":"q-en-kubernetes-de5f2c37dc83b8e3c71ab19779fa4067a89caa68b2b68175821af26181b15923","text":"/kind bug What happened: After upgrading 1.11.1 -1.12.0 , when a node gets terminated in AWS, the controller-manager doesn't delete the node in the API. This has lead to several side effect for us : pods don't get GC Prometheus alerts firing because 100 of nodes are in the NotReady state for a long period What you expected to happen: kube-controller-manager deletes the node when it stops reporting, after checking with the cloud provider. How to reproduce it (as minimally and precisely as possible): Looking at the code, I believe this is what introduced the behaviour change : - The node gets tainted with because it's in a Terminated state. But it doesn't get removed because it still \"exists\" in the cloud provider. The behaviour is probably correct for a shutdown node, but in AWS a terminated node can't be restarted so there's no good reason to keep it there ? Environment: Kubernetes version : 1.2.0 Cloud provider or hardware configuration: AWS OS (e.g. from /etc/os-release): CoreOS (sorry for the all the sigs, not sure who it belongs to) /sig aws /sig cloud-provider /sig cluster-lifecycle\nas I see it should delete node if instance is in terminated state have you tested that change in that row? I do not see what could be the issue there. Or could it be that the is not in place for all binaries in 1.12.0?\nI might be wrong, but I think that part of the code never gets executed once the node is tainted :\ngood finding. So we should remove terminated state from this row. Then it does not add shutdown taint if node is terminated - instead it will delete node. also this should return false.. because if it returns true node is not deleted\nI had the same issue. After I scale down my worker node instance number from 3 to 1. One node is stuck in status and never got deleted. However, one interesting thing is one of the other node is removed successfully. do you see similar behavior too?\n- can we make sure this gets cherry picked for 1.12 releases ? I think we already missed the cut for 1.12.2"}
{"_id":"q-en-kubernetes-3c94567dd7c22153dcd3c3d3fd52ec6b7d816a73e9f5b4c6749f041044699dee","text":"Is this a BUG REPORT or FEATURE REQUEST?: What happened: After upgrade of Kubernetes 1.11.3 to 1.12.1 via kubeadm in bare-metal setup I see these errors in kubelet logs: What you expected to happen: Since this is a bare-metal setup I expect no messages concerning Azure cloud How to reproduce it (as minimally and precisely as possible): Do an upgrade from 1.11.3 to 1.12.1 Anything else we need to know?: kubelet-1.12.1- was downloaded from Environment: Kubernetes version (use ): Client Version: version.Info{Major:\"1\", Minor:\"12\", GitVersion:\"v1.12.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2018-10-05T16:46:06Z\", GoVersion:\"go1.10.4\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"12\", GitVersion:\"v1.12.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2018-10-05T16:36:14Z\", GoVersion:\"go1.10.4\", Compiler:\"gc\", Platform:\"linux/amd64\"} Cloud provider or hardware configuration: bare-metal OS (e.g. from /etc/os-release): Oracle Linux Server 7.4 Kernel (e.g. ): 62 Install tools: Others:\nLines from kubelet log which say that no cloud provider was specified (that is a bare-metal setup): ..\n/sig azure\nhave the same issue related to the azure.. also a bare-metal cluster.. these messages are flooding my centralized logging elasticsearch (though they aren't really problematic more than being an annoyance). i also upgraded from 1.11.3 to 1.12.1 on a 3-master HA set up. i also noted (perhaps worthy of a separate issue) that the two of my three control-planes that are showing the azure error also previously had been having cert authority errors from the api-server container, when having used kubeadm upgrade.. (on the second two control plane systems, with the issue) (and on the first control without the issue) (and i needed to kubeadm reset and then kubeadm init all over again with 1.12.1 on the 2nd two control-plane nodes).. the first control-plane node that didn't need to do a reset/init didn't have the azure errors (nor the cert errors). so i kind of suspect it might not be an kubeadm upgrade issue but a general issue going from 1.11.3 to 1.12.1 ?\nSame problems +1\nSame problem on brand new bm 1.12.1 cluster built with kubeadm and latest 1alpha3 config based off of customized kubeadm config print-default example\n/assign\nSame problems +1\nOct 9 11:11:32 localhost kubelet: E1009 11:11:32. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:11:42 localhost kubelet: E1009 11:11:42. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:11:52 localhost kubelet: E1009 11:11:52. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:12:02 localhost kubelet: E1009 11:12:02. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:12:12 localhost kubelet: E1009 11:12:12. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:12:23 localhost kubelet: E1009 11:12:23. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:12:33 localhost kubelet: E1009 11:12:33. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:12:43 localhost kubelet: E1009 11:12:43. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:12:53 localhost kubelet: E1009 11:12:53. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1 Oct 9 11:13:03 localhost kubelet: E1009 11:13:03. 3342 ] failed to get azure cloud in GetVolumeLimits, kubernetes-1\nhow often did you hit this error log: I am planning to make this log as , is that ok for you? Usually V(2) is the default info(see ) log for k8s on azure I am not sure what the default info log by kubeadmin installation.\nIt is logged every 10 seconds. I disagree. At least previous incarnation of GetVolumeLimits in version 1.11.3 : would have logged an error \"No cloudprovider present\". Which I didn't see in kubelet log. In fact, I didn't see any message in kubelet logs with 'azure' substring at all. I think it means that AzureDataDisk plugin was not instantiated at all in 1.11.3. So the question is: why azuredd is instantiated now in 1.12.1 in bare-metal setup (that is without cloud provider)? Do you mind checking this idea? (Besides your method of lowering message priority won't work as I said this is a bare-metal setup where kubelet systemd unit file has no --v option at all )\nThat's because feature reaches beta in v1.12, in v1.11.3, even the code is there, it won't be invoked since feature is still alpha in v1.11\nAnd in v1.12, aws, gce, and azure disk plugins are all loaded, take aws ebs for example, it will return error directly: And logging as V(4): I suppose V(4) logs won't display in logs. Let me make the same change as aws and gce.\nalso tries to address this error.\nThe same mistake, I also encountered NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME 172.16.16.13 Ready <none5m27s v1.12.1 172.16.16.13 <noneCentOS Linux 7 (Core) 4.18.12- docker://18.6.0 failed to get azure cloud in GetVolumeLimits, 172.16.16.13 failed to get azure cloud in GetVolumeLimits, 172.16.16.13 failed to get azure cloud in GetVolumeLimits, 172.16.16.13\nI have cherry picked the fix to v1.12\nI understand that the feature Node-specific Volume Limits has now reached beta in 1.12, but for those of us running Kubernetes on-premises, this log error is pretty useless. In fact, I would even want to disable this plugin alltogether, since it will never be used on-prem. Is there a way to stop the plugin from loading, or in the very least make sure the log does not appear? Right now, it is polluting the system log file quite a lot.\nI second that. That was the key message of my bug report. That is we must not even instantiate cloud-specific plugins if no cloud provider is configured.\nI'm having exactly the same problem, doing an installation from scratch on DigitalOcean\n+1\n+1\nHi I have the same issue. It would be really nice to know how to disable this error message that floods the logs for no reason.\nFor those using logstash it was sufficient to just temporarily add this to your syslog filter: if [message] =~ \"\" { drop { } }\nAfter a fresh installation of Raspbian and kubernetes, same issue is occurring in my 4 node cluster. Client Version: version.Info{Major:\"1\", Minor:\"12\", GitVersion:\"v1.12.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2018-10-05T16:46:06Z\", GoVersion:\"go1.10.4\", Compiler:\"gc\", Platform:\"linux/arm\"} Server Version: version.Info{Major:\"1\", Minor:\"12\", GitVersion:\"v1.12.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2018-10-05T16:36:14Z\", GoVersion:\"go1.10.4\", Compiler:\"gc\", Platform:\"linux/arm\"} flood of this: epsilon kubelet[2412]: E1024 12:25:58. 2412 ] failed to get azure cloud in GetVolumeLimits, epsilon\ndisable in controller plane should work: \"--feature-gates=AttachVolumeLimit=false\"\nI have tried \"--feature-gates=AttachVolumeLimit=false\" and it's not working. I still see messages kubelet[]: E1024 22:51:02. ] failed to get azure cloud in GetVolumeLimits,\nI have the same problem here.., ] failed to get azure cloud in GetVolumeLimits,\n+1 same case here, also tried with --feature-gates=AttachVolumeLimit=false\nWill this be disabled by default in the next release?\nIt's already fixed in v1.12.2 and v1.13.0. do you know how to disable in kubelet in v1.12? Would add in kubelet command arguments work?\nI haven't updated to 1.12.2 yet (still on 1.12.1) but just to confirm, disabling that feature works. For those that are also running on CentOS and/or RHEL, just to the following: Edit file on your master server (control plane), and add the following: If you already have extra arguments in there, just add this one at the end of the line. Then restart kubelet on your master server:\nYes that will work.\n--feature-gates=AttachVolumeLimit property doesn't exist in the documentation\nUpdate tho 1.12.2 resolve for me Em qui, 1 de nov de 2018 9:22 PM, Abhilash Pallerlamudi < escreveu:\nadd follow line to \"your kubelet.service\" file. like this: [root ~]# systemctl daemon-reload [root ~]# sudo systemctl restart kubelet then it's ok\nSo to get rid of a bug we should disable an entire feature? Says it is supported by Azure-Disk. Why isn't node reporting the limit? It is getting instance metadata okay as all my nodes are tagged with SKU. Being that SKU determines attach limit as well, I do not understand why this cannot be made working? I am on 1.14 in Azure on Ubuntu.\nhow do you deploy k8s on Azure? To get volume limit, kubelet on agent node needs azure identity, e.g. service principal, usually it's under file\nI deployed with kubespray. I have cloudconfig set as such: KUBELETCLOUDPROVIDER=\"--cloud-provider=azure --cloud-config=/etc/kubernetes/cloudconfig\" Service principle is set in cloudconfig file.\nI triple checked and location was misspelled on some of the nodes for the cloud provider. Thanks!\nthe get vm size list call depends on location, so now it works?\nYes it does. Thanks!"}
{"_id":"q-en-kubernetes-20a4ccf4adcc587e9b1d97fc6976c0accb49e7f495374adff17fffda9a16848e","text":"Is this a BUG REPORT or FEATURE REQUEST?: /kind bug What happened: e2e test \"[sig-network] Network [It] should set TCP CLOSEWAIT timeout\" requires the test target OS supports /proc/net/nfconntrack, but Ubuntu 16.04 doesn't support it. Then the test fails today like the following: What you expected to happen: The test succeeded or was skipped. How to reproduce it (as minimally and precisely as possible): Just kick Anything else we need to know?: The existence of /proc/net/nfconntrack depends on the Linux kernel config CONFIGNFCONNTRACKPROCFS, but that is disabled on Ubuntu 16.04. In addition, the test is skipped if the node number is less than 2. So I guess that is one of reasons this issue was not present until today. Environment: Kubernetes version (use ): v1.11.3 Cloud provider or hardware configuration: OpenStack OS (e.g. from /etc/os-release): Ubuntu 16.04.4 LTS Kernel (e.g. ): 4.4.0-134-generic Install tools: kubeadm Others:\n/sig network\nSome kube-proxy functionalities depends on conntrack, is it possible to run a conformance cluster without having conntrack enable?"}
{"_id":"q-en-kubernetes-00d9fc51e2d19fb7f62f6dafef99b2750dc74c15fbc41108be6037bea54208ac","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: What happened: the for a CRDs Established condition is always null What you expected to happen: it should have a timestamp How to reproduce it (as minimally and precisely as possible): create a CRD and then examine its yaml/json. For example: Anything else we need to know?: Environment: community master\n/assign\n/area custom-resource\n/sig api-machinery /area custom-resources\n/assign\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\n/remove-lifecycle rotten\nedit: actually, that issue is about kubectl display and clock skew. this issue is about the condition timestamp updating, which still requires correction\nClosing this issue. <detailsIn response to : duplicate of Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nHello I'm a bug triage shadow for the 1.16 release cycle and considering this issue is tagged for 1.16, I'd like to check its status. Code freeze is starting on August 29th (about 3 days from now), which means that this PR must merged by then. Do we still target this issue to be fixed for 1.16?\nyes, will take a look at it tomorrow /assign"}
{"_id":"q-en-kubernetes-1aa2b788ee8dcf8d8598c7b30272a10240375971a3f4c89bc972fae10098a876","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: What happened: Here's an example from an e2e test: What you expected to happen: leader election success logs to be at a higher log level.\nLeader election library is not a part of the scheduler. The scheduler is only a user of the library. It is located in client-go. It is currently . I am not sure if it is a good idea to reduce verbosity further. The log messages seem to be excessive in the scheduler logs when the scheduler is idle in our tests.\nI think this is a common problem for all components, not just for scheduler.\nIn that case, we should consider reducing its verbosity to 5 or 6.\nFor some reason I thought that this was something new but I went back to 1.9 tests and checked the logs and I also see it printed every 2 seconds. So maybe this is fine for test environments.\nThis is certainly not new, but I am still in favor of lowering it verbosity unless there is a good reason for having it in tests by default.\nI'm ok with changing the verbosity but the code now lives in the api-machinery land.\nI'd like to pick this up. Could I do that?\nThe log level seems correct following these guidelines: I am not sold on the value of glog's infinite log levels feature, however I could be convinced that our logging guidelines need a TRACE level.\nThank you for your response. I'm on your side, How about ?"}
{"_id":"q-en-kubernetes-688a1363ab44b38e73304094013c6bfb676fae85b86e65acef0c6925731c996d","text":"<!-- This form is for bug reports and feature requests ONLY! If you're looking for help check and the . If the matter is security related, please disclose it privately via --Is this a BUG REPORT or FEATURE REQUEST?: /kind cleanup What happened: In PR we tightened the --provider parameter checking of the binary such that it only accepted known providers. This turned out to be problematic () because people had started to depend on the previous behavior, which was to accept anything and merely warn about it. For the sake of giving people time to adapt, PR reverted to the older behavior. What you expected to happen: In 1.14 we want to allow only known providers. While at it, the error handling for unknown providers should be enhanced (print proper error message instead of calling Failf, which panics). /assign /sig testing\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nkubeadm was one of the tools which got broken by the tightening of the parameter checking. Does kubeadm still rely on the acceptance of unknown provider parameters or can the current warning be turned into an error now? /remove-lifecycle stale\nyes the kubernetes-anywhere deployer is still bound to provider=kubernetes-anywhere (unknown) in test infra jobs. we went back and forth on this topic, but if you wish to replace this support for unknown deployers i can switch those jobs to . LMK.\ngiven one release has passed with the deprecated message on the CLI, it's probably safe to remove this functionality now. :)\n\"Lubomir I. Ivanov\" <: Yes, please do. Once that is complete, I'll turn the warning into an error.\ndone:"}
{"_id":"q-en-kubernetes-fcc3a4f0546a69a996e9acca03b96cb5d6a4520581f6771770c5248bc191f84b","text":"<!-- Please only use this template for submitting enhancement requests --What would you like to be : a new addon to addon manager \"that automatically installs CSI CRDs if CSIDriverRegistry or CSINodeInfo feature gates are true\". But this is either broken or this addon manager isn't active in a cluster brough up with: This should work automatically also for such local clusters. Why is this needed: The CSI E2E tests are now expecting that the cluster has the CRDs and fail when that isn't the case. /kind feature (or perhaps a bug, not sure) /cc\n/assign\nIs this expected to be installed in other clusters? A lot of clusters are not going to be using cluster/* scripts."}
{"_id":"q-en-kubernetes-78b37d9fe8b38ac1317f8aaccb649830f24b3fef9915f4a2195d4feb676a7b80","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: gce-new-master-upgrade-cluster-new-parallel Which test(s) are failing: Varies, but all subpath failures, including: [sig-storage] CSI Volumes [Driver: csi-hostpath-v0] [Testpattern: Dynamic PV (default fs)] subPath should fail if subpath with backstepping is outside the volume [Slow] [sig-storage] CSI Volumes [Driver: csi-hostpath] [Testpattern: Dynamic PV (default fs)] subPath should fail if subpath directory is outside the volume [Slow] [sig-storage] CSI Volumes [Driver: csi-hostpath-v0] [Testpattern: Dynamic PV (default fs)] subPath should fail if subpath file is outside the volume [Slow] [sig-storage] CSI Volumes [Driver: csi-hostpath] [Testpattern: Dynamic PV (default fs)] subPath should fail if subpath file is outside the volume [Slow] ... and pretty much every other subpath test, but never all of them at once. There's also a few other storage tests failing, such as: [sig-storage] Volume expand [Slow] Verify if editing PVC allows resize [sig-storage] Detaching volumes should not work when mount is in progress Since when has it been failing: 11/22 Testgrid link: https://k8s- Reason for failure: These flakes started around the time that merged, but doesn't match up with the exact merge stamp, so it's probably coincidental. The subpath test failures seem to be mostly timeouts: ... so possibly this is just GCE fail. Anything else we need to know: This test job has always been flaky, with around a 40% failure rate. <!-- DO NOT EDIT BELOW THIS LINE --/kind flake /sig storage /priority important-soon cc\ncan you please help investigate this. It looks to me like the tests are flaking for and . I wonder if the v0 tests are racing with the non-v0 tests for something in this test?\nThe other tests including for the other CSI driver and non-CSI in-tree plugins are solid green which also makes me think this is a issue with the hostpath tests or driver.\n(just adding additional similar flakes here for tracking) Similar flakes in :\nadding to the milestone for triage since the flakes look new\nThese all 4 subpath tests call and it fails in finding a specific failed event by using . As uses to check if the specific error event happens, and , I guess that tests may flake if other failed event happens before the expected one. Also, already wait for the event, we might be able to delete  in . I will create a PR to fix above.\n/priority critical-urgent /remove-priority important-soon /milestone v1.13\nwhile we should update the test with your PR, I'm not sure if it's going to fix the failure. Looking at one failing test log, these are the events I see on the failing pod: So it seems to be failing before even reaching the subpath logic.\nTimeline of events: Test pod started getting processed by kubelet: It took a minute after the test pod was created for the csi driver to finish registering and mount to succeed, and then subpath event to trigger: So should address the issue. I was initially confused because the test logs didn't show the subpath event for some reason\nI opened up to investigate if 1 minute for csi plugin registration is expected\nI've a tracker issue for the complete list of storage flakes: This is so that we can track whether we're addressing most of the flakes with the subpath fixes, or if additional fixes are required.\nkeeping open until flakes are verified cleaned up\n/reopen Reopening this issue until we can verify the fix in CI\nAs of this morning we are seeing CSI \"subpath should fail\" flakes in the following job Multiple \"subpath should fail\" flakes in this slow job alone Multiple \"subpath should fail\" flakes in the 1.13 slow job, similar to master. The following 2 tests are flaky only in this upgrade job can you help us determine if these are cases are flaky tests that you are fixing or is there an underlying CSI bug that needs addressing?\nroot cause of this is a delay in an external sidecar containers and the tests not waiting long enough to accommodate the delay. There are no changes required in (outside of test) to handle this. These should not be 1.13 blocking. We'll continue to look in to getting them non-flaky today. In the meantime, we can mark these tests as flaky to reduce the noise, if you'd like. looks like the same issue () is still being hit. See : And logs from (): We need to make the tests handle this condition more gracefully. The suggestion here is one possible solution.\nWould be good to capture this as a known issue in release notes.\nto followup on capturing this in Release notes\nper , can you create a targeted fix increasing the timeout? Looks like marking these tests as flaky will be difficult for 1.13, but if that doesn't fix it we will need to do that.\nShort term fix to bump the test timeout:\nAs per discussion in release burndown meeting, this is not a 1.13 blocker. please move this issue to 1.14 if there are no pending issues you are concerned about\nOk looks like the test still flaked. Looking at , the mount succeeded after a minute and proceeded to subpath processing, which (correctly) failed. However, for some resaon, the test is not seeing the subpath event and still failed after 4 minutes. On kubelet, subpath processing failed after a minute and generated the event: But the test doesn't see it for some reason, and still times out after 4 more minutes:\nI saw the same on CSI volume mount succeeded at: Kubelet correctly failed the pod because of subpath But test still failed: Because the test failed to find the pod event. This looks like a test issue. Not a 1.13 blocker.\ncan you PTAL at this\nIs there a way to check if events are overloaded and not able to handle new events? The MountVolume retries while the driver is not registered is very frequent and could have overloaded the event server.\nChecking for the existence of an event seems like the wrong way to test this. Would checking pod status alone be insufficient? Ultimately that is what we really care about.\nChecking for the specific event guarantees that we are failing with the correct error. But perhaps events are not reliable enough to depend on. Checking if the Pod never came up running would probably reduce the flakes but could potentially mask false negatives. cc for any thoughts.\nTests are still flaking and actually it seems like in general storage flakes in slow are flaking more since 11-21 or so: https://k8s- My theory is that the driver registration delay is causing a flood of events, which causes tests that check for events to fail. I think this is revealing multiple issues in both the tests and product, and we should work on fixing them all, although any one of the fixes should mitigate the test flakes: Driver registrar can sometimes take a minute to come up: Volume manager operation executor doesn't seem to be throttling retries or backing off: Tests should not rely on events: Tests should wait for CSI driver to be registered on all nodes before creating test pods:\nand have been merged and will address the biggest issues. has been backported to 1.13 already, backport is still pending.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-testing, kubernetes/test-infra and/or . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-654ece199003f278d006ccf57d00b1bb1a0a4ea2c01afabbffd3256d622e5402","text":"What happened: POST or PUT to a custom resource with a JSON body missing \"kind\" results in a 500 error: A missing \"apiVersion\" also results in a 500 error: Then with both \"kind\" and \"apiVersion\" defined we get the normal 4XX return codes. What you expected to happen: All bad inputs for create and update requests should be 4XX return codes. E.g. Creating a with body of results in the 422 error: How to reproduce it: Create a CRD or use a cluster that already has a one prepared (GKE 1.11.0+ clusters have the CRD installed which you could test against). HTTP to that custom resource with a JSON body of . E.g. Environment: Kubernetes version (use ): 1.11.2 Cloud provider or hardware configuration: GKE OS (e.g. from /etc/os-release): Kernel (e.g. ): /sig api-machinery <!-- DO NOT EDIT BELOW THIS LINE --/kind bug\nThis is fixed on master, but I can reproduce it on 1.11. Posting a custom resource with a JSON body of : on master: on 1.11:\nwe are transforming all non-status error into 500 status error. in this case, it's the decoder returning a which is non-status. a quick fix by . (i like this number)\ninstead of fixed, it's rather been shadowed by i suppose. now when that empty unstructured obj going thru the decoder, the gvk is explicitly coerced, so this case is not reproducable on master for now.\n/cc\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\nI get the same response when trying to update the custom resource's 'status' sub-resource. On k8s v1.11.4\n/assign\nsubmitting the specified object on master returns a 400 badrequest, as desired /close\nClosing this issue. <detailsIn response to : submitting the specified object on master returns a 400 badrequest, as desired Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-8827ca63004a47357cfefee3cd648dabc7eed493bce201b21f63d05efb6b62be","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks!--What happened: If kubelet plugin registration fails during the registration process, the kubelet never attempts to register the socket again (assuming the error was transient), unless the socket for the driver is deleted & recreated to trigger a new registration. What you expected to happen: Maybe retry with exponential backoff when kubelet plugin registration fails instead of just logging and waiting for another filesystem event like it does today. Of course, if a socket returns a terminal error (does not implement the kubelet plugin registration service, for example), we should not retry. Follow up on /milestone v1.14 /sig storage How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: <!-- DO NOT EDIT BELOW THIS LINE --/kind bug\nI'd like to work on this issue\n/assign\nRe opening. was a temporary work around for this. is the permanent fix.\nHello! We are starting Bug Freeze for 1.15. Is this issue still planned on being incorporated into 1.15?\nthe dependent PR is not yet merged. can we get this in for 1.15?\nsaid that is the \"permanent fix.\" and that has merged. So let's close this out /close\nClosing this issue. <detailsIn response to : said that is the \"permanent fix.\" and that has merged. So let's close this out Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-754a337748c77aa830912aa5c369fd083977cc37177bb1280fd450d5c939bfe3","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks!--What happened: In the test failed due to namespace leakage: we created e2e-tests-load-30-nodepods-1-pfn9m, but never got response from api server (got 0-length response instead) we retried and created e2e-tests-load-30-nodepods-1-m9tq5 e2e-tests-load-30-nodepods-1-m9tq5 to f.namespacesToDelete and leaked -pfn9m next test case failed because e2e-tests-load-30-nodepods-1-pfn9m wasn't deleted. What you expected to happen: While it's unclear why we got 0-length response instead ( investigates that), we always can get error while the namespace was created (e.g. connection reset, timeout, io error). We should handle this case in retrying correctly. How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): v1.14.0-alpha.0.793+ Cloud provider or hardware configuration: gce OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: <!-- DO NOT EDIT BELOW THIS LINE --/kind bug\nThere are no sig labels on this issue. Please add a sig label by either: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. See the . The in method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/cc"}
{"_id":"q-en-kubernetes-53cfc136f9589b29086751241b2b69f45850f5b53dcdf9ee898447ca23815858","text":"What happened: AAD support in kubectl does not work for sovereign cloud The 'environment' parameter was introduced to support sovereign cloud, but it works in first round. The 'environment' value is not persisted, so follow-up requests will fail. What you expected to happen: 'environment' parameter persisted in kubectl config file. How to reproduce it (as minimally and precisely as possible): Use kubectl to connect to Azure sovereign cloud, make the token expire, and do auth. Anything else we need to know?: Environment: Kubernetes version (use ): 1.11 Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: <!-- DO NOT EDIT BELOW THIS LINE --/kind bug\nThere are no sig labels on this issue. Please add a sig label by either: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. See the . The in method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-0ed0c9c8e2de14d9bea2caa689fb3d4db2dcd076956a35485cb763c276972f01","text":"What happened: removed reporting of the OutOfDisk condition in the kubelet. A 1.12 kubelet upgraded to a 1.13 kubelet leaves the previously reported OutOfDisk condition in place, and it eventually gets switched to Unknown status. What you expected to happen: If this condition is no longer reported, it should be removed by the kubelet. How to reproduce it (as minimally and precisely as possible): a 1.12 kubelet, let it report OutOfDisk condition status. to 1.13 the condition is no longer updated a while, the node controller flips the status to Unknown Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: <!-- DO NOT EDIT BELOW THIS LINE --/kind bug /sig node\nThe fix for this should be picked to 1.13\n/assign happened to have a fix handy, if no one is working on it.\nafter upgrade kubelet should ignore this condition . or change it to 'DiskPressure'\nBetter not. What if both condition and exist?\nshould be picked to 1.13 also, I think.\n13.2 doesn't have this patch; would it be possible to make it for 1.13.3?\nUnfortunately this was not mentioned in the release notes (or I missed it), got hit by this too. 13.2 is affected too, would be good for me to see this getting fixed in 1.13.3"}
{"_id":"q-en-kubernetes-536ebee2479645b5b719f09972b970f703777471342fc75981f2b5f9dd5ef7ea","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks!--What happened: When using the function from Python's Kuberneets API, we noticed that Pods are terminated bythe cluster without respecting their preStop hooks or terminationGracePeriodSeconds seconds. What you expected to happen: When using the eviction API, just as when using delete, Pods should be allowed to run their preStop hooks, as well as be given terminationGracePeriodSeconds before being terminated with SIGKILL. How to reproduce it (as minimally and precisely as possible): the following pod: the pod logs in another window to evict the pod. Here is a sample curl equivalent of what (1.13) is doing - happy path: Here is a sample curl equivalent of what Python's is doing (when not passed ) - broken path: The log output generated by the guinea pod on the happy path looks like this: The \"Got signal 15\" message shows up only after the preStop hook is finished, so ~10 seconds after the evict call is made What happens on the broken path is: We seem to be getting 2x SIGTERM, there are no messages indicating continued processing after the signal, and the signal itself is registered immediately after the eviction call is made, suggesting that preStop didn't run. Anything else we need to know?: Environment: Kubernetes version (use ):Cloud provider or hardware configuration: AWS OS (e.g. from /etc/os-release): CentOS Linux release 7.5.1804 Kernel (e.g. ): 3.10.0-862.14.4.el7.x8664 SMP Wed Sep 26 15:12:11 UTC 2018 x8664 x8664 x8664 GNU/Linux Install tools: in-house Others:\n/sig api-machinery\nthis is because of this code: invoked from here: it is passing in the eviction delete options field as-is, meaning nil gets propagated and defaulted to mean delete immediately. when calling delete APIs normally, delete options defaults to a non-nil empty set of options:\nFixed in\npatch looks legit. Thanks for the quick reaction and for the pointers!\nI've a elasticsearch stateful set which have a preHook which never completes and the pod is never killed despite terminationGracePeriodSeconds being set. Is it related ? If so can this be considered to be cherry-picked on v1.13.x release branch ?"}
{"_id":"q-en-kubernetes-8a691359afd12eed0a607a985abb219c9f1a4fc754f792f7946964f4e4d965ba","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks!--What happened: I'm running a K8s 1.13 cluster and a partial netsplit happened due to a defective linecard dropping only TCP packets above a certain sizes between two masters, but the third could still communicate with both (so the fault made the network non-transitive). etcd was run on the physical network, but Kubernetes traffic runs on a Wireguard overlay for security and routing is provided by OSPF (so internal communication was not affected since it was tunelled over UDP). Etcd experienced some temporary issues due to the connections dropping only certain packets (above ~1000 bytes), but quickly dropped one host out of the quorum, at which point it was degraded but all the remaining hosts could communicate freely. Each master node has an etcd instance and an apiserver. All apiservers are anycasted inside the overlay to balance the load and provide redundancy and the announcement is coupled to the health check, so in case any apiserver fails no traffic will be routed to it. The problem was that the apiserver did not fail its healthcheck even though all etcd write requests timed out. The non-working apiserver was still accessible and got routed its share of the anycast traffic since that was routed over the UDP overlay. Thus ~1/3 of all workloads and nodes were stuck on a non-working apiserver until I nuked the non-working apiserver. TLDR: One etcd server started timing out all writes due to a non-transitive netsplit and the Kubernetes apiserver healthcheck did not fail, causing 1/3 of K8s clients to not have a working control plane until manual intervention. What you expected to happen: The health check should've failed (in my opinion, I'm happy to discuss if you think otherwise). How to reproduce it (as minimally and precisely as possible): Set up 3 etcds with corresponding apiservers, introduce a split between two (shouldn't have to be a packet size limit, a full disconnect should reproduce cleanly), observe that all three report healthy even though at least one of them no longer works. Anything else we need to know?: Environment: Kubernetes version (use ): v1.13.0 Cloud provider or hardware configuration: Physical x86_64 servers OS (e.g. from /etc/os-release): Custom Kernel (e.g. ): Custom, based on 4.19. No patches affect this functionality. Install tools: Custom Others:\n/sig api-machinery (maybe cluster-ops?)\ncc\nLooks like the etcdctl endpoint health command does a random read after fetching the list of members - - we should probably do the same?\nLooks sane and would have prevented my issue. Should I PR this?\nsure. thanks\n/assign"}
{"_id":"q-en-kubernetes-54c721787801f302bf73cc1b2dd5a4c401204e68c80dfcb6e42ea3d8ce05c21e","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks!--What happened: azure file remount on Windows in same node would fail What you expected to happen: If user delete a pod with azure file mount in deployment and it would probably schedule a pod on same node, azure file mount will fail since command would fail if file share is already mounted on the node: It's related to bug: Error logs would be like following: How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): v1.9-v1.12, master Cloud provider or hardware configuration: OS (e.g. from /etc/os-release): Kernel (e.g. ): Install tools: Others: /assign\n/sig windows\n/sig azure\n/milestone v1.14 cc This may be good to backport to 1.12 or 1.13 too if customers are using those versions."}
{"_id":"q-en-kubernetes-8e63dbf625d4bef689d56756744a33822ff52bb047451e9e4f46ad247787e451","text":"We need to add Windows-specific DNS tests to cover the following fields: Test cases needed for : Default, ClusterFirst, None Test cases needed for Test cases needed for Test cases needed for These need to run on Windows containers because the domain resolution policy is different, so different results need to be checked. Windows assumes that any name with a in it is a FQDN, not a PQDN. There is an to cover simple PQDN resolution, but they don't cover the Pod DNS configurations above. /sig windows /milestone v1.14\nFYI I'm working on this. I plan to start out by porting over to Windows.\nI just moved it into the in progress column. THanks!\nWhat test cases are needed for , , and ? My thoughts were that covers and . I'm adding test cases for Default and ClusterFirst. Does cover the test cases needed for and or do I need more? Or should I just be porting over the entire file?\nFor the first one - yes. The main differences will be checking the output of instead of , and maybe using instead of but the flow is the same Checking second one now\nAgree on second one. I think that will cover it. It may make sense to simplify it a bit because the common methods used there are all based on running bash + dig. If it works with busybox, that may be feasible. The description at matches those test cases\nIf at all possible, could we use native Windows cmdlets such as instead of relying on Linux cmdlets that have been ported to Windows like ? Am not 100% confident if they are going to behave consistently.\nyeah, I think what David's suggesting makes sense if the test cases are forked for Windows.\nOpened PR for enabling custom DNS on Azure CNI:\n/milestone clear This will need to merge after 1.14 as we're waiting on CNI updates for these tests to pass. I this status note to the Windows kep\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\n/remove-lifecycle rotten"}
{"_id":"q-en-kubernetes-8baab40e69f0284fe537fe8f20f37c00d552721258221dd50325dab698334f76","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks!--What happened: In an otherwise-stable bare-metal setup, with only LocalVolume and NFS storage, and without any volume mount activity, I get 80,000+ of these warning messages per day: What you expected to happen: No warnings; and if there is a reason to generate these warnings, more info about what operation is being attempted should be to the warnings message. How to reproduce it (as minimally and precisely as possible): I don't know. This issue emerged after I about 40 services and 300 volumes on a 3-worker cluster. Anything else we need to know?: Environment: Kubernetes version (use ): 1.13.0Cloud provider or hardware configuration: Bare-metal OS (e.g. from /etc/os-release): Ubuntu 18.04.1 bionic Kernel (e.g. ): 4.15.0-42-generic Install tools: kubeadm Others:\n/sig storage"}
{"_id":"q-en-kubernetes-a58991ec54e0de3f67ffc848889c93ae955a2930e4c50ed7fa5e4ebd3e04ff11","text":"Which jobs are failing: Which test(s) are failing: Job Reason for failure --- --- Since when has it been failing: 2019-02-11 Anything else we need to know: started failing with commit started failing with commit cc /kind failing-test /priority important-soon /sig api-machinery"}
{"_id":"q-en-kubernetes-da635c92143e1506eed36bd24ea751bd342a1a4539f250ba5d99681805edffe2","text":"What happened: The preparation step of the e2e tests were failed sometimes on my local environment like What you expected to happen: These tests can be passed without any errors. How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): v1.13.1 Cloud provider or hardware configuration: OpenStack OS (e.g. from /etc/os-release): 16.04.4 LTS (Xenial Xerus) Kernel (e.g. ): 4.4.0-138-generic Install tools: kubeadm Others:\n/sig storage\nAfter discussing on the PR I have known I need to enable StorageObjectInUseProtection plugin of admission controller which is disabled by default. The corresponding e2e tests require the plugin is enabled, so I think it is better to enable the plugin by default for most users.\nInvestigation of StorageObjectInUseProtection plugin of admission controller under kubernetes/kubernetes: cluster/centos/config- Enabled cluster/gce/config- Enabled cluster/gce/config- Enabled cluster/kubemark/gce/config- Enabled hack/local-up- Enabled cluster/juju/layers/kubernetes- Disabled This is the reason why we don't face this issue on the upstream CI system, and I it on my local environment which is deployed with kubeadm."}
{"_id":"q-en-kubernetes-a34100f2eb8400ab82f47190ab9c18caed04a9eb22795d23b67020c7edc6c623","text":"I have been deploying master-branch kubernetes on Windows using flannel/host-gateway networking, with one master, one linux worker, and three windows worker. Recently, kube-proxy stopped running correctly. It exits with an error about not finding the HNS Network. I have confirmed that the HNS network exists on the host with the correct name (cbr0), and have identified the last commit where kube-proxy works. What happened: Starting with the commit merging , kube-proxy outputs the following error message: This is confusing because does show : (nat network removed from above output) What you expected to happen: kube-proxy to work in host-gateway mode. How to reproduce it (as minimally and precisely as possible): Build kubernetes at or after . Start flanneld, kubelet, and kube-proxy on a windows 2019 vm. See kube-proxy fail despite flanneld having created the network. Anything else we need to know?: Environment: Kubernetes version (use ): kube-proxy breaks on commit and later; works on Cloud provider or hardware configuration: vsphere OS (e.g: ): windows server 2019, with up-to-date patches Kernel (e.g. ): windows nodes are , linux master/worker are Install tools: deployed using /sig windows cc\nI'm aware of this issue. The fix will be out within an hour."}
{"_id":"q-en-kubernetes-5560ce61eb1481212ee336127c07ce979a7d506aa207b599921813846f59ed3b","text":"Which jobs are failing: All new tests introduced for api change Which test(s) are failing: New alpha subpathExpr tests [Feature:VolumeSubpathEnvExpansion][NodeAlphaFeature:VolumeSubpathEnvExpansion] Existing tests are working fine and the whole suite works correctly in the gce suite https://k8s- Since when has it been failing: 21st February 2019 (since merge of PR) Testgrid link: https://k8s- Reason for failure: Investigation shows that node e2e tests use a different framework pod creation pattern Doing some comparisons of tests that work and those that don't reveals differences in the standard client set as and the existing ones which do work go through the intermediate which uses The version has an additional function specifically for node tests\n/sig storage\n/sig node /sig test\n/assign\n/cc\n/sig test"}
{"_id":"q-en-kubernetes-b8ea5b149739df9d1153565d68823c4c5efdc3c47d47da1083c1f8303bea2ce0","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: in Which test(s) are failing: Since when has it been failing: 2019-02-27 () -- since it was Testgrid link: Reason for failure: Timeout: Anything else we need to know: /kind failing-test /sig storage /priority important-soon /milestone v1.14 cc\ncc Could you help with this please?\n/assign\nNeither of the failing tests mentioned in this issue are present in testgrid anymore. is all clear"}
{"_id":"q-en-kubernetes-aa432f08c029c053dd7cfcb048f4042f5867f2876fa4810ba8a4c59a6311440b","text":"What happened: Currently starting Kubectl failing on Azure Stack. What you expected to happen: Kubectl should start on Azure Stack environment. How to reproduce it (as minimally and precisely as possible): Try to deploy Kubernets on Azure Stack environment.. Anything else we need to know?: Environment: Kubernetes version (use kubectl version): master Cloud provider or hardware configuration: OS (e.g: cat /etc/os-release): Kernel (e.g. uname -a): Install tools: Others: /sig azure /assign\nis it ? is a tool ...\nmy bad its kubelet. I update the title.\nthen could you provide the kubelet log? I am not sure kubelet on Azure stack is running as a container or process, you could get the logs by: And how do you deploy k8s on azure stack? cc"}
{"_id":"q-en-kubernetes-892e0241d063b18638cd5cbf78924bd2bec214bc99c4a6b3f534188c99a38c9e","text":"<!-- Please only use this template for submitting enhancement requests --What would you like to be : 1) add metrics into kubelet to monitor ; 2) add throttling to . Why is this needed: Kubelet provides lots of its functionalities through its http server, and currently there is no server-side monitoring and throttling mechanisms. Adding metrics into kubelet to monitor its http server will improve its observability and debuggability; adding throttling will prevent kubelet being overloaded from serving too many http requests.\n/sig node"}
{"_id":"q-en-kubernetes-4d0998f98cafdaec4256d215130088b83422619ea1d1b2223d6ff9d851b2d4d0","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: When referencing init container's env value from its fields, kubelet complains What you expected to happen: Init container should be supported to get env values from its own fields, as to How to reproduce it (as minimally and precisely as possible): a pod with init container refering its fields value as env value kubectl create -f test-pod-1.yaml cannot be created pod events Anything else we need to know?: Environment: Kubernetes version (use ):Cloud provider or hardware configuration: OS (e.g: ):Kernel (e.g. ):Install tools: Others:\nReiterating the mentions to trigger a notification: <detailsIn response to : Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThe err happens at this line , it seems that we should search for init containers as well"}
{"_id":"q-en-kubernetes-70af5a1cedf5015eefb116c0451b90a7456fd23ff69c19081db26ac7f1d62745","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: N/A Which test(s) are failing: Since when has it been failing: Ever Reason for failure: Vm on Azure with no public IP assigned aren't able to perform a command to the internet. Anything else we need to know: The test (see ) tries to test the internet connection for a pod by pinging 8.8.8.8. But azure does not provide ICMP support for VM without a public IP therefore this test cannot succeed.(see ) A fix would be to change the test method to a tcp request instead of a ping request (the test already performs a http get to in its BeforeSuite (see ))\n/sig network /sig testing /sig azure\nyou may refer to (Pods don't have connectivity to external networks )\ncc WDYT about the proposal to change this from an ICMP call? We probably need input from sig-architecture as well since this as well /sig architecture\n/triage unresolved\nAn alternative is to issue a TCP/UDP request to 8.8.8.8 on port 53.\nICMP is dropped by Azure for security reasons. This is intended. Can use curl as a replacement. I will fix the corresponding test.\nSGTM\nCan you please review the related PR? I'll in the PR again. Thanks!"}
{"_id":"q-en-kubernetes-995d9901bec7e27fab2cf79fc167a930ec1037256d4fd506ff1df2260d69ee9e","text":"What would you like to be : Remove any logic in kube-proxy that tries to automatically clean up proxy rules based on the proxy mode. Only attempt to clean up if is set. Why is this needed: This is a continuation of a discussion from There's a lot of complexity that comes with switching the proxy mode on kube-proxy because it tries to delete proxy rules created from other proxy modes. This has a lot of side-effects and we should be encouraging users to reboot thier nodes along with switching proxy modes. One example of this can be found\n/sig network\nI can take this on - does this issue have/need a KEP?\nI think not but will defer to\nI think a micro-KEP is appropriate. Specifically to talk about compatibility with regards to flags and to talk about docs (not sure which, but we have to write this down :)\nalso to talk about --cleanup (I think it should probably clean up its own mode, not the others -- or maybe I am wrong, please think about the right thing here :)\n/assign (Yay, I can do this now!)\n/priority important-soon"}
{"_id":"q-en-kubernetes-ff8db826b2d26ffcaf2a48d3e4e9fc88e37882937f2286a9da1d70678ae9597b","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: LoopbackHostPort code is calling the which depends on the machine on which this test is running. Especially in this particular testase it looks for the ipv6 ip address which does not exist because the machine is not configured for ipv6. Because of the LoopbackHostPort function bahaves based on the interfaces configured, test will fail if machine is not really configured with ipv6. Need to rewrite the testcase to handle such situation. What you expected to happen: Testcase should pass\n/sig network\n/cc\n/triage unresolved I am a bot run by ‍\nIf this issue has been triaged, please comment . If you aren't able to handle this issue, consider unassigning yourself and/or adding the label. I am a bot run by vllry. ‍\n/remove-triage unresolved"}
{"_id":"q-en-kubernetes-7d23cd435e6e7fb198369752ef8cd3422b1f8a106c042cb30a911f62f0f37b32","text":"What would you like to be : Change excludeCIDRs from []string to []*IPNet in Proxier and perform the parsing at construction of Proxier. Why is this needed: This was proposed by over comments in Pr This improves the performance of isIPInExcludeCIDRs since the cost of parsing becomes one time.\n/sig network /kind cleanup\n/remove-kind feature\n/triage unresolved I am a bot run by ‍\nping"}
{"_id":"q-en-kubernetes-06090bc29717f53ed35b5f857d1a7f93dac3313da76c3b3f75481458fd51444a","text":"What happened: We're running pod behind and sending logs via UDP to it. It is k8s v1.13.3 with ipvs services. When pod gets restarted logs getting blackholed. Simplified example on a node: So, old pod had IP 192.168.129.116, new pod - 192.168.128.193. Weight correctly forwards all new traffic to new pod, but there are some which are still sending traffic to non-existing IP: Minimal value for UDP connection expiration is 1s: But logs are being sent at much faster rates, so these broken connections are never expire. Issue introduced by Looks like this supposed to keep old pod running, while there are still connections to it But is there a reason to still keep these realServer entries, while old pod already cleaned up and no such IP exist? What you expected to happen: would cleanup all non-existing realServers from ipvs entries. How to reproduce it (as minimally and precisely as possible): Repro steps are same as for host-port issue: Anything else we need to know?: sysctls are default: Environment: Kubernetes version (use ):Cloud provider or hardware configuration: on-premise OS (e.g: ): Debian 9 Kernel (e.g. ): 4.9.0-8-amd64 Install tools: Others:\n/sig network\n/triage unresolved Comment when the issue is assessed and confirmed. I am a bot run by ‍\ncc\nWe observe the same behavior in k8s v1.12.7 and it hits us hard. In our case Nginx in a pod uses coredns service IP as DNS resolver address and every time coredns pods get new addresses DNS queries are blackholed. Query rate is higher than ipvs UDP connection timeout (30s in our case), so the connection to a non-existent ip address stays forever. Kube-proxy keeps non-existent real server entry in ipvs service and conntrack entry is never getting removed either. Kube-proxy logs: ipvsadm -ln:\nI've been putting some thoughts on the issue (we actually have the same one). Currently, we handle graceful termination the same way for udp as we do for tcp. However on heavily loaded UDP services, (DNS for instance) udp ports are reused quickly and old backends are not deleted fast enough. There are two possible solutions: use more aggressive timeouts (30s instead of the default 300s). Not perfect but better immediately delete real servers and associated connections when a backend is not ready anymore (very good for DNS queries but this means no graceful termination for udp) / what do you think? I'm not sure what the exact behavior with iptables I can create a detailed issue with options to have a wider discussion. Or even directly send an email to the sig-network group\nI like the idea. BTW, iptables also does not support graceful termination for UDP connections.\nthis is very easy to implement: change the test here: make sure is set (which we do : )\nIf we agree this is the way to go, I can implement it (if iptables works this way this is the best solution, I agree) We could still tweak timeouts for TCP (defaults are very long: 900s, which can lead to complete blackholing for 15mn on a crash of a target because the connection is never closed)\nI agree.\n/area ipvs\nIf you have a test environment I have a test image with the fix discussed just above: Commit here: This is not perfect, but I believe it is much better. What happens is the following: udp target pod becomes NotReady IP is removed from RealServers immediately Because of any new packet reusing the port will generate a write error and the connection will be removed from the ipvs connection table (if the application retries you are fine) I think in addition to that we should set much more aggressive UDP timeouts (5s? instead of the default: 300s) and make it configurable with a flag for specific use-case (I haven't seen a use case for long lived udp \"connections\" with very low amount of traffic)\nSorry, i was unable to quick test this on local dev cluster. Tried to replace image with but it seems it does not work with k8s 1.13. It then generates a lot of the same messages, and no any ipvs VIPs are created. I've also tried on fresh new cluster with the same result. Then tried but was unable to even start the cluster (with modified config from our 1.13) Can you please cherry-pick this to 1.13 or 1.14 or am I doing something wrong?\nIf this issue has been triaged, please comment . If you aren't able to handle this issue, consider unassigning yourself and/or adding the label. I am a bot run by vllry. ‍\nWe tested this patch to kubernetes v1.12.7 and it works well in production. BTW, I don't know much about SCTP. But question is do we need to threat it as UDP or TCP in this case? But it seems more close to UDP.\nThanks for testing! The PR is limiting graceful termination to TCP which seems the safest. I don't know SCTP but making it behave like TCP would be an easy change.\nThis feels like a fundamental problem w/ endpoints... ideally the proxier would remove the realserver when the pod is nonexistent rather than relying on arbitrary timeouts. Otherwise, there is no way to support graceful termination with UDP based protocols. Not a big deal for most UDP things, but I imagine e.g. QUIC users would be affected. Immediately terminating udp connections is in line with the iptables behavior, which seems to be okay for most people, so the change seems reasonable until there is a better way to determine pod state.\nThe PR that was just merged will improve things a lot for UDP but I completely agree with you on the fundamental problem with endpoints. I think we should have 3 different states: Ready (accepting traffic), in which case we should route traffic to the endpoint NotReady (terminating for instance), in which case we should stop routing new traffic to the endpoint but should keep existing connections \"Gone\" (no in the endpoint object anymore), in which case we should remove the endpoint altogether Today the endpoint object is too simple to handle these differences. I hope the new EndpointCondition status in the EndpointSlice API will help address this problem. More than happy to integrate this in IPVS when it is released (cc / )"}
{"_id":"q-en-kubernetes-a5fd3f8c97b71c1b6febd1e30a31e7696fb42552edc46c1b3df644e51c67f804","text":"we need to add such a file under: test/e2e/framework see this file as an example: xref: /help /good-first-issue /area e2e-testing-framework /sig testing /kind cleanup\n/remove-good-first-issue /remove-kind feature\n/priority important-longterm /area e2e-test-framework\n/assign\ni think it might be a good idea to create the dep graph as JPG and have a chat about that during a meeting.\nWhat are the restrictions that we are trying to place in the e2e framework? In the example given it looks like the only thing blacklisted is provider-specific implementations or the internal client. I know one of the recent changes was ensuring we dont use the internal client, is that all we want to blacklist for now? Or also all specific provider implementations? Some other dep we want to prevent?\nI'd like to help on this issue, this would be my first issue so I'd need some help as well if that's OK.\n/cc"}
{"_id":"q-en-kubernetes-38db129a694acec080229fb8864bdeffe34dab3c8bb74b84a209c796e077985e","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Make release of v1.14.2 failed. After facing the error () we did 4 : eaafa5952f8073f51b13dbeafe406019995c8c0ba commit in 's opening PR () Then the build process failed again: What we did was: we ran two commands called , after that, we did . And right after this command, the error happened. Environment: Kubernetes version (use ): v1.14.2 OS (e.g: ): ubuntu 16.04, bash version 4.3\n/sig release"}
{"_id":"q-en-kubernetes-7dd992e79bede1333eea29d91300bc35053996291a11517e577968869cb33fdc","text":"<!-- Please only use this template for submitting enhancement requests --What would you like to be : Introduce a mechanism that allows specifying the IP protocol version used in the e2e testing. Why is this needed: This is a followup on , my understanding is that in order to discriminate between IPv4 or IPv6 we have to use the test labels and this adds complexity to the testing automation and doesn't allow to send this information to the context of the tests. Having a mechanism to specify the IP protocol will simplify the testing for IPv4 and IPv6 and will allow reusing most of the code.\n/sig testing /cc /assign Does this make sense? If affirmative, what should be the best way to add this functionality? Checking the code I think that or can provide this, don't know if checking some of the cluster config parameters, ...\nthe testing-commons group (sig-testing) maintains the framework, but ideally the decision on how to proceed in terms of adding ipv6 detection should come from sig-network. possibly having this as agenda for a sig-network meeting is a good start, then discussing it in a testing-commons meeting. /sig network the linked issue is a big one, so i don't have all the context here. but IMO existing tests should just work for ipv6 without the requirement for a new \"IPV6\" tag in test names. dual-stack or protocol specific tests should have their descriptions contain the details properly (i.e. what is this test doing) we might want to add an utility function under: (there is a PR to move it BTW ) then such an utility might be useful only to tests that need to detect the protocol. one of the goals is to refactor the code in testcontext / framework, so adding this logic there is less preferred.\n/area e2e-test-framework\nI think that we can add an option to test_context to specify the cluster IP family the same way that exists one for the cloud provider this is simpler because when setting up the CI jobs you will know beforehand what IP family you want to test, it also will allow reusing all the e2e code for IPv6 and Dual Stack (cc: and by default will be IPv4, so it´s backward compatible.\nBeing able to override and specify exactly what you want the test to beleive about your cluster is nice, but ideally it should generally work without specifying more flags. Surely we can detect this?\nI was thinking during some time about this, detecting the IP family is very simple having IPv4 or IPv6 only clusters, but with the introduction of dual stack I´m afraid about the supportability and reliability of the autodetecting mechanisms in the long term. If you see the dual stack PRs there are still a lot of unknowns about the cluster behavior. My point is, let´s have something first stable that works and is simple, and once dual stack is implemented, we can replace the flag with a function in the framework . I guess that migrating from a flag variable to a function is straight forward, is just replacing by , what do you think?\nin is as part of the context. do you mean adding as a function which will auto-detecting the ip family. less flags for the framework binary is preferred.\nonce we have this I will open a new issue to start cleaning up and , we no longer need to discriminate based on tags"}
{"_id":"q-en-kubernetes-844631212d61672d588215671b80c058c6787d965af7cbd5a18c20bcd00e1381","text":"What happened: The externalName of a Service is not allowed to end with a trailing dot. What you expected to happen: The externalName of a Service should allow a trailing dot. This is very useful to denote a fully qualified value of the CNAME record created by the service and prevent DNS-trial-and-error by appending different (cluster-local) search domains. How to reproduce it (as minimally and precisely as possible): Try to set as a externalName for a Service of type ExternalName. More details The reference reads: RFC-1123 section 6.1.4.3 also states: Also in the errata of RFC-1123 a reference to RFC-1034 is to reiterate on what a legal hostname is. Section 3.1 of RFC-1034 explains the use and validity of a trailing dot. So while one could argue that a valid RFC-1123 subdomain may have a trailing dot, it might still be best to not change the which is used in a lot of places, but instead change the validation locally. Anything else we need to know?: I'll send a PR. Environment: Not much applicable.\n/sig network\n/triage unresolved Comment when the issue is assessed and confirmed. I am a bot run by vllry. ‍"}
{"_id":"q-en-kubernetes-0c82c8af2735bd0287155b4d0e1c197e36b2d063499f94b97cd0f5b47c16329a","text":"What happened: Kops runs from a pod manifest stored on the node, having the annotation. We've experienced the case where the cluster autoscaler turned on a new node and, once the node has been to the cluster, application pods have been admitted before . Once it was the turn of , all system resources were utilized, so the pod has not been regularly admitted. However, since it's a critical pod, the executed the , which has tried to kill () a pod whose container wasn't started yet (it was downloading the Docker image), which has caused to timeout and return the error . The admission of failed with the following error: The pod execution has not been retried, so we've got a node without running and its networking broken. What you expected to happen: The should be eventually admitted. Anything else we need to know?: A minimal log extraction for comprehension: From the \"Failed to admit pod\" log I have been able to trace back the core code involved: () Logs: \"Failed to admit pod [POD] - [ERROR]\" () Error message: \"preemption: pod [POD] failed to evict [ERROR]\" () -() Error message: \"timeout waiting to kill pod\" Environment: Kubernetes version: Cloud provider: AWS OS: ubuntu 16.04.6 LTS (Xenial Xerus) Kernel: Install tools: kops Network plugin and version: kubenet\n/sig node\nI think the following change may alleviate the situation you encountered:\nThanks for your reply. To my understanding, when the is called it will return the minimum set of pods required to free up resources, so even if only 1 pod eviction fails it's likely that at the end of the the will not have enough resources anyway to admit the critical pod. In your opinion / experience: be an option introducing retries in the in case the eviction of 1 pod fails? yes, do you have any suggestion of a similar design adopted somewhere else in the ?\nLooking at getPodsToPreempt, I am not sure it returns the minimum set of Pods. For , retrying may not help much since the pod to be killed may not respond to the subsequent kill command.\nSorry to reiterate on this, but I double checked the code and - if I'm not missing anything - returns the minimum set of pods. Let's look at logic: pods by QOS class using that the total sum of evictable pods resources is enough to satify the requirements (resources to evict for the critical pod) the list of pods to evict, as the merge of three sublists: : the minimum set of pods to evict assuming we've already evicted all and all : the minimum set of pods to evict assuming we've already evicted all and the required : the minimum set of pods to evict to satify the requirements left Am I missing anything? A possible different approach to solve the issue is to refactor to return all input pods sorted by \"should be evicted first\". Then iterates over the full sorted list of potential pods to evict until the requirements are not satisfied (skipping to subtract pod resources from requirements if a pod fails on ) and breaking the loop as soon as the requirements are satisfied. What's your take on this? /cc who originally implemented this\nw.r.t. approach outlined above, it makes sense.\nWhether this evicts the \"minimum set\" depends on how you define the \"minimum set\". At the time this was implemented (k8s v1.5.0 IIRC), we didn't have PodPriority, so we used QoS to mirror the kubelet's eviction algorithm. It should probably prioritize low-priority pods for eviction today given the recent changes to the eviction algorithm to incorporate PodPriority. The current approach minimizes the eviction of guaranteed pods, and prioritizes evicting few larger pods over many smaller ones. Since eviction marks the evicted pod as failed, it should suffice to just admit the pod regardless of what is returned by the evictPod function. In future syncPod loops, the kubelet will retry the pod deletion steps that it was stuck on. It would eliminate this class of bugs entirely, which was the purpose of the feature in the first place.\nSee if the following aligns with your suggestion:\nSomething like that, yeah. I'm not sure the \"Failed to admit pod\" log warning is still required, but we can iron those things out during code review.\nDoing like this it means we do always admit pods, while we should do it only for critical pods, correct?\nPlease see this The change is for critical pod(s)."}
{"_id":"q-en-kubernetes-5eb0ac65b3d30c0951e0da1f97bdb8adf05f535c840f17ca4a42ad70e61c0160","text":"What happened: logs a nil error on success, EG: . What you expected to happen: kube-proxy shouldn't log an error on success. Either a nil error should be interpreted as success, or an explicit exit boolean should be returned & evaluated. How to reproduce it (as minimally and precisely as possible): Run in master/1.15, or 1.14.2+. Anything else we need to know?: Code indicates that this is a problem ever since exiting was introduced (kube-proxy previously hung). It calls log.Fatal() on the return result, under the now-false assumption that an exit -a crash. Environment: Hyperkube, 1.14.2\n/sig network /priority important-longterm /cc\n/triage unresolved Comment when the issue is assessed and confirmed. I am a bot run by vllry. ‍\n/assign\nGitHub didn't allow me to assign the following users: johscheuer. Note that only and repo collaborators can be assigned and that issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI would start looking into this but it seems I can't assign myself\nthe bot only works on org members\nGitHub doesn't allow assigning people unless they are either members of the organization or repo \"collaborators\". There's no way around this. Kubernetes doesn't use the latter mechanism in favor of the org membership process. (some notes / links in the bot comment). You can certainly take a stab at this without assigning it though"}
{"_id":"q-en-kubernetes-083944d00199b655f25fb35f51e7ea3786d4181f487cac5cdbda9f2140ce4497","text":"<!-- Please only use this template for submitting enhancement requests --The built-in and cluster roles here are deprecated in 1.13 when this issue was resolved: /sig storage /milestone v1.17 /priorit important-longterm\nYou must be a member of the GitHub team to set the milestone. If you believe you should be able to issue the /milestone command, please contact your and have them propose you as an additional delegate for this responsibility. <detailsIn response to : <!-- Please only use this template for submitting enhancement requests -- /milestone v1.17 /priorit important-longterm Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/cc /priority important-longterm\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale /assign\nBug triage for 1.17 here with a gentle reminder that code freeze for this release is on November 14. Is this issue still intended for 1.17?"}
{"_id":"q-en-kubernetes-ba159de838e276a42915cbe831ad277f6b9150cbf7cfdd40b9245b21c2eb7560","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: kubelet terminated unexpectedly during hack/local-up-cluster kubelet log as a result apiserver cannot find the node to schedule pods What you expected to happen: kubelet runs successfully and hack/local-up-cluster creates a single node cluster How to reproduce it (as minimally and precisely as possible): run hack/local-up-cluster from HEAD Anything else we need to know?: additional cgroups check seemed to be introduced in ; hack/local-up-cluster works when I check out the parent commit have CI for hack/local-up-cluster and , so I'm not sure if it's an error on my side. It would be greatly appreciated if someone could explain it to me Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Debian GNU/Linux Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others: /sig node cc\nThanks for raising such a detailed issue! cc any thoughts here? Confusingly, I think the kubelet has been assigned a different cgroup for memory and for cpu... however the fact that its using makes me think the user is relying on system defaults. out of curiosity, have you made any mods to how your system assigns cgroups?\nIt sounds like we might want to handle the error more gracefully. If we can't get the kubelet's cgroup, we should fall back to collecting metrics from all cgroups.\ncollecting metrics from all the cgroups on the system? Any concerns about how many cgroups there could be? Also, or any interest in implementing this fix (assuming we go with the fallback behavior)? If not, I can try.\nthanks for looking into this! Sorry I'm not familiar with cgroups. I'm don't know any modification to my system assigns cgroups in particular\nI can put out a PR this week\nxref\nThanks!"}
{"_id":"q-en-kubernetes-3f173d1022a4408031efac17c8299f04d835794ab8a77d69bb7667b0107a08bf","text":"Some kubelet metrics use non-standard prometheus collectors unnecessarily. Specifically, we have which can be simple gauges. Since we cannot migrate non-standard prometheus collectors using the internal metrics stability framework, we will need to convert these metrics into standard prometheus variants. /sig instrumentation /assign\nGitHub didn't allow me to assign the following users: solodov. Note that only and repo collaborators can be assigned and that issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : Some kubelet metrics use non-standard prometheus collectors unnecessarily. Specifically, we have which can be simple gauges. Since we cannot migrate non-standard prometheus collectors using the internal metrics stability framework, we will need to convert these metrics into standard prometheus variants. /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nrecommends that we generate those metrics in the GenericPLEG during the relist. We can generate the data directly from the pods call there ().\nI would be interested to work on this issue if no one else is working on it currently\nGo for it!\n/assign"}
{"_id":"q-en-kubernetes-f5b1c2afeb6c20ee6d20a93f01654fa0f4acb253484063e97ce2510ee236c81b","text":"What happened: tried promoting a test with a tag in its full test description. The hierarchy of ginko-relevant AST's was . The tag was a in the top level . The full test description only came from , so the tag was missed. What you expected to happen: The conformance test list generator should have logged and ERROR and failed a test on the PR, calling out Disruptive as an invalid tag How to reproduce it (as minimally and precisely as possible): Check out locally and follow the directions in to generate the conformance list. The invalid test is in the list. Anything else we need to know?: Is there a function we can use to get the full test description from Ginkgo instead of rolling our own? If not, we should probably be maintaining a stack rather than just the last FYI who ran into this FYI who wrote the code to keep track of last\n/sig architecture /area conformance\n/milestone v1.16\n/priority important-soon\n/assign\nGitHub didn't allow me to assign the following users: devaips. Note that only and repo collaborators can be assigned and that issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-551b6084bfaac05cef6a997a25e281e330936f51feb2c4bba89f5a0e79ea7f67","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: attach/WaitForAttach timeout value is too small in CSI driver, it's all 15s which is really too small: Compared to built-in driver timeout, it's really too small: is , total is What you expected to happen: How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): 1.13 - 1.15 Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others: /sig storage /assign"}
{"_id":"q-en-kubernetes-e2ad17169ceb1098c807ca3a4621d8a63ce2959afa0affc3c6c5e92b949508b1","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: cd $ $ make ../image- bin webhook ../image- line 145: GOARM: unbound variable Makefile:23: recipe for target 'bin' failed make: [bin] Error 1 What you expected to happen: The binary or image should be built successfully. How to reproduce it (as minimally and precisely as possible): see my desc above, type \"make\" will see the error message. Anything else we need to know?: Environment*: Kubernetes version (use ): latest Cloud provider or hardware configuration: OS (e.g: ): Ubuntu Kernel (e.g. ): all Install tools: Network plugin and version (if this is a network-related bug): Others:\nThere are no sig labels on this issue. Please add a sig label by either: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. See the . The in method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\nI'm pretty sure this is supposed to be built via from which does default this environment variable.\n, thanks for the information! I think the harness of the shell scripts could bring us a little more, GOARM should only been set when the ARCH is ARM Make a binary is possible where a Makefile is presented\nThat seems reasonable, but we should probably avoid computing this in many places, currently the GOARM=${GOARM:-7} is going to always be 7 when building in the currently supported way. We should change that if we're going to compute it elsewhere. cc-ing for test/images fun ..."}
{"_id":"q-en-kubernetes-1a9bb0fa341de4f203eba4b1a2198b6b5b48cdf4b2b045035cfa84b6b4f2a114","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: login system root, and run below command from \"test/images\" dir, Will fail with below error message, ./image- line 91: : command not found This is cause by the empty string like \"\" cannot be understood as a shell command. What you expected to happen: image should be built successfully. How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): latest Cloud provider or hardware configuration: OS (e.g: ): all Kernel (e.g. ):all Install tools: Network plugin and version (if this is a network-related bug): Others:\nThere are no sig labels on this issue. Please add a sig label by either: a sig: e.g., to notify the contributor experience sig, OR the label manually: e.g., to apply the label Note: Method 1 will trigger an email to the group. See the . The in method 1 has to be replaced with one of these: bugs, feature-requests, pr-reviews, test-failures, proposals. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign"}
{"_id":"q-en-kubernetes-5c2fa5d68b30b20426b3cc550ef8c760b15b7eae41d09aebcd2cc187fe5717fd","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: pull-kubernetes-bazel-test Which test(s) are failing: Test_buildClientCertificateManager Testgrid link: Reason for failure: Anything else we need to know:\n/remove-kind failing-test /kind flake /sig scheduling\n/sig node"}
{"_id":"q-en-kubernetes-6bd1d79418aa070c75893275b9a56665000ecdc164e6d23c45498a1bfe6deca4","text":"What happened: The new TestScorePlugin for the scheduler framework is flaky How to reproduce it (as minimally and precisely as possible): An example: Anything else we need to know?: The test evaluates that a pod should be scheduled on specific node with a higher score. I think the test sometimes fails because the higher score node that the test creates didn't propagate yet to the scheduler (i.e., the node exists in etcd, but scheduler hasn't been notified that it exists yet, and so it is not evaluated at all). /kind bug /sig scheduling /assign\n/kind flake"}
{"_id":"q-en-kubernetes-f2cf3fb164978f0a8c62bbea5f6f0d6f6c61435524bb489ae50eec1a242fdf48","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Error msg override when saveVolumeData fails. What you expected to happen: return the right error. How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig storage\n/assign"}
{"_id":"q-en-kubernetes-b5b68ca719178ee7e086c976c8ddfc4c67500e8ea5e1389b8d55d179ba472c77","text":"What happened: When using a yaml file for input, silently fails at the first missing resource, not returning any further results. What you expected to happen: Searching with cli arguments properly ignores missing resources and continues the query, eg: Would expect similar results with a file input, eg: How to reproduce it (as minimally and precisely as possible): With an input file of : the command fails at apparently the first missing resource: Note: that rearranging the inputs in (so that the missing resource is last) produces the \"expected\" output: Anything else we need to know?: Environment: Kubernetes version (use ):Cloud provider or hardware configuration: local OS (e.g: ): fedora 30 Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others: x-ref:\n/sig cli\n/assign\nGot around to reproducing it and can confirm that I'm getting the same behavior. I'm looking into fixing it now\nthanks, any updates on this?\nI've learned that in the case of a fake entry, kubectl stops sending API get requests altogether. Learned by running kubectl get -f --ignore-not-found --v=10, the result is shown below. Till now I've been looking into the cobra command framework but now I'm trying to find out where kubectl is making these API calls from(working on finding out where this connection is happening from Slack+debugging, I've asked a more detailed question here ). <img width=\"1440\" alt=\"Screen Shot 2019-08-12 at 4 13 41 PM\" src=\"https://user-\"\nthat's an interesting find, definitely helpful. When I looked into this I got to the section of get's main run function, and it seems like it should be continuing on any errors (see ) but maybe the ResourceBuilder is actually stopping when it hits a 404?\nThis behavior comes from where the not found errors are being ignored and where the input list is flattened. Errors are ignored here: While flattening occurs here: As the visitor iterates through the objects in the list if an error occurs, such as not found, it will stop iterating over the objects in the list and return that not found error. The logic will then ignore it and continue. Any objects that came before the not found object in the list will have been visited and will then be included in the output. To make the behavior consistent between the two ways of getting the set of objects, the would need to be updated to support filtering errors.\nJust submitted a PR adding that functionality to the FlattenListVisitor. The behavior is now consistent between the two methods."}
{"_id":"q-en-kubernetes-293cdea3c6ec5a83fbc9e02a1638fe58db21159d7cc077448d29d2c4fb193ff6","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: CRI-O e2e Which test(s) are failing: e2e suite [sig-network] Services should be able to change the type from ExternalName to NodePort e2e suite [sig-network] Services should be able to create a functioning NodePort service 19s See link: https://openshift-gce- Since when has it been failing: Anything else we need to know:\n/sig network\n/triage unresolved Comment when the issue is assessed and confirmed. I am a bot run by vllry. ‍\n/milestone v1.16 /assign FYI Looks like two other release-blocking tests too: - The changes link testgrid gives () suggests two other PR's, but I agree that seems the likely culprit\nBased on the , Service is reachable on and but failing to reach on node's IP over nodePort i.e. . From the log, it is not clear why its not reachable on node's internal IP. Could you provide from the CRI-O based node when trying to reach nodePort service. Secondly, need to make sure, node is listening on allocated nodePort. As this issue seems much on system-level, I need to setup k8s cluster/node with CRI-O runtime and troubleshoot to find the root cause.\n[sig-network] ESIPP [Slow] [DisabledForLargeClusters] should work from pods [sig-network] Services should only allow access from service loadbalancer source ranges [Slow] seems flaky to me. Most of time execPod starts running and might be terminating immediately after some time which causes this issue. [1.] e2e is failing as might have terminated and so no PodIP leading to invalid . [2.] empty . The same issue had been observed in for E2E and to fix the issue, we introduced execPod deployment to make sure execPod would remain in desired running state. It's better to address this change in another PR and emphasize using pausePod deployment.\nThanks, for helping me setting up CRI-O based cluster. Kubernetes e2e suite [sig-network] Services should be able to change the type from ExternalName to NodePort e2e suite [sig-network] Services should be able to create a functioning NodePort service 19s From log, it is clear that, these services are failing to reach over allocated nodePort. But on my CRIO based testing cluster, they are reachable. or Could you please check if they failing at your end too or it's testgrid Job-specific issue?\nHey, I did some local tests and was able to reproduce the failure. I started the cluster via (check that the variable is correct and does not point to ): Executed the tests by (again, watch out for the variable to be correct): Error output: Since I'm now able to reproduce the issue I can go further in debugging and find out how to fix this. Is anyone furhter, or If not, then I will target for fixing the issue this week. :)\nThanks for reproducing this. FYI we've seem similar failure on , but I haven't reproduced the problem locally yet. We are looking to at least turn on verbose on the test command for debugging:\nThe issue is that we try to automatically discover the node IP by obtaining its status here: In our setup the node has IP which for sure does not work if we try to access it within the pod. I'm thinking about a proper solution and will propose it as a PR.\nIf this issue has been triaged, please comment . If you aren't able to handle this issue, consider unassigning yourself and/or adding the label. I am a bot run by vllry. ‍\n/assign"}
{"_id":"q-en-kubernetes-f32b7c5428ec40f8176edd9ab6a3a19435c1e0344527c7b4f5c2634d6944e927","text":"[x] write e2e tests for all admission operations [x] (for resource APIs) [x] (for discovery docs) [x] switch to v1 admission APIs in e2e - [x] remove skips/time.Sleeps - and promote to conformance tests - AdmissionWebhook should have tests for all stable operations. Those tests should go through the conformance process before being released. This ticket is to help track what operations are yet to be covered by conformance tests. These stable endpoints should be hit by e2e tests tagged with [conformance]. There are tests currently hitting 6 of the corresponding beta endpoints that could be updated to use stable. Tests should be written to exercise the remaining 9 untested stable endpoints as well. These twelve tests could be updated to use the stable API, and subsequently promoted to conformance. / ! ! These will need tests written for them.  untested beta/admissionregistration !\n/sig api-machinery\n/assign\n/area conformance\n/assign ^ This is a requirement for graduation to GA.\ncan we get another apisnoop run? We've coverage for admissionregistration and apiextensions and I'd like to get confirmation from apisnoop that we've got full coverage now.\nKeeping this open until tests are actually promoted to conformance\nis ready for review. It adds our admission e2e tests to conformance as part of graduation to GA.\nis updated currently not hit by any tests, one will need to be updated or written to cover it. Might want to ensure you've promoted one test that hit's each endpoint. !\nThe PR to do that is at"}
{"_id":"q-en-kubernetes-16d20469cfbbebd3a63b2abd343d03ad47978a2d98b642d88d94294a2a31e03e","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: With kubectl v1.16.0-alpha.2, completion with override flags fails. then hit TAB will produce error \"resource type must be specified\" What you expected to happen: It must show the namespaces without error. How to reproduce it (as minimally and precisely as possible): With kubectl v1.16.0-alpha.2, then hit TAB key. Anything else we need to know?: It looks like introduced this regression. Environment: Kubernetes version (use ): kubectl v1.16.0-alpha.2 Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig cli\nI also reproduced this bug with the following steps:\nI'm fixing it."}
{"_id":"q-en-kubernetes-9496dc8b8e00d14800df6ff70d9d0598864626d6dbfb05ec6d0deacd8726a6f4","text":"<!-- Please only use this template for submitting enhancement requests --What would you like to be : We need to enhance Azure cloud provider code to support both AAD and ADFS authentication. Why is this needed: On Azure Stack customer can have two type of deployment .\n/sig azure\n/assign"}
{"_id":"q-en-kubernetes-dcfa7bd39ff05f8f481c3621cc1f09ee2806ddcbff04ca803688eceda95fbb5e","text":"What happened: In reviewing a test for promotion to conformance I noticed the kubelet API was being accessed via the API server's node proxy endpoint (ref: ). This seemed off, so I reviewed . It doesn't explicitly call out kubelet API access as allowed or disallowed. . and agreed kubelet API access seems like a bad idea. I'm now surveying what we have that does_ assume kubelet API access. It's definitely used by parts of e2e framework and some Conformance tests. I'm trying to verify how critical its use is. It appears that much of it is related to debugging or verifying whether kubelets and the control plane agree on who is running what. What you expected to happen: Find a blanket policy that bans kubelet API access for Conformance tests, whether directly, or indirectly through the API server's node proxy endpoint, at any time. Alternatively, find a policy that says kubelet API access may be allowed, but only for debugging or failure purposes (must not be on the critical path to make the test pass) Alternatively, find a policy that says kubelet API access is allowed, but only indirectly through the API server, and evaluated against the same constraints we use for control plane access (GA, non-optional) If we want to say kubelet API access is banned, we can use this as a tracking issue to demote whichever tests rely on it, and come up with a plan to verify the same behavior without kubelet API access. /area conformance /sig architecture /priority important-soon /milestone v1.16 /assign\nLook for anything that ends up calling calls this in BeforeEach, which will happen before some Conformance tests (if this fails, the tests will fail) Conformance test requires kubelet API access to pass may end up calling it if the test fails, via: Some Conformance tests may end up calling this if the test fails specifically Other areas this is used: only if the test fails required for a pre-stop hook test to pass required to make sure the readonly kubelet apis aren't exposed via api server proxy resource usage tracking tests require kubelet api access to pass tests call this in\nLooking for doesn't turn up any additional conformance tests, and I don't feel compelled to link the two other tests that use it as they are unlikely to ever be eligible.\nThe examples pointed out by on slack, the kubelet's endpoints is required to be available for these to pass\nHello! I'm the bug triage lead for the 1.16 release cycle and considering this issue is tagged for 1.16, I'd like to check its status. The code freeze is starting on August 29th EOD PST (about 5 days from now) and once the code freeze starts, only release-blocking changes will be considered to be merged. Is this issue still planned to be fixed for the 1.16 release?\nare there further changes for this targeted at 1.16?\nAny update on this?\n/milestone v1.17\nBug Triage Release Team checking in. Code freeze for the 1.17 release is on November 14. Will this make it in? Thanks!\n/assign\nhi is this still targetting for 1.17 or can we move to 1.18? Code freeze is Nov 14th. thanks!\n/milestone v1.18\nSo we've updated the documented policy. My suggestion to close this out is to drop the [Conformance] tag from the tests that currently require the kubelet API to pass, and open issues that they need to be rewritten. Ignore whether the API is needed for diagnostic info in failure modes. SchedulerPredicates [Serial] calls this in BeforeEach, which will happen before some Conformance tests (if this fails, the tests will fail) Conformance test Pods should be submitted and removed [NodeConformance] requires kubelet API access to pass\ni'd support this ^^\nHello , Bug Triage team here for the 1.18 release. This is a friendly reminder that code freeze is scheduled for 5 March. Is this issue still intended for milestone 1.18?\nHello There's been no movement on this issue this year. Should this get pushed to the next milestone?\nyes, please John T Skarbek <:\n/milestone v1.19\n/cc"}
{"_id":"q-en-kubernetes-c5612df6fc8150ff4e6c5aa45fb73dc64d52cd4a3c4604a56174a837d5c098a3","text":"This issue was reported in the Description The container manager used in kubelet checks for docker daemon process either via pidfile or process name. While the pidfile points to the docker daemon process PID, the dockerProcessName constant stores a docker cli name (docker) instead of docker daemon name (dockerd). Figure 30.1: Constants in file. Recommendation Correct the docker process name to dockerd. Anything else we need to know?: See for current status of all issues created from these findings. The vendor gave this issue an ID of TOB-K8S-025 and it was finding 31 of the report. The vendor considers this issue Informational Severity. To view the original finding, begin on page 77 of the Environment: Kubernetes version: 1.13.4\n/sig node /kind bug /remove-kind feature\n/lgtm\nlol, this is the issue. my bad!"}
{"_id":"q-en-kubernetes-af2c92409aa3d43952a89129e5e1ec7b247d66b5bf6b0f901e05f1e1149218e9","text":"<!-- Please only use this template for submitting enhancement requests --What would you like to be : The scale client () currently works fine with namespaced resources. However, it doesn't work for non-namespaced resources e.g. cluster-scoped CRDs. The current implementation should be extended in order to support non-namespaced resources. The current interface: /area custom-resources /sig autoscaling\nI think scale subresource support for CRDs falls more into sig-apimachinery domain. After DirectXMan12 moved on from sig-autoscaling I don't think there is anyone in our sig-autoscaling with knowledge of scale subresource internals. /sig api-machinery\n/assign\nThe bug report sounds legit. Does the generic scale client need to be changed (I think so), or does solve your issue?\nNo, the dashboard PR doesn't solve this. I discovered this issue while working on that PR so when gets merged, it will only work for namespaced CRDs. I think the generic scale client needs to be updated, maybe by adding another getter to the ScalesGetter interface: /kind bug\nPR:"}
{"_id":"q-en-kubernetes-a8f763c740d3fb11e6b67f36722750f0a39e674fe923e77cebdc2dd4df02a0e7","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: ci-kubernetes-e2e-gci-gce-alpha-enabled-default Which test(s) are failing: CSI hostpath-v0 tests Since when has it been failing: 8/8 Testgrid link: https://k8s- Reason for failure: Anything else we need to know: We need to update client-go to get the fix for selfLink, build new nfs provisioner images, and update them in the e2e test. The fix for client-go is being cherry-picked to 1.15 if you want to use a tagged release:\n/assign\n/assign\nThis implies rebase to 1.15.x in external-storage, which seem to be challenging (at least to me :-) )\n14 cherry pick also in progress:\nOK, I guess this line in the lib is the culprit: So I need to update lib then external-storage to pull it in\nI don't think the lib itself needs to be updated. Only updating go-client to get the fix to is needed.\nRight, lib doesn't dictate what version of client-go to use. Will wait for client-go release/commit\n- thanks a lot for bumping the dependency. Did you have a chance to build the new container with it so that we could update it here:\nyeah, opened . (Not sure how to trigger test, or we can just merge and see.)\nYou can manually create a PVC on an alpha cluster to test."}
{"_id":"q-en-kubernetes-36ab61066741b9409307c7a95f47ca9d3b5b7f4d54f14d141fffedbc5c4e7548","text":"<!-- Please only use this template for submitting enhancement requests --What would you like to be : Add a sort parameter (or similar concept) to the command. Why is this needed: With the numerous amount of API resources (not counting custom ones), it is sometimes difficult locating them in the output. By default, API resources are not sorted by the name column (the first on the left) and often this is the column holding the information one wishes to query. A sort parameter would help more easily identify the object to be understood.\n/sig usability"}
{"_id":"q-en-kubernetes-02830743648dba6d35b2a5372ac09c56b9865c74b55a0ac8ad4b1c83d339c9c0","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: ci-kubernetes-node-kubelet-orphans Which test(s) are failing: [sig-storage] EmptyDir volumes [It] pod should support shared volumes between containers [Conformance] Since when has it been failing: Testgrid link: Reason for failure: Anything else we need to know:\nplease don't CC sig-testing on every test failure :( we own the infrastructure etc. the sig-storage owns these tests."}
{"_id":"q-en-kubernetes-49dfc85fc8498e0db279cbaabc579f0a0b63f738c4ccea222e5a413fa241b959","text":"What happened: There's a typo typo in and What you expected to happen: “depreciated” should be “deprecated”. How to reproduce it (as minimally and precisely as possible): View the release notes Anything else we need to know?: /lifecycle active I have a PR to fix this /sig docs Environment: Kubernetes version (use ): v1.16 Cloud provider or hardware configuration: not applicable OS (e.g: ): not applicable Kernel (e.g. ): not applicable Install tools: not applicable"}
{"_id":"q-en-kubernetes-99cf00288bb2f26fdcbec632d47295247bf238cc0435db3bc55f75aa0f1ddc53","text":"What would you like to be : Currently the apiserver has request latencies metrics, which are a histogram of requests to the apiserver. However, these request metrics do not include healthz. While we have recently the ability to get back health check metrics from the healthz endpoint directly, we should also emit this data as a metric, so that we can monitor health check latencies. /assign /sig api-machinery instrumentation\nGitHub didn't allow me to assign the following users: jktomer. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : What would you like to be : /sig api-machinery instrumentation Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI noticed this is informally assigned. Has there been any progress? Is this a place I can fit in?\nYes, I'm working on a PR.\n/assign\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale"}
{"_id":"q-en-kubernetes-0a2762540f2b531dffff6c46a9c4e32e93e7219960664c0ebd76521a168b5c1a","text":"What happened: Provisioning of a volume with failed and this event were created on a pod: The event does not say anything useful and only confuses users. What you expected to happen: Either no event (event on PVC about failed provisioning could be enough) or something generic about \"waiting for provisioning to finish\". How to reproduce it (as minimally and precisely as possible): Tested on AWS. It should be the same with GCE and other block-based in-tree volumes. StorageClass with PVC requesting RWX (so provisioning fails): pod that uses the PVC. Pod events: Anything else we need to know?: I've been able to track down what happens: Scheduler adds to the PVC Provisioner tries to create a volume for the PVC and fails. The annotation is removed in  The scheduler expects that the annotation exists here: I am not sure how to fix it though. Ignoring the error might be the easiest thing to do. Environment: Kubernetes version (use ): today's master /sig storage /sig scheduling /assign\nyes, the error message is confusing. \"FailedScheduling\" event will be emitted if the scheduler failed to schedule the pod. How about wrapping the internal error message in function, e.g. ? Users will know the pod failed because pod volumes not bound successfully. Sometimes, users can't just wait for the scheduler to recover from the error state, manual intervention might be required. The internal reason will be useful. In this scenario, .\n's suggestion sounds good to me"}
{"_id":"q-en-kubernetes-a42868a9b37565bd02ba94bbf4c6ffa3c24eec324dcbcc175becdaa7054bb2f9","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: The kubelet metrics related to , except for one, are missing from the kubelet metric endpoint. This applies to these metrics: What you expected to happen: I expect that the metrics mentioned above should be found in the endpoint on the kubelet. How to reproduce it (as minimally and precisely as possible): Run kubelet with a valid , and (defaults to true). or run the node e2e tests: Anything else we need to know?: All e2e tests related to DynamicKubeletConfig fails in master (and in v1.16): This bug/error was introduced in Environment: Kubernetes version (use ): && Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig node /cc logicalhan tallclair mtaufen\n/cc\n/assign WIP proof of concept solution to fix (make it a bit easier to understand the problem) here:\n/assign\nedit: I misunderstood the issue. discussed with offline, and he has an idea to restore previous behavior\nThat sounds perfect! He answered in the PR, so hopefully we can get this working again. I have wrapped my head around this issue for quite a few hours now, so I understand that it is kinda hard to understand by just reading this issue. Here is some more background info that hopefully makes sense.\n/priority important-soon /milestone v1.17\nLooks like the CI jobs are cleaning up now! ! Finally down to single digit number test failures, and away from the 30+ :smile:"}
{"_id":"q-en-kubernetes-747d699b97976a530bdb275e803e2be1bb68ac7168a6ff8b59806b000dbdf33f","text":"<!-- Please only use this template for submitting enhancement requests --What would you like to be : Change scheduler metric from Gauge to Histogram Why is this needed: is recorded for each preemption attempt. A histogram is better. /sig scheduling /help /good-first-issue\n/help /good-first-issue\nThis request has been marked as suitable for new contributors. Please ensure the request meets the requirements listed . If this request no longer meets these requirements, the label can be removed by commenting with the command. <detailsIn response to : /help /good-first-issue Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI can work on this issue.\n/assign\nI could work on this as well if doesn't have time to work on this.\nI am working on it, I will create the initial PR in less than a week. Sorry for the delay."}
{"_id":"q-en-kubernetes-f44ef0bcf2f9c8d0399d4efb9b080a7dd533e835c60e6b9400ac5a442e4901db","text":"Currently custom predicate and priorities can be created through the . The problem is that a custom predicate/priority, like ServiceAffinity, can be multiple times in the Policy with different names, possibly with different arguments. This semantic (i.e., specifying multiple instances of the same custom predicate with different arguments) is only useful if different s were specified because in all cases of custom predicates/priorities the arguments are merge-able (for example, from multiple CheckServiceAffinity custom predicates can be merged into one set). In the world of framework plugins, we don't currently have a way to maintain those semantics. to run. The question here: do we really want to maintain the old semantics of custom predicates and priorities? in other words, do we need to change the ComponentConfig API to somehow allow specifying multiple instances of the same Plugin with different arguments/weights, or should we consider that this is not supported? I always lean towards simplifying things, so my suggestion is to not support that, but I would like to know if there are actual deployments that rely on this. /sig scheduling /cc\n/cc\nWe discussed this in the sig meeting, and put the issue for discussion on the scheduler and kubernetes-dev mailing list, I believe this behavior is not intentional, and so we should add logic to fail the scheduler configuration when a policy includes more than one custom predicate/priority of the same type. /assign\n/assign\nany progress on this one?\nI haven't made any progress on this yet, but it's on my priority now\nIn migrating custom predicates to filter plugins , I am gonna ignore the custom predicate name and use the default name."}
{"_id":"q-en-kubernetes-3e344d7170961bb0b5ea1388aadba785b1fc66be8acebee6a3bdcfe8f657da4d","text":"<!-- Please only use this template for submitting reports about failing tests in Kubernetes CI jobs --Which jobs are failing: cloud-provider-openstack jobs Which test(s) are failing: hack/local-up- itself fails. Since when has it been failing: Recent change , checks for node ready by doing a grep of KUBELETHOST. But when HOSTNAMEOVERRIDE is set , its unable to find the ip and exits. Testgrid link: Reason for failure: Anything else we need to know:\nRepalce KUBELETHOST with HOSTNAMEOVERRIDE shoud fix this issue, I have created a pull request\nThanks. But what if HOSTNAME_OVERRIDE is not set , then again it would fail? I guess its optional parameter\ncc\nIf not set, the default value is 127.0.0.1, won't fail\nHi ramineni, nodpad is correct. HOSTNAME_OVERRIDE do have default value which is 127.0.0.1.\nThanks. Then grep HOSTNAME_OVERRIDE is ideal solution. That would work for me. Hopefully PR gets merged soon"}
{"_id":"q-en-kubernetes-047f808e04b31271a571d492a12584e78df3565487565320590578c7b2755139","text":"What would you like to be : A unit test that verifies the evaluation order of the nodes in the scheduler. Why is this needed: The scheduler evaluates the nodes in a specific order until it finds a pre-defined minimum number of feasible nodes. To give all nodes a fair chance of being evaluated, in each scheduling cycle, the scheduler starts evaluating from where it left off in the previous cycle. This is captured by the parameter in  We need to add a unit test that captures regressions related to this behavior to avoid bugs like /sig scheduling /priority important-soon /help\n/assign\n/remove-help"}
{"_id":"q-en-kubernetes-12f7047d8d873a7d6bac7023fc38517c350449071b3fe2b6d53f3ea14de5351f","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: EndpointSlice for masters was not created after enabling EndpointSlice on a pre-existing cluster. Only those EndpointSlices managed by kube-controller-manager were created. What you expected to happen: EndpointSlice for masters should be created after enabling EndpointSlice on a pre-existing cluster, the same as the EndpointSlices managed by kube-controller-manager. How to reproduce it (as minimally and precisely as possible): kubeadm to create a cluster: the EndpointSlice feature by editing the kube-apiserver, kube-controller-manager, kube-proxy manifest kube-apiserver and kube-controller-manager is restarted, the EndpointSlice in namespace will be missing. Anything else we need to know?: Environment: Kubernetes version (use ):Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools:Network plugin and version (if this is a network-related bug): Others:\n/triage unresolved Comment when the issue is assessed and confirmed. I am a bot run by vllry. ‍\n/remove-triage unresolved /assign"}
{"_id":"q-en-kubernetes-c62d28c0cab8ac8c425758491c93f526d0f109b8db3aef1e60198c10830f3c12","text":"What happened: Housekeeping sync loop finds an orphaned pod and tries to clean it up. The kubepod slice, as seen below, is not removed by with no indication as to why. This leads to the docker scope timing out and entering a failed state. The returned error(s) from runc are swallowed making debugging difficult. What you expected to happen: As per other areas of the sync loop, useful errors are logged at debug level due to the async calls. How to reproduce it (as minimally and precisely as possible): N/A Anything else we need to know?: N/A Environment: Kubernetes version (use ): master\n/sig node\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nHave you seen the comment above pcm.Destroy() call ? Did you have a chance to look at kubelet log to see if the above was true ?\n/remove-lifecycle stale\nThis would give us some clue about the error:\nyes, definitely looked! In this specific situation, the pods are never retried. What we see happen with these pods: ADD -RECONCILE -Started Volumes -Created slice -No sandbox for pod -Started libcontainer -ContainerStarted -ContainerStarted -RECONSILE -ContainerDied -RECONCILE -Unmounts -Killing unwanted pod -Orphaned removing pod cgroups -container kill failed because of 'container not found' or 'no such process' -cleanup: failed to unmount secrets -ContainerDied -Container garbage collection succeeded -SyncLoop (SYNC): 141 pods (including this one) -stopping timed out. Killing. -Stopped libcontainer -Removed slice -DELETE -REMOVE -err: pod not found Destroy all cgroups of the pod that should not be running, by first killing all the attached processes to these cgroups. We ignore errors thrown by the method, as the housekeeping loop would again try to delete these unwanted pod cgroups Despite running, the slice is not removed and we do not log or indicate why. Subsequently, the pod is NOT determined to be either an active pod or for some reason not in the cgroup list that is passed into kl.cleanupOrphanedPodCgroups(cgroupPods, activePods) during housekeeping loops. A more detailed look, which I believe is accurate ;): housekeepingCh -HandlePodCleanups -cgroupPods=UID+CGroupName=GetAllPodsFromCgroups -allPods=GetPodsAndMirrorPods -activePods=filterOutTerminatedPods -cleanupOrphanedPodCgroups I do not think desiredPods and runningPods are relevant here.. So either: a) the pod is not found and to allPods (seems unlikely) but we can't know for sure.. b) the pod is incorrectly determined to be active via activePods which again we can not determine.. Any pod that is not Failed, Succeeded or has a DeletionTimestamp set and a status of Terminated or Waiting - anything else returns false and is deemed active c) the pod cgroup is not found in m.subsystems.MountPoints and to cgroupPods - unlikely, but could be a bug but the root of this issue is the fact that all of the scopes and slices exist and keep building up until things start failing... GetAllPodsFromCgroups could use some logging... Maybe a V 5 log that prints out foundPods at the end... From there, they're passed into the cleanup: kl.cleanupOrphanedPodCgroups(cgroupPods, activePods) 1) All activePod UIDs are put into a podSet 2) A range loop of uid and cgroup name from cgroupPods looks to see if the uid from cgroupPods also exists in the podSet (active pods) 3) If the uid from cgroupPods exists in the podSet we skip out of the loop 4) If it doesn't, the pod is orphaned (not active w/ cgroups on the filesystem) and we call pcm.Destroy(val) on the cgroup name. Doesn't make sense why pcm.Destroy(val) ignores errors... it would be helpful! tryKillingCgroupProcesses logs any error and also returns the an error which is always swallowed.. we don't see any issues here More problematic, the call to cgroupManager.Destroy only returns an error which would be swallowed... Could probably use a similar level 3 log here like what you have for tryKillingCgroupProcesses because as you can see, were totally blind here...\nyour change is of a similar nature to what I proposed in my PR for this issue, just a level higher. I think there are a few areas noted above the could help further debug this.. Still haven't been able to get my PR merged with additional logging.\nThe following code was relatively new (compared to when you logged this issue):\nIndeed... I looked into it a while ago and those changes are new via: - Could definitely make a different if the issue is related to pod getting filtered out from activePods when they have not been fully cleaned up yet on the subsequent housekeeping syncloop runs.\nI think is promising, but I still think it makes sense to add some logging here. Do you want want to proceed with my open PR, your suggestion in or both?\nYou can proceed with your PR."}
{"_id":"q-en-kubernetes-2e980982f61b83c27c7ae4bdf180c05d465fe500056a9b182ab82110619c2947","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: It returns an error when I create a pod of yaml file over 1M. I checked the codes and found that the coded wrong. It should be “json.Unmarshal(p.patchBytes, &v)”, not “json.Unmarshal(p.patchBytes, v)” in the 344 and 368 line. What you expected to happen: It can be parsed correctly. How to reproduce it (as minimally and precisely as possible): Look at the code and you'll see or you can test it. Anything else we need to know?: no. Environment: Kubernetes version (use ): Community latest code Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/kind bug\n/sig api-machinery\n/assign"}
{"_id":"q-en-kubernetes-754626d269fd40f941348e993264e68393021cfd29fde0f04b37efaba8a1dae3","text":"What happened: While testing out the flag, I created a manifest with the following contents: I ran , and the resources appeared on my minikube cluster: Then, I removed the deployment from the manifest: And re-ran the same command, . To my surprise, however, the deployment wasn't removed; the only output from kubectl was . What you expected to happen: I expected the pruner to find the deployment with the matching label, , and prune it. How to reproduce it (as minimally and precisely as possible): I believe my example above is minimal and easily reproducible. Anything else we need to know?: This small change to the code for resolves the issue: Environment: Kubernetes version (use ): v1.16.3 Cloud provider or hardware configuration: N/A OS (e.g: ): N/A Kernel (e.g. ): N/A Install tools: N/A Network plugin and version (if this is a network-related bug): N/A Others: N/A\n/sig cli\n/assign cc"}
{"_id":"q-en-kubernetes-aaa23a6dfbac4923cc88d1230b36a82c16015c99fdfc56d40a0659c9be5002f8","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: When trying to use the webhookConversionServer, im getting . I was able to figure out that the unicode character displayed maps to port 443. see code What you expected to happen: If the service reference doesnt define 443, it should show an error that the service doesnt have the port number (whatever the number may be). How to reproduce it (as minimally and precisely as possible): your service (missing the default port 443) your CRD (one version no conversion webhook) your resource at this old version crd with new version and conversion webhook, and use new version for storage. to get the resource yaml, and see error Anything else we need to know?: current implementation will convert port number into a unicode character. ie. im seeing where . Must use instead of to maintain decimal. Environment: Kubernetes version (use ): v1.15.4+IKS Cloud provider or hardware configuration: IBM IKS OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig api-machinery\nAre you planning to send out a PR ?\ni was going to.. but didnt know which repo was relevant and didnt feel like reading through all contibution gorp.\nCreated a PR to fix the message output."}
{"_id":"q-en-kubernetes-de464a32cf340b19edfa7e83f192f3b1617eeff5aab3c9a4e1e526a7d46b9840","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Azure cloud provider is sending unexpected LoadBalancer update requests even if the LoadBalancer rules are already identical. The reason is GetIdleTimeout() returns nil when the annotation is not set While find rule doens't check whether it is nil or not. What you expected to happen: Azure cloud provider shouldn't send unexpected LoadBalancer updates. It is easily to get throttled if there're too many services running on the cluster. How to reproduce it (as minimally and precisely as possible): kubectl annotate service nginx testlabel=value Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others: /kind bug /area provider/azure /assign\n/sig cloud-provider"}
{"_id":"q-en-kubernetes-f4017862a33fbac1520d07447f039f12ce4e46548fea44bde023d16fb063d85c","text":"If you give a prerelease version to , it'll match both and . Based on this logic: We'll end up trying to pull binaries for this prerelease from the CI bucket () rather than the release bucket (). These pre-release versions are usually* pushed to the release bucket as part of the release process (e.g. ), and so should probably prefer these over the CI versions (which may or may not exist, depending on whether a commit was pushed to the release branch between when the tag was created and the build occurred). More explicitly, the current logic is: the release bucket by default. a version matches the CI regex, use the CI bucket. , if a version matches the release regex, we're good - continue to use the default (release bucket). , bail out - invalid version. The correct logic would probably be something like the release bucket by default. a version matches the release version regex, continue to use the release bucket. , if a version doesn't match the release version regex, but it matches the CI regex, use the CI bucket. it matches neither, bail out. Important note: due to our hotfix for , there is a somewhat fake prerelease tag () which doesn't have any bits in the release bucket, and we expect others soon. We should not fix this bug in until the latest pointers in the CI bucket do not point to valid release/prerelease versions.\n/sig testing /sig release I guess?\n/area release-eng /priority critical-urgent /assign /milestone v1.18\nHello With code freeze approaching for 1.18 (5 March), can you please check if the current issue is still intended for milestone 1.18?\n-- I'm pretty sure I can get to this before the cycle is out.\nif our current approach for tagging release branch pre-releases doesn't trigger this, this seems like it is not critical for the milestone, is that correct?\n-- Good point! /milestone v1.19 /remove-priority critical-urgent /priority important-soon\nGiven the priority dip on this one, I think it's fine for someone in Releng other than Tim or I to pick this up. /unassign -- This is a good one to work on. /help\nI can look that, after the release 1.18, if is that fine then can we have a short discussion about this one? /remove-help /assign\n-- SGTM"}
{"_id":"q-en-kubernetes-d229adfbea12aef3b49af22ccc6144a0f39cff69fa180ff4396be871aeac05e1","text":"What would you like to be : There are many useful scripts in \"\". For example, I often use and for creating simple k8s cluster quickly. However, there are only few explanation or nothing how to use it. I think it becomes more useful and helpful if some instruction or guidance is explained in each script. Why is this needed: It aims and expects to increase new k8s developers by describing how to use each script. My proposal Add the following description under the copyright for each script file. Overview of the script Simple usage Supplementary explanation(This is option) Other concern I would like to know the purpose of use about hack directory. (e.g. Where to put tools for developers(therefore there is no guidance?).)\nI don't know who is in charge of each scripts of hack, but if he agrees the idea of this feature request , I am going to do this task myself!!\n/sig testing\nI believe there is a script generating swagger/openapi specfilters under hack, but I am still struggling with it. It could be of great help if there is some how-to doc for each scripts.\nthis basis, why not create a file under the hack folder, because a lot of kk CIs are derived from multiple script commands under the hack. There are already 268+ files under the hack file. For contributors unfamiliar with these, these comments and instructions are helpful to them, and it is also helpful for new contributors to handle pr ci error reports.\n/cc What do you think\nI think it's good idea to create (It may be easy to understand if the script's explanation is described by category or use in ) Apart from that, it is helpful that the script file itself has a brief description.\nWhen we create , it may tough work if creating it including all scripts explanation at one time. (There are many scripts and I understand only some scripts now.) Regarding , current my idea is to add a description to each script little by little and update the at the same time. This means that will be completed little by little in conjunction with adding the description of each script. In the future, if someone create new script in hack, he also may have to add description of his script into\nSelf-documenting commands sound useful. A high level overview also sounds useful.\n/cc /cc Regarding adding help information, I suggest that we can submit pr in one or several files as a unit, so that we can track and statistics. For the documentation, I think it should be concise and not too long. How about this advice? We can claim tasks through , the pr title yes , i agree with you. [x] kubernetes/hack/benchmark- [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] kubernetes/hack/boilerplate/BUILD [ ] kubernetes/hack/boilerplate/test/BUILD [ ] [ ] [ ] [ ] [ ] kubernetes/hack/BUILD [x] kubernetes/hack/build- [x] kubernetes/hack/build- [x] [ ] kubernetes/hack/conformance/BUILD [ ] [x] kubernetes/hack/dev-build-and- [x] kubernetes/hack/dev-build-and- [x] kubernetes/hack/dev-push- [ ] kubernetes/hack/e2e-internal/BUILD [ ] kubernetes/hack/e2e-internal/e2e-cluster- [ ] kubernetes/hack/e2e-internal/e2e- [ ] kubernetes/hack/e2e-internal/e2e-grow- [ ] kubernetes/hack/e2e-internal/e2e-shrink- [ ] kubernetes/hack/e2e-internal/e2e- [ ] kubernetes/hack/e2e-internal/e2e- [x] kubernetes/hack/e2e-node- [ ] kubernetes/hack/generate- [x] kubernetes/hack/generate- [ ] kubernetes/hack/gen-swagger-doc/gen-swagger- [x] kubernetes/hack/get- [x] kubernetes/hack/ginkgo- [x] kubernetes/hack/grab- [x] kubernetes/hack/install- [ ] kubernetes/hack/jenkins/benchmark- [ ] kubernetes/hack/jenkins/OWNERS [ ] [ ] kubernetes/hack/jenkins/test- [ ] kubernetes/hack/jenkins/verify- [ ] kubernetes/hack/lib/BUILD [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [x] kubernetes/hack/lint- [x] kubernetes/hack/list-feature- [x] kubernetes/hack/local-up- [ ] kubernetes/hack/make-rules/BUILD [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make-rules/helpers/go2make/BUILD [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with- [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with- [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/skipme/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/skipme/subsubdir/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/skipme/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/subdir/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/testdata/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/testdata/subsubdir/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/testdata/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/underscore/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/underscore/subsubdir/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/underscore/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-with-gofiles/README [ ] kubernetes/hack/make-rules/helpers/go2make/testdata/dir-without-gofiles/README [ ] kubernetes/hack/make-rules/Makefile.manifest [ ] kubernetes/hack/make-rules/make- [ ] kubernetes/hack/make-rules/test- [ ] kubernetes/hack/make-rules/test-e2e- [ ] kubernetes/hack/make-rules/test-e2e- [x] kubernetes/hack/make-rules/test- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/make- [ ] kubernetes/hack/OWNERS [x] kubernetes/hack/pin- [x] kubernetes/hack/print-workspace- [x] kubernetes/hack/run-in- [ ] kubernetes/hack/testdata/auth- [x] [x] [x] kubernetes/hack/testdata/CRD/foo-- [x] kubernetes/hack/testdata/CRD/foo-deleted- [x] kubernetes/hack/testdata/CRD/foo-updated- [x] [x] kubernetes/hack/testdata/CRD/multi-crd-list-- [x] kubernetes/hack/testdata/CRD/multi-crd-list-deleted- [x] kubernetes/hack/testdata/CRD/multi-crd-list-updated- [x] kubernetes/hack/testdata/CRD/multi-crd- [x] [x] [x] kubernetes/hack/testdata/deployment-label- [x] kubernetes/hack/testdata/deployment-label- [x] kubernetes/hack/testdata/deployment-label- [x] kubernetes/hack/testdata/deployment-multicontainer- [x] kubernetes/hack/testdata/deployment- [x] kubernetes/hack/testdata/deployment- [x] kubernetes/hack/testdata/deployment- [x] kubernetes/hack/testdata/deployment-with- [x] [x] [x] [x] [x] [x] kubernetes/hack/testdata/filter/pod-apply- [x] kubernetes/hack/testdata/filter/pod-dont- [x] kubernetes/hack/testdata/frontend- [x] kubernetes/hack/testdata/frontend- [x] kubernetes/hack/testdata/invalid-rc-with-empty- [x] kubernetes/hack/testdata/kubernetes- [x] [x] [x] [x] [x] [ ] kubernetes/hack/testdata/multi-resource- [ ] kubernetes/hack/testdata/multi-resource-json- [ ] kubernetes/hack/testdata/multi-resource- [ ] kubernetes/hack/testdata/multi-resource-list- [ ] kubernetes/hack/testdata/multi-resource- [ ] kubernetes/hack/testdata/multi-resource-rclist- [ ] kubernetes/hack/testdata/multi-resource- [ ] kubernetes/hack/testdata/multi-resource-svclist- [x] kubernetes/hack/testdata/multi-resource-yaml- [x] kubernetes/hack/testdata/multi-resource- [x] kubernetes/hack/testdata/null-propagation/deployment- [x] kubernetes/hack/testdata/null-propagation/deployment- [x] kubernetes/hack/testdata/OWNERS [x] kubernetes/hack/testdata/pod- [x] kubernetes/hack/testdata/pod- [x] kubernetes/hack/testdata/pod-with-api- [x] kubernetes/hack/testdata/pod-with-large- [x] kubernetes/hack/testdata/pod-with- [x] [x] [x] [x] [x] kubernetes/hack/testdata/prune- [x] kubernetes/hack/testdata/prune- [x] kubernetes/hack/testdata/recursive/deployment/deployment/nginx- [x] [x] [x] [x] [x] kubernetes/hack/testdata/recursive/pod/pod/busybox- [x] [x] [x] kubernetes/hack/testdata/recursive/pod- [x] kubernetes/hack/testdata/recursive/pod-modify/pod/busybox- [x] kubernetes/hack/testdata/recursive/pod- [x] kubernetes/hack/testdata/recursive/pod- [x] [x] kubernetes/hack/testdata/recursive/rc/rc/busybox- [x] [x] [x] kubernetes/hack/testdata/redis-slave- [x] kubernetes/hack/testdata/retainKeys/deployment/deployment- [x] kubernetes/hack/testdata/retainKeys/deployment/deployment- [x] kubernetes/hack/testdata/rollingupdate-daemonset- [x] kubernetes/hack/testdata/rollingupdate- [x] kubernetes/hack/testdata/rollingupdate-statefulset- [x] kubernetes/hack/testdata/rollingupdate- [x] kubernetes/hack/testdata/scale-deploy-1.yaml [x] kubernetes/hack/testdata/scale-deploy-2.yaml [x] kubernetes/hack/testdata/scale-deploy-3.yaml [x] [x] kubernetes/hack/testdata/service- [x] kubernetes/hack/testdata/service- [x] kubernetes/hack/testdata/sorted-pods/sorted- [x] kubernetes/hack/testdata/sorted-pods/sorted- [x] kubernetes/hack/testdata/sorted-pods/sorted- [ ] [ ] [x] kubernetes/hack/test- [ ] kubernetes/hack/test- [x] kubernetes/hack/update- [ ] kubernetes/hack/update- [x] kubernetes/hack/update- [ ] kubernetes/hack/update-generated-api-compatibility- [x] kubernetes/hack/update-generated-device-plugin- [ ] kubernetes/hack/update-generated-device- [ ] kubernetes/hack/update-generated- [ ] kubernetes/hack/update-generated-kms- [ ] kubernetes/hack/update-generated- [x] kubernetes/hack/update-generated-kubelet-plugin-registration- [ ] kubernetes/hack/update-generated-kubelet-plugin- [x] kubernetes/hack/update-generated-pod-resources- [ ] kubernetes/hack/update-generated-pod- [ ] kubernetes/hack/update-generated-protobuf- [ ] kubernetes/hack/update-generated- [ ] kubernetes/hack/update-generated-runtime- [ ] kubernetes/hack/update-generated- [ ] kubernetes/hack/update-generated-swagger- [x] kubernetes/hack/update- [x] kubernetes/hack/update-import- [ ] kubernetes/hack/update-openapi- [x] kubernetes/hack/update- [ ] kubernetes/hack/update-vendor- [ ] kubernetes/hack/update- [ ] kubernetes/hack/update-workspace- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-api- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-cli- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-conformance- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-external-dependencies- [ ] kubernetes/hack/verify-flags/excluded- [x] kubernetes/hack/verify-flags/OWNERS [x] kubernetes/hack/OWNERS [ ] kubernetes/hack/verify-flags- [x] kubernetes/hack/verify-generated-device- [x] kubernetes/hack/verify-generated- [x] kubernetes/hack/verify-generated-files- [x] kubernetes/hack/verify-generated- [x] kubernetes/hack/verify-generated- [x] kubernetes/hack/verify-generated-kubelet-plugin- [x] kubernetes/hack/verify-generated-pod- [x] kubernetes/hack/verify-generated- [x] kubernetes/hack/verify-generated- [x] kubernetes/hack/verify-generated-swagger- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-import- [x] kubernetes/hack/verify-import- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-no-vendor- [x] kubernetes/hack/verify-openapi- [x] kubernetes/hack/verify-pkg- [x] kubernetes/hack/verify-publishing- [x] kubernetes/hack/verify-readonly- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-staging-meta- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-test- [x] kubernetes/hack/verify-test- [x] kubernetes/hack/verify-test- [x] kubernetes/hack/verify-typecheck- [x] kubernetes/hack/verify- [x] kubernetes/hack/verify-vendor- [x] kubernetes/hack/verify- [x]\ni will pick up , i will do it step by step.\nI agree with your idea, and thanks the checkbox to check progress. there are many scripts, so let's work together!\n/assign /assign\nI will pick up hack/verify-\nI will pick up hack/verify-\nI will pick up hack/verify- (all of verify scripts). But there are large number of scripts(over 40), I want to split it into two or three PRs.\nI will pick up hack/update-\nI will pick up these scripts. kubernetes/hack/benchmark- kubernetes/hack/build- kubernetes/hack/build- kubernetes/hack/dev-build-and- kubernetes/hack/dev-build-and- kubernetes/hack/dev-push- kubernetes/hack/e2e-node- kubernetes/hack/generate-\nI will pick up these scripts get- ginkgo- grab- install- lint- list-feature- local-up- pin- print-workspace- run-in- test- test-\nHow about this now? /help Marking this tag so that others can easily access it.\nThis request has been marked as needing help from a contributor. Please ensure the request meets the requirements listed . If this request no longer meets these requirements, the label can be removed by commenting with the command. <detailsIn response to : How about this now? /help Marking this tag so that others can easily access it. Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI think we've finished adding simple explanations to the almost scripts. It seems that several scripts of update- does not complete yet. I think we may able to close after that work.\nI will pick up these scripts:\n/assign"}
{"_id":"q-en-kubernetes-21f6b622d037a3471bbf2d53c2f298df7029b0752f80faeb870e98f01c0e5129","text":"In a few utility function in current test/e2e/storage/utils, we found that the err variable gets shadowed. Because of this, it might get nil pointer error. See code\nfixed by"}
{"_id":"q-en-kubernetes-22956ba2135ab6f3796239a359e3855984bd4cb916049da6b7ec54d0501cca5c","text":"What would you like to be : An integration test for NodeResourceLimits plugin. Why is this needed: No end-to-end test coverage current exist for this alpha feature. /sig scheduling /help\n/assign /remove-help"}
{"_id":"q-en-kubernetes-d1fe3cf19c1c7d9bc5268ad985f5c595dbb9bea2e0f983d49a1ad7b835b3259e","text":"When trying to see why pods don't start using either kubectl get events or kubectl describe pod , the base message is not detailed enough. Warning FailedScheduling 5s (x4 over 65s) default-scheduler 0/1 nodes are available: 1 node(s) had taints that the pod didn't tolerate. What would you like to be : Details of the taint and details of what it didn't tolerate Why is this needed: To get much much better debug information by default when trying to understand why pods can't launch.\n/sig cli\n/sig scheduling /priority important-longterm\n/cc\n/remove-sig cli\nThose labels are not set on the issue: <detailsIn response to : /remove-sig cli </details\n/good-first-issue\n/assign starizard\nCan I get an idea of what this would involve? I might be able to put some work on it.\nI assume is taking this already. The error message is produced in We probably also require changes in to get the name of the missing toleration.\n/remove-help"}
{"_id":"q-en-kubernetes-6bd3b0b5b88d2c4e52fa7e7988f3bc34e4f6c96dcec415eb9240194b302773b8","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: I am testing API Priority and Fairness. I launched it like this: After the logging said the cluster is running and told me how to set KUBECONFIG, I did so in another shell and went looking for FlowSchema objects. But kubectl erroneously told me there is no such thing. However, looking in the kube-apiserver log shows flowschemas and prioritylevelconfigurations being read and written; for example: Also, I am able to read them with against the insecure port. Interestingly, the kube-apiserver log also includes this complaint: I will attach the kube-apiserver log. What you expected to happen: I expected to list them. How to reproduce it (as minimally and precisely as possible): I think the report above is pretty minimal. Anything else we need to know?: Environment: Kubernetes version (use ): this is the branch, git commit 916edd922e528f2f27542d3ca5967ca11fe153f8Cloud provider or hardware configuration: bare metal, intel, 16 \"CPUs\", 64 GB of memory OS (e.g: ): 18.04.2 LTS (Bionic Beaver) Kernel (e.g. ): Linux mjs-dev-1 4.15.0-74-generic -Ubuntu SMP Thu Dec 19 08:06:28 UTC 2019 x8664 x8664 x86_64 GNU/Linux Install tools: hack/local-up- Network plugin and version (if this is a network-related bug): Others: Docker version 18.09.7 go version go1.13.4 linux/amd64\n/sig api-machinery\n/cc /cc /cc /cc\n/assign\nthe \"\" API group is not included in the list of known groups to register at startup: That means resources inside that group will not appear in discovery, and clients like kubectl will not know how to find them.\nYou can see all resources visible via discovery by running\njust as said, for now we have to manually create an apiservice into the cluster to get kubectl work. i dont think we're adding this to the list until the beta stage, can we make an exception for the case?\nIt is fine to add alpha things to the list I linked so they get created automatically if needed. They only get registered if the relevant runtime-config is enabled.\nInterrupts control plane upgrade in lazy scenario where it's assumed dns name isn't changed properly to active node or or single node control plane\noh cool actually it succeeds on the second main node but the kube apiserver pod doesnt start\nthe hacky way being to adjust that api creation based on this?\nAnd I removed the flags too\nWait what is this? This was to my api manifest during kubeadm upgrade?"}
{"_id":"q-en-kubernetes-31635c902aff9f8a2a2d4b7ec7f766a933cc4b8a93f6aa67e00afd6d1cd1a418","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: With the exec credentials plugin configured in my kubeconf, and a stale/expired token present in the same file issuing e.g. triggers the authentication flow as handled by the plugin - once the flow is returned to kubectl (by printing the ExecCredential object) the new token is stored to the kubeconfig but not immediately used to resolve the request, leading to: Immediately issuing again works as kubectl now seem to use the credentials (token) stored following the first request. What you expected to happen: That kubectl would read the new token as fetched by the exec credential plugin and use it already for the first command issued. How to reproduce it (as minimally and precisely as possible): the kubernetes API server with the OIDC authenticator enabled. kubectl to use an exec credential plugin printing the token received out of band. that the token was indeed written to the kubeconf file. that the first kubectl command failed. that the next identical command succeeds. Anything else we need to know?: This works as expected when there is no token previously stored in the kubeconf - the behavior is only triggered when there is a token stored but it expired, or for whatever other reason not accepted by the API server. Environment: Kubernetes version (use ): 1.14, also tried with 1.17 Cloud provider or hardware configuration: OS (e.g: ): Mac OS for kubectl client Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig auth /sig cli\ncould you provide more details around how is configured. For example, it is unclear to me if you are using the built in oidc plugin or a custom exec plugin. I think you are using a custom exec plugin but its also trying to write tokens to the file (which it should not do)? /triage needs-information\nThanks for looking into it . You are indeed correct in your assumption that I'm using a which stores the token in the after retrieval. I'd love to learn why this should not be done, but the problem remains also without persisting the token after retrieval. Basically the steps to reproduce could be reduced to these: containing an expired token. same configuring an exec plugin which retrieves a valid token out of band and echoes it. Expected behavior: The valid token retrieved is used for the subsequent authentication. Possibly expected behavior: The exec plugin flow not started (or exited with a warning) due to an expired token being present in the . Current behavior seems to acknowledge the invalid token and hence tries to retrieve a new, valid one - which it then discards as soon as it arrives.\nSeems we can reduce it even further by removing the token altogether from the file and instead provide it directly. So the only step needed to reproduce is essentially this: configuring an exec plugin which retrieves a valid token out of band and echoes it as part of an . Now any will trigger the exec plugin to retrieve a valid token, but will still not use it for subsequent authentication.\nThis one is caused by config merging with clientcmd overrides: and exec.Auth not being in this switch:\nInteresting, thanks Will add it to the switch along with some tests and if everything works out well I'll submit a PR next week.\nHaving spent some time on this in the debugger, it seems somewhat more intricate than I first thought. Turns out the exec plugin is executed whether the presented is invalid or valid - the exec action is actually called before the first call is even dispatched. This due to the exec credential plugin possibly being needed to setup the client certificate requested for the transport. Once the exec plugin's is called it assign the authenticators function to : ..which is then called when setting up the transport, thus triggering the exec plugin regardless of a token being present or not: Since the exec plugin has no way of knowing whether a token or a client certificate will be returned by the exec action, this kind of makes sense. It however defeats letting any present to override the exec plugin, which clearly seems to be the intent: Possible options here that I can think of: (and fix) behavior of letting override the exec plugin. This looks doable by adding a check for token auth early in the 's method body: But might be there are consequences to this I'm not aware of. the behavior of letting override the exec plugin being called. third, more complex option, would be to keep the token override but dispatch to the exec credential plugin should the response from the first call with the token come back with a 401. This was what I originally believed was the intent as my first tests always included an invalid token. While this behavior seems nice I'm more and more thinking it may violate the principle of least surprise. Either you present a token and use that for auth or you let the exec plugin fetch the credentials for you. What do you guys think?\nAs a colleague just made me aware of, certain tools like kubetail seem to have kubectl call the exec plugin so many times that eventually the threshold of 1000 is reached and warn logs (exec:201) start spewing - even though no certificate (and hence no rotation) should be involved. This has me more convinced that the first option is the right one, skipping the client TLS ceremonies should a token be present for authentication.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale"}
{"_id":"q-en-kubernetes-1374383302c8b2f5e1d9227ad3e6a5cd40003fd3af4f018f460ca706631944c5","text":"What happened: Used kube- to bring up a GCE cluster and then logged on to the master. Run kubectl on the master, is hangs for a while (30 seconds) and then returns an error. $ kubectl get nodes Error from server (InternalError): an error on the server (\"\") has prevented the request from succeeding What you expected to happen: I expected to get the node list. If I use one of the system account's kubeconfig files it works. $ sudo kubectl --kubeconfig=/mnt/disks/master-pd/srv/kubernetes/kube-bootstrap/kubeconfig get nodes NAME STATUS ROLES AGE VERSION kubernetes-master Ready,SchedulingDisabled <none26m v1.18.0-alpha.2.55+-dirty kubernetes-minion-group-jfk6 Ready <none26m v1.18.0-alpha.2.55+-dirty kubernetes-minion-group-mqnp Ready <none26m v1.18.0-alpha.2.55+-dirty kubernetes-minion-group-x540 Ready <none26m v1.18.0-alpha.2.55+-dirty How to reproduce it (as minimally and precisely as possible): Bring up a GCE cluster from head, ssh to master and run \"kubectl get nodes\" Anything else we need to know?: Environment: Kubernetes version (use ): Client Version: version.Info{Major:\"1\", Minor:\"18+\", GitVersion:\"v1.18.0-alpha.2.55+-dirty\", GitCommit:\"\", GitTreeState:\"dirty\", BuildDate:\"2020-01-23T00:38:39Z\", GoVersion:\"go1.13.6\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"18+\", GitVersion:\"v1.18.0-alpha.2.55+-dirty\", GitCommit:\"\", GitTreeState:\"dirty\", BuildDate:\"2020-01-23T00:36:29Z\", GoVersion:\"go1.13.6\", Compiler:\"gc\", Platform:\"linux/amd64\"} Cloud provider or hardware configuration: GCE OS (e.g: ): COS m_77 Kernel (e.g. ): 4.19.76 Install tools: cluster/kube- Network plugin and version (if this is a network-related bug): Others: I believe this used to work based on talking to KAS on the insecure port using no authentication. The insecure port is now turned off and we require authentication materials. We should probably setup an admin account specifically for the master.\n/sig cli\n/assign\nI also would not have expected a hang, but an immediate error, and I would have expected an authentication error, not an internal error\nI believe the internal error is due to it defaulting to port 8080. Nothing is listening on that port. This used to work but I would guess it stopped working when we stopped listening on the insecure port.\nchanged behavior of client-go on connection refused errors, which changed a short wait and an \"invalid host/port\" error into a long retry wait and a confusing \"internal server\" error /assign\nif kubectl is intended to work automatically against the local API server, then a kubeconfig file should be set up, but the 30 second delay and incorrect error are a bug that should be fixed in client-go /sig api-machinery\n/remove-sig api-machinery /unassign actually, moved the client-go issue to will keep this one for the GCE $KUBECONFIG resolution\nI agree that seems like the right fix.\nI'm not sure this belong to sig-cli... looks like made this change /sig cluster-lifecycle /cc\nRe: set up a kubeconfig That's what we do for GKE: create an authenticated kubeconfig and set the env var. E.g. if you have an authenticated kubeconfig at :\nYes, GKE sets up an admin system account for the master(s) as part of the internal-configure-helper. I believe we should do the same thing for GCE.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale"}
{"_id":"q-en-kubernetes-143b0f1aae2935f6e5093d1fc3fbf7fef6f1f25a0d6c2f039aa34d7fd3e325f8","text":"What happened: When running external CSI Driver tests with the Kubernetes E2E: We seem to be getting the following error. From initial investigation it looks like this happens in the during check but only when the test is being skipped. Hunch is that maybe since the test is skipped some resources are never created or they are cleaned up already and they are nil when the is called and expecting them. Kind of low priority since it doesn't make any tests actually fail. /cc /sig storage\nsee\nI think should address this"}
{"_id":"q-en-kubernetes-915370850f922c52baff913af6698146cfcbf304105ee540e15c6286e1599323","text":"Not sure if this is a bug, feature requests, or a documentation issue. What happened: I am trying to apply some configs with --prune, and was surprised by the behavior. I have a config file with objects in multiple namespaces, and apply with (note: no flag). By kube context was set to a namespace that was not in the manifests. Here is what I found: If my context is set to a random namespace, nothing is pruned If its set to namespace X, only objects in namespace X are applied If I pass a flag I can control which namespace gets pruned, but that will break the apply because the file has multiple namespaces in it What you expected to happen: I expect some way to be able to apply a config file with objects in multiple namespace and having pruning across all namespaces. Seems there are a few options: all namespaces referenced in the file, prune these across all namespaces warn about why we are not pruning due to namespace mismatch How to reproduce it (as minimally and precisely as possible): Run the script from namespace default, a, and b. You will see different behavior on each run. Related? Environment: Kubernetes version (use ): 1.17.0 Others: /sig cli\nI encountered this issue. I think has the bug that o.namespace gets \"default\" if no namespace is given. Then only a single namespace is visited with or without namespace. If my observation is correct, it changes the bahaviour drastically. In my work, pruning just stops working."}
{"_id":"q-en-kubernetes-0f75b4e031b6dedf26dcd668791c9b4bd7f0f9329b05e0f830ed2db15c090fe1","text":"What happened: Using leaderelection with ConfigMapLock, I get a panic when I fail to renew the lease. What you expected to happen: No crash How to reproduce it (as minimally and precisely as possible): Create a leaderelection setup with ReleaseOnCancel=true Run the program, allow a lock to be acquired. Change RBAC permissions to block configmap access Eventually we fail to acquire the lock and run into the panic above\n/sig api-machinery\n/assign\nlet me know if you need help getting more info, I am not sure if its easy to reproduce or not. I can add some extra debug logging, etc\nI will see if I can reproduce via unit test.\n/cc\nCan you check the log around the time of crash to see if there was the following ? Checking the events around the time of crash would also help: thanks\nI have the log snippet at the top, there is a\nI can also hit race conditions from a unit test. I can make this a new issue if preferred. and similar but slightly different:\nThere is some race around accessing observedRecord. We can use a Lock to protect the access. Please log a new issue. Thanks"}
{"_id":"q-en-kubernetes-5e5b950d571506f3e848bbc335531428e82786cb9d209d69a84a78892574f681","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Kubelet dies with a panic exception if CSI plugin returns NodeId == \"\" and nil error. Exception log see at the end of the message What you expected to happen: Kubelet should just raise an error to the plugin and ignore the registration. How to reproduce it (as minimally and precisely as possible): Create a CSI plugin with only Node that would implement GetNodeInfo as follows: Register it as a DaemonSet and see kublet crush Anything else we need to know?: Environment: Kubernetes version (use ): 1.14.10-gke.17 Cloud provider or hardware configuration: GKE OS (e.g: ): Container-Optimized OS (cos) Kernel (e.g. ): - Install tools: Network plugin and version (if this is a network-related bug): Others: Related exception Related line of code: - no check whether error is not nil before calling Error() on it.\n/sig storage\n/sig node is this the version of the kubelet too? 14 is out of support. is the issue reproducible with a 1.15+ kubelet?\n/assign\nI checked the code - same defect still exists in current codebase.\nthanks /priority important-soon\nWow, that was fast. Thanks!"}
{"_id":"q-en-kubernetes-99d5f10dcf25cc2fbe358f74e31ff561469628287156fd4c3b1b75db0492fdf9","text":"Which jobs are failing: Since when has it been failing: Testgrid link: Reason for failure: Anything else we need to know: /milestone v1.18 /priority critical-urgent /cc /sig cluster-lifecycle /area provider/gcp\n/reopen Reopening, to observe the job.\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nSeveral of the recent failures appear to be this: cc\nnext run should pick up the change in we have had boskos outage clearing up now.\nThis is still not going to work because k/k has removed the .deb files we've been downloading from the CI artifacts:\nFYI we are discussing in Slack how to proceed to resolve this. I doubt it will be a simple, quick fix, though.\nme and did mention this on the k-dev: !msg/kubernetes-sig-cluster-lifecycle/q934Ah3SD64/JukDA9wsAgAJ given there were no replies we assumed there are no users. but this is definitely for the better as we removed the Bazel build of packages from k/k. 's PR seems to use binaries directly: i will have a look.\nClosed via There is a possibility we will publish debs again in the future (via ), but while that tool is in flux, I don't want anyone depending on it. Thanks for the fix, /close\nClosing this issue. <detailsIn response to : Closed via /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-23453cfe9279d45b88b1b76bf59719d6724b15443ac3ffc3c4eacd0d07ffeb70","text":"Problem: Preferred pod/node affinity and preferred PodTopologySpread may not be respected because of conflicts with other scoring functions, like node utilization or image locality. Scheduler Scoring functions have weights, by default they are all set to 1 (apart from one): Having equal weights is problematic because of the way we score: each Scoring plugin returns a score [0, MaxScore], which get multiplied by the configured weight of the plugin, and a node score is the sum over all scores. Therefore, even if the user explicitly indicates that they prefer to place pod x on node y, this preference may not be respected and the scheduler may decide to place the pod on node z because it has lower usage and the container that the pod is requesting. Proposed Solution: Scoring functions that allow users to explicitly express preference should have higher weights than generic scoring functions. This includes: inter-pod affinity, node affinity, taint/toleration and pod topology spread. This issue is proposing to change the default behavior of the scheduler, I am not sure how k8s deprecation policy applies here, although I would argue that this change should help the scheduler make decisions closer to users expectations. We will start with pod topology pod spread since as of creating this issue, the feature is still in beta. /sig scheduling\nQuote from the : Rule : Deprecated behaviors must function for no less than 1 year after their announced deprecation. I think the rule doesn't apply to this proposal since it is not impacting the correctness of applications or impact administration of the cluster.\nAs nodepreferavoidpods already exercised to set its weight to 1000 by default, it sounds fair to me to increase the default weight for inter-pod affinity, node affinity, taint/toleration and pod topology spread. BTW: to some extent, we can use e2e tests verifying the to-proposed default weight. Right now we have some e2e tests covering priorities (scoring), but they depend on to ensure cpu/mem are at the same ratio of usage: and hence IMO a bit fragile.\nStrongly agree with this change. But, IMO, we may need to reconsider the default impact weights of each scoring plugin. Maybe they should have more weight ladders, rather than simply 1 or 1000. For example, from practical experience, in most cases should not be set to the same weight as , , etc. (Yes, we need to consider this for users, because most users may not notice this change).\nIMO a reasonable criterion to guide which Score plugins should have non-default weight is: whether the Score plugin is associated with an API (e.g., PodTopologySpread Score plugin is associated with .\nIf we give a higher weight only to the API-facing scores, then workloads that don't use these features are not affected, which is a good thing. I think a good weight to give to these scores is one that avoids it to be diluted among the rest of the scores. So, maybe a weight that is half the sum of all non-API-facing scores. Now, if a user uses the API for more than one score, I think we can safely assume that they want equal weights for them.\n/cc Daniel, can you comment on whether or not the proposed change conforms with k8s deprecation policy?\nI hate to say it, but I think you have to make the new set of weights an option and use the deprecation process to move people over. It's quite difficult to predict what a change in these weights might do to the overall scheduling behavior, no? If it was already possible to specify the settings (via flag or config file), then maybe there's a case to be made that the deprecation process is heavier than needed here. To make that case, you'd need to test and be VERY sure that the new settings are actually an improvement. The change sounds like an improvement, but I'd like to see that quantified / tested somehow. Perhaps even more importantly, I'd like to know the conditions under which it makes things worse, and along which dimensions. can give a second opinion if you don't like this answer :)\nThis may have an impact on availability (since by default the scheduler will prefer less under utilized nodes) and pod startup (less preference to node that already has the image) IF the pod have explicit preference for being on a specific node or with a specific pod. Those weights are actually configurable via the component config file, but most cloud providers don't allow customers to provide a custom configuration, they just use the default. Regarding the change of behavior, this effectively create two level scoring: first level is by user-specified preferences (affinity), second level is by workload-agnostic ones (e.g., image existence). If users are not specifying preferences, then this change have no impact.\nThose are good points, but how many users specify preferences? Should we encourage those users to use non-optional constraints, instead? I can imagine arguing that making the preferences more likely to be obeyed makes it rarer and therefore more surprising and worse when they can't be honored.\nI'm not sure if I'm understanding this correctly. Examples of non-optional scores are \"topology spreading\", \"resources utilization\" and \"image locality\". Users (aka Pods) can not opt-in or opt-out from them. Cluster operators (on prem or cloud-providers) can opt-out or change weights, but for a whole cluster. Remember that we are talking about node scoring. The nodes to be scored should have already passed the filters. Let's take an API-facing score (node affinity) and non-API-facing score (image locality) for illustrating the effect of weights: Imagine a pod is using \"preferred\" node affinity and there are no nodes available that satisfy that affinity. That means that in this case all nodes would get a low (or zero) score for node affinity. In this case the score given by image locality will provide the difference in score between all nodes. Now, let's say some nodes satisfy the node affinity. Those nodes will have the same high score for node affinity. But they will defer in image locality, which will be (in this case) the determining factor to select a node. However, in real (default) clusters we have more than 2 scores. And the problem is that pod preferences end up being diluted with all the other default scores summed up. What the proposal would achieve is that we first honor pod preferences and then we differentiate among the resulting nodes using the default scores.\nOK, so I guess the case where this hurts users is: They specified preferences The preferences are actually bad (plot twist) they're diluted so much that they're not having much effect but now the scheduler starts honoring them more often But it's still probabilistic, so we can expect some percentage of the time, the preferences continue to not get honored? Numerically, suppose 50% have preferences, 2% of those got them wrong, 80% of those are currently diluted (i.e., not already broken), and 10% of those are in environments where they'll suddenly get honored 100% after this change (becoming broken). That's .5.02.8*.1 = .08% of users, or 1 out of 1250 users this change could break. That's still maybe a little high for my taste, but maybe you have an alternative estimate?\nI think we can tweak how many of the currently diluted cases become honored. In I proposed that we use a weight of half the sum of the non-API-facing weights. We can be less aggressive and do a quarter or just 2 (we currently put weight of 1 for all).\nOn another note, I would say weights are actually part of the component config API. In both v1alpha1 and the new v1alpha2 (), you can tweak the weights of individual scores. Users might be setting up these weights assuming that the defaults are 1. This is not a problem with the old Policy API, because you have to specify all the scores with their weights. So, if we decide to change the weights, better do so now for v1alpha2.\nDaniel, I don't expect that 50% of k8s workloads have preferred pod/node affinities, but I don't have data to suggest an alternative estimate. I guess we could make the change behind a feature flag that gets graduated over multiple releases.\nIt is permissible for the default to change over versions, in fact that's the way to do it for API types. Since the type is alpha, I guess it's OK to just do this. So, if the config file is v1alpha1, do the old defaults; if it's v1alpha2, use the new ones. How does that sound?\nsgtm. But the question of what's a good weight for api-facing scores still remains. By default we only have 4 non-api-facing scores: NodeResourcesBalancedAllocation ImageLocality NodeResourcesLeastAllocated DefaultPodTopologySpread So my two proposals are actually the same number: 2 :sweat:\nWhat about the case where a config file is not provided at all (which is very common across all cloud providers)? I guess we will have to keep the old defaults, and so we didn't actually solve the issue, right? it will also be confusing to users who pass a config file that doesn't change any of the weights.\nIf there's no config file at all, I think it's fair game, users can easily get the old behavior if needed. Others might disagree with me, but that's what I think. A completely separate thought: are you sure changing the weights is good? What about changing the way they're evaluated? E.g. in ML it's common to use e.g. a softmax function at the last layer to force a NN to make a choice / assign a consistent set of probabilities to each option. I think the way that might look here is instead of using F(N) as F's priority for node N, use exp(F(N)) / SUM (N0-Nn) { exp(F(N)) }. This would have the effect of magnifying minor preferences. Basically, I'm claiming that priority functions should be distributing the same amount of \"probability mass\" over the N possible nodes. It's not fair for one function to say \".3\" for everything while another says \".6\" for everything. You maybe need to normalize this before applying the weighting function anyway. But if you do the softmax, you might not even need the weight, as priority functions that give each node the same score will basically be no-ops, letting ones with differing opinions win.\nInteresting proposal, but don't you think weights will still be needed because unlike NN, the scheduler didn't go through a training phase to discover which features (in this case priority functions) provide the most important signal?\nLet's put an scenario with three simplified scores: selected: a score that prefers a particular set of nodes (maybe a particular zone). : existence of an image. : low utilization Now let's assume they give the following scores: Notice that they all have the same distribution (so they all give good signal). Even when using softmax the sum of scores (with the same weight for every score) will be: Nodes 1, 4 and 7 have the same score and thus the same probability to be chosen. I would argue that the user intention was that node 4 would be selected. It's still called a preference because if would give 0 to all nodes, then the pod can still go to the nodes, and the automatic scores take all the decision. Now, in the default scheduler, we have 4 automatic scores, so the user preference is quite unlikely to get respected. It's not necessarily a matter of good signal from the score.\nYup good points-- I agree a weight is still needed. If all the numbers are 0 or 1 (I had thought that it was a floating point and not a binary signal) then I don't think softmax is going to be different than a linear normalization (?). I guess it's debatable whether one wants a priority function that selects fewer nodes to have its opinion on those nodes be weighted more highly, which is what normalization would do. Anyway sorry if this is a rabbit hole. :) I think this idea would make more sense if the priority functions could express a strength for their preference and not just yes/no.\nIt's not binary necessarily... I was over simplifying. Node affinity and image locality certainly are, but node utilization is not.\nIt might be worth some experiments, then... if you have some way of telling whether or not it's actually better? Aldo Culquicondor <:\nWe have an HA use case that involves spreading pods across nodes and zone with preferred scheduling since we'd rather schedule the pods than not. Before 1.18, we've been using the something like the following: Now with the 1.18 beta, we tried the following: The later yields better scheduling results than the former on a 9 nodes multi-zone cluster with 3 nodes in each node. So while was better it surprisingly didn't yield a perfect spread. So I decided to combine both and ... boom ... we had a perfect pod spread for our deployment with 9 replicas. I'm not sure how these results fit into the discussions so far but I would have expected either option by itself to yield a better spread. Combining the two lets me reinforce my desires to the scheduler. IMO, I would like to see the scheduler give my explicit request more weight.\nthanks a lot for sharing. Giving more weight to explicit requests is what we are proposing. However, I'm a little surprised that you need these constraints. The scheduler already spreads by default when using services, replica set or stateful sets. Are these pods not members of any of those? I assume you are not modifying the scheduler configuration? But, of course, there are other scores in play. It makes sense that spreading constraints are more effective than antiaffinity, because there is a lot of antiaffinity with the node that is ultimately selected as well. Adding both specs works too. However, this increases scheduling latency, as we have to calculate more things. The main motivation for this issue was actually to make scoring more effective for the opposite case, which is not done by default: \"packing\". But your use case is just as valid.\nThese pods are part of a deployment. I tried without any scheduling preferences and the results are similar to podAntiAffinity preferred scheduling. Here is the baseline deployment yaml used for the tests: I have not modified the scheduler configuration.\nYes, deployments are affected because they create a replica set that owns the pods. But there are always other scores at play that might also be important, which is why a \"preferred\" spreading is never going to be perfect. This is kind of the point that was trying to make: guarding users from over-fitting and degrading the clusters in some way.\nWhat if we reduce the scope to just PodTopologySpreading, which is beta since 1.18? We could give it a weight of 2 now, before going to GA.\nAldo, one thing to take into account is the new default spread logic which is based on default pod topology spread constraints. For default spread constraints, I think the weight shouldn't increase, so things will get a little complicated with how we treat this case.\nGiven that default spread constraints is explicitly specified by users in , maybe we can assume they are already aware of adjusting the \"Weight\" in ? While for regular users, they won't touch , so a default non-1 weight is a no-op for them.\nthe default constraints (which we plan to introduce to replace the current defaultpodspreading score plugin) are not explicitly specified in PluginConfig.\nMaybe I missed something, are we talking about the same default constraints defined here?\nyes we are, but isn't the plan to have a default value for DefaultConstraints when not set: two constraints, one for zone and one for hostname, which mimic the current defaultpodspread behaviour and allows us to deprecate it?\nI think the only way around that is to introduce a weight for default constraints in the plugin config. And then set it to 50% for the k8s default.\nhmm, this will be a confusing configuration parameter to users. Another approach to consider is to bake it into the TopologySpread plugin instantiation without needing to pass it through PluginConfig (similar to what we did for volumebinder in the past: ).\nurgh, this will be tricky to do since defaulting will be done via component config, and so we don't have a way to communicate whether the values were defaults or explicitly set by the user. If defaulting is done in then it might be possible, but we shouldn't do that.\nDefaulting is going to happen in in the near future\nWe can have it in the internal type, but not the external one then. This will hide it from the users. But how do we communicate that to conversion logic?!\nIt doesn't seem there is a way because defaulting happens before conversion...\nTo make the new parameter semantically clear, we can treat it as a weight override rather than a percentage of the plugin weight, and so it's default would be 1 rather than 50%, internally the plugin computes the ratio as the override weight divided by the plugin weight, which it uses to scale scores computed for pods that don't have topology spread specified in their spec.\nGot it now. I thought the default constraints would default to nil if not specified by users. I think we should introduce a field called \"multiplexer\" and default to 1, so it's friendly to defaulting and conversion.\n+1.\nWhat if we start giving meaning to MaxSkew for scoring? If you set a low MaxSkew, you probably want it to have a higher effect than if you set a bigger value. We can discuss the exact formula, but I think adding these semantics make sense. On Sat., Apr. 18, 2020, 5:50 p.m. Wei Huang, <:\nThat's possible. Current Score function doesn't take into consideration - it's just following a simple \"the less number of matching Pods a topology has, the higher score the Nodes on that topology are given\" principle.\nNote that if we add another parameter to the args, it would only be relevant for scoring, adding to the confusion.\n: This score calculation is somewhat complicated, but can be resumed as: Calculate number of matching pods per topology. There is some new feature that takes in consideration the skew Normalize by . The effect is that pods with the lowest count get a score of . Then the rest of the scores go from a certain number to . A score is given to a node with when is . This formula has the effect of reducing differentiation among nodes as the increases. : The formula is very simple, the score is equal to the percentage of free space (averaged for cpu and memory). Each score on their own give the differentiation we want, but they kind of have different scales. As a result tends to be stronger. I'm using a 9 nodes (3 in each zone) cluster as an example. I've set different initial loads, and in all cases I simulate a Deployment of 30 pods (each assumed to take 1% of node resources). Here are the scenarios, where numbers are percentage of use: For equal score weights (current configuration), this is the distribution of new pods: for both nodes and zones Note how a node with only 40% utilization gets only 1 or 0 pods. For double score weights for both nodes and zones Only at 80% utilization we observe bigger skews, which is reasonable. But the spreading is still pretty good at 40% For planned default spreading See , but this might change, depending on feedback here. I think this is enough justification to increase the weight of to 2 in its 2nd beta release.\nSounds reasonable. It's mathematically equivalent to changing the formula to e.g. , right? Have you considered doing that instead of changing the weight? It seems like that'd help everyone except those who specifically already increased , instead of just people who are using the defaults?\nIt'd make sense to me if the available functions were in the same scale / units so that they'd compose better, but it's not obvious how to do that--what would the units be? One thought is to make the units \"std deviations from the mean\"--if we knew the output frequency of each function, we could maybe do that. So basically this would measure how unusual a preference is, I guess. (I don't think this is an actionable thought. It really bugs me that the outputs of these functions are combined when they are not really in the same units, though!)\nYes 100 is the maximum score we have defined any plugins to have. I'm not convinced this would work. The variance in my examples for initial resource usage is quite high. So it would still get a lot of importance. But, you have triggered me to start thinking about the linearity of the scores. The formula is good because it reduce the variance as increases. However, that reduction is too aggressive. On the other hand, is completely linear. I'll experiment changing the linearity of either function, but preferring to do so for the beta plugin.\nHere go 2 alternatives, with the same situations as We either solution, we wouldn't need to modify the weights Although we are not getting as good spreading as simply doubling the weight. But this is only in the last scenario, where we have usages of 80%. So I would call this rather a win (we should be paying greater attention to resource usage at this point). I think option B is simpler to understand and gives slightly better results (only noticeable in the last case). So this would be my algorithmic preference. However, I'm fine with option A.\nYeah, that's a great idea. You can generalize that and give each function a both a linear weight and a \"gamma\" (name from image processing): or even go for (but in this case I'm in favor of just making a targeted fix to that one function)\nAwesome, Any concerns against option B? Given that it's an older feature.\nhow would option B impact node ranking in the absence of spreading? would image size scores start to have bigger impact?\nI think SIG Scheduling should probably have a discussion. I could go either way, my naive inclination is to just go for the change, but I'm sure it'll break someone. We probably need a way to get back the old behavior, or a suggested weight change that would approximate it (not clear how well that could do though). I also have question, which is how well does this new value compose with other functions? I feel like the proper answer looks like an evaluation criteria + {long list of scenarios} + {list of workloads to schedule} + randomized testing / benchmarking. Basically the multi-dimensional version of the analysis here. That sounds like a lot of work but it would permit actual optimization of these functions.\nconfigurable gamma is an obvious way... :/\nI don't think we want another config parameter, we already have the weight concept and it is configurable, so I think we should go with that, just double the spreading score weight by default, and users who don't favor this behavior can configure it back to 1.\nYeah, that sounds expedient. ahg-g <:\nIn short, yes. Although that score has a mechanism to avoid the \"node heating problem\" we could have undesirable consequences unless we carefully analyze the impact. But that same reasoning doesn't apply to option A (improving PodTopologySpreading), for the following reasons: The score in 1.18 had pretty weak signal, that was only useful with very small number of pods. We have drastically improved this . So it can only be compared to other scores now. The score can now be tweaked in the PodSpec through maxSkew I would argue that option ~B~ A is better that simply doubling the weight because its signal is more conservative when it matters (nodes have high usage).\nI am not clear on the following: you say \"we could have undesirable consequences unless we carefully analyze the impact.\" but then you argue for the option at the end, can you clarify? We had default spreading that was giving similar strong signal, we should compare with that. My concern with modifying the is that we may break someone. Increasing the weight is something that users can revert, while option B can't be reverted.\nSorry, I meant option A, improve PodTopologySpread\nOne potential advantage to doubling the weight is that topology spread currently offers a signal comparable to default spreading, which we can think of as a baseline for spreading signal since it has been used for a while now. Doubling the weight will allow admins to revert back to the old behavior if they wish to do so. Also, from my perspective, doubling the weight is giving better placement at high utilization ( vs )"}
{"_id":"q-en-kubernetes-da2fde1c088200d5610f06fd7955a1151e1a7aecd68788e3d49161d2a8de99d3","text":"Once we upgrade to go1.14, we can use the new CipherSuites and InsecureCipherSuites functions and CipherSuiteName function to build the supported cipher suite flags. /sig api-machinery\n/sig auth\n/cc\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-testing, kubernetes/test-infra and/or . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nas the project was upgraded to go-1.14+ is that still something that needs to be done?\nyes, these manually maintained maps should be swapped to the stdlib impls:\n/assign\n/triage accepted Thanks,"}
{"_id":"q-en-kubernetes-7bf8455a3a807df05b77f0e3ec29c1e5ea4153da582875cc4d458bc01acfad36","text":"What happened: https://k8s- These tests failed consistently: Anything else we need to know?: Volume snapshot tests need to be reworked to use / helper functions that support block volume mode. In addition, csi-driver-host-path fails when restoring block volume snapshots. /assign /sig storage\nThe label(s) cannot be applied, because the repository doesn't have them <detailsIn response to : What happened: https://k8s- Anything else we need to know?: Volume snapshot tests need to be reworked to use / helper functions that support block volume mode. In addition, csi-driver-host-path fails when restoring block volume snapshots. /sig stroage Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nhostpath driver fix:\nE2e test fix: (needs hostpath driver release first)."}
{"_id":"q-en-kubernetes-4a74dc91d29df1600cf3137c4656606581d4ad54be7e0389561e161808edd7df","text":"<!-- Please only use this template for submitting reports about flaky tests or jobs (pass or fail with no underlying change in code) in Kubernetes CI --Which jobs are flaking: pull-kubernetes-e2e-gce Which test(s) are flaking: Testgrid link: Reason for failure: Anything else we need to know: Triage board shows flakes here spiked dramatically on 3/4 /sig node /priority critical-urgent /milestone v1.18\nare you able to assist if this is related to recent topology work?\nsure, I'll take a look\nis this the test you just in ?\nAs mentioned above, this is a new test that was just , and the test sometimes fails. I don't see that any of the recent topology manager changes would cause these failures. I'll be happy to help out and dig more if or others think otherwise.\nThis is me, the bugs are untriaged containerd / Kubelet failures with low occurrence rates that this test now exposes. Investigating.\nstill flaking in non-dockershim environments: pod pod-submit-status-0-14 on node kind-worker2 container unexpected exit code 128: start=1970-01-01 00:00:00 +0000 UTC end=2020-03-06 16:16:54 +0000 UTC reason=StartError message=failed to create containerd task: OCI runtime create failed: starting container process caused \" container init caused \" mounting \"/var/lib/kubelet/pods/-819e-4840-8369-\" to rootfs \"\" at \"\" caused \"stat /var/lib/kubelet/pods/-819e-4840-8369- no such file or directory\"\"\": unknown `\nDifferent reason this time. Will continue to enumerate failure modes."}
{"_id":"q-en-kubernetes-27ad9593cabf1355cf849e43656d227454220bbe6ecc2987e545b250f692c951","text":"What happened: A SystemOOM event was not reported even when the memory limit hit was the root cgroup memory limit. What you expected to happen: A SystemOOM event should have been reported. How to reproduce it (as minimally and precisely as possible): Do not set or , and set to disable memory eviction on the kubelet. Run a pod without memory limits which slowly consumes memory, such as with from Use arguments -mem-alloc-size\", \"100Mi\", \"-mem-alloc-sleep\", \"10s\". This should trigger an OOM on the root cgroup, but no event will be generated. Anything else we need to know?: cAdvisor populates ContainerName and VictimContainerName from matching the regexp: instead in the kubelet. cc\n/sig node /priority important-soon"}
{"_id":"q-en-kubernetes-1a6279e17e106d236da2df8ae406c43b54aa2a31f961ca6490921619691d8981","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: k8s 1.17.2 config kubelet with those parameters cgroupDriver: cgroupfs cgroupRoot: /docker cgroupsPerQOS: false kubelet log always prints those E0303 17:25:03. ] Failed to get system container stats for \"/docker\": failed to get cgroup stats for \"/docker\": failed to get container info for \"/docker\": unknown container \"/docker\" E0303 17:25:03. ] eviction manager: failed to construct signal: \"allocatableMemory.available\" error: system container \"pods\" not found in metrics What you expected to happen: kubelet does not print those error information How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig node"}
{"_id":"q-en-kubernetes-7f207c102634d5fe622f73c0749707049f74cd1865ef73e2bb5f934cd8861b8b","text":"Which jobs are flaking: pull-kubernetes-integration Which test(s) are flaking: TestNominatedNodeCleanUp Testgrid link: Reason for failure: /sig scheduling\n/assign\nany progress on this one?\nI was working on it and tried to find the root cause. I will give my solution within this week\nAnother one\nStill happening cc\n/assign\n/close dupe\nClosing this issue. <detailsIn response to : /close dupe Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-a166ddd65322b2b4c2d8cbe35891249f4d033568413f3b77fc1f46ee1d1ee8b8","text":"/kind cleanup /sig storage /sig node Description: Feature gate VolumeSubpathEnvExpansion was GA in 1.17 and is marked for removal in 1.19 Related Issues: GA Release:\n/assign\n/cc"}
{"_id":"q-en-kubernetes-4d425eb67b342ad1bffa32583e9ce70b5b7f385bf982ed87181deda55eeafb05","text":"When using the package, one can easily add a flag to a CLI tool. However, the output of invocation with is hard-coded to . As such, the package is not very usable in non-Kubernetes projects. /wg component-standard /assign"}
{"_id":"q-en-kubernetes-82438b7d3ab1a91c738ff981f5a7edec77d479950cb2b6233ab4c9b673847adb","text":"What happened: there are many kubelet warnings in the systemd logs I initialized the cluster with kubeadm init (no modifications) What you expected to happen: Not to have that many kubelet warnings in the system logs Anything else we need to know?: Environment: Kubernetes version (use kubectl version):Cloud provider or hardware configuration:OS (e.g: cat /etc/os-release):Kernel (e.g. uname -a):Install tools:\nI have the same issue\nshould add a option or an annotationo to resolve this /sig node /area kubelet\nSame issue kops, kubernetes 1.18.3, aws cloudprovider\nSame issue on v1.17.4 installed with rancher\nSame issue on v1.18.0 ~\nSame issue on 1.17.6\nHappens on v1.18.3 as well. This issue only occurs when container has some fsGroup set in its securityContext AND volume being mounted is either secret or a configmap. I guess they are being re-mounted every 70s or so. It's okay when the volume being mounted is a standard PV - log only appears once and doesn't get repeated.\nI'm seeing the same issue. I'm willing to fix it.\nI've digged into this a little bit further. I believe this is just a annoying warning message. AFAICS, it does not show any issue... I was reading the kubelet, its volume manager and the volume plugins source files. It seems that's the expected behavior. The desiredStateOfWorldPopulator sync loop always call the . The finds the volume plugin and check if the plugin requires mount. Secret plugin, wich is the plugin in discussion, always returns . Thus, the of the volume manager remounts the volume every time. I think the problem is in the which prints the warning message. It's more and I believe it does not take these plugins into account. The secret plugin is as it is now since . Is that make sense to you?\n/assign"}
{"_id":"q-en-kubernetes-6ef31dc84da87e31587c45884e44fac62afa58423be65897f264bf2a209b64be","text":"What happened: Pods get stuck in terminating state, as kubelet failed to unmount volume: nodev on /var/lib/kubelet/pods/-120c-45c4-aeac- type hugetlbfs (rw,relatime,pagesize=2Mi) What you expected to happen: Pod to get deleted and volumes to be cleaned up. How to reproduce it (as minimally and precisely as possible): kvm based virtual machines on ubuntu 16.04. create a pod with hugepage volumes () Delete the pod. Anything else we need to know?: Changes are related to mentioned commit id: In this file : , line no 72 : pageSize, err := resource.ParseQuantity(strings.TrimPrefix(opt, prefix) + \"i\") The mount already has a pagesize 2Mi in my case as shown here : (rw,relatime,pagesize=2Mi) But the code here suffix/appends \"i\" assuming pagesize can be only of type 2M or 1024M resulting in failure to clean up this volume. Documentation from code: / NOTE: Adding suffix 'i' as result should be comparable with a medium size. // pagesize mount option is specified without a suffix, // e.g. pagesize=2M or pagesize=1024M for x86 CPUs pageSize, err := resource.ParseQuantity(strings.TrimPrefix(opt, prefix) + \"i\") if err != nil { return nil, fmt.Errorf(\"error getting page size from '%s' mount option: %v\", opt, err) } when the pod is deployed, kubelet creates the mount and we can see from logs that it already has pagesize=2Mi for my environment. Environment: Kubernetes version: kubectl version Client Version: version.Info{Major:\"1\", Minor:\"18\", GitVersion:\"v1.18.0\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2020-03-25T14:58:59Z\", GoVersion:\"go1.13.8\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"18\", GitVersion:\"v1.18.2\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2020-04-16T11:48:36Z\", GoVersion:\"go1.13.9\", Compiler:\"gc\", Platform:\"linux/amd64\"} Cloud provider or hardware configuration: Kvm hypervisor based virual machine setupOS (e.g: ): NAME=\"Ubuntu\" VERSION=\"16.04.6 LTS (Xenial Xerus)\" ID=ubuntu IDLIKE=debian PRETTYNAME=\"Ubuntu 16.04.6 LTS\" VERSIONID=\"16.04\" HOMEURL=\"\" SUPPORTURL=\"\" BUGREPORTURL=\"\" VERSIONCODENAME=xenial UBUNTUCODENAME=xenial Kernel (e.g. ): Linux master 4.4.0-169-generic -Ubuntu SMP Tue Nov 12 10:38:00 UTC 2019 x8664 x8664 x8664 GNU/LinuxInstall tools: Network plugin and version (if this is a network-related bug): Others:\n/assign bart0sh\n/sig storage\nWhat about this change ?\nThe change looks good to me. Would you be willing to submit a PR?\n& : Thanks for the quick turnaround.\nWe are seeing this issue in 1.18.6 version, can you please push the fix to this branch as well.\nCould we know when the problem was introduced? As I suppose we don't have the same issue with K8S 1.17.x release?\nHi same issue found in 1.18.2\nIt is not yet ported into v1.18.6 as well.\nCan you submit your PR for release-1.18 branch?"}
{"_id":"q-en-kubernetes-0fc661983815db7faba77c39fd6b3524bd9cdac0d42845f086e33294e3a9b733","text":"Currently we don't have an easy way of determining what number of requests are apiserver self-requests. This has historically bitten us, since we used to have a 50(?) QPS limit on apiserver self-requests (which thus has been removed). Even so, it would be a good idea to know this generally, since this may be helpful in pinpointing performance regressions for apiserver. /sig api-machinery instrumentation\nCan this issue be considered a newbie one? I'm hunting for issues which might be a good first PR.\nIn the triage meeting was mentioned that would be good to have similar issues to track other well known actors like controller-manager and kubelet. But let's keep this one for api-server calling itself to have it done. /cc\n/cc"}
{"_id":"q-en-kubernetes-d3ce650be53c7e46a0ac0bcc623a0dec5f6925c142d33c15ed61fa9c7810444a","text":"[x] Original approval issue: 09 July 2020 [x] Two weeks soak end date: 24 July 2020 This issue is created to allow edit by /wip /hold According to this APIsnoop query, there are still an endpoint which is untested. a set of 3 Pods with a static label within a Namespace. that all 3 Pods with the label in the Namespace are created. the set of Namespaced Pods with a label via DeleteCollection. that all Pods with the label have been deleted. Discover useragents: List endpoints hit by the test: Display endpoint coverage change: Two other endpoints tested here have tests in progress to conformance already. If a test with these calls gets merged, test coverage will go up by 1 points This test is also created with the goal of conformance promotion.\n/sig testing /sig architecture /area conformance\nDuring Conformance Meeting: Is there a way to write this test so it is resource agnostic? If we are just changing the resource... can be parameterize the resource-type\n/assign\nGitHub didn't allow me to assign the following users: heyste. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nTest to be created using \"Polling\" and not \"Watch tooling\"\n/reopen Merged Issue open, waiting for Promotion PR\nReopened this issue. <detailsIn response to : /reopen Merged Issue open, waiting for Promotion PR Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\nThe test appears to be flaking, need to reach out to #sig-node for some feedback.\nCan you point us to the link of the flaky tests? Thanks!\nThe Promotion The testgrid Thank you!"}
{"_id":"q-en-kubernetes-23c5335b00aca0eec51c917ea7725032775eef25c4e2b03a4a38794b0414d296","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Same symtoms as The detach is not successful until controller-manager is restarted controller-manager logs: What you expected to happen: The detach operation should know that a node is not active or its been deleted and should assume the disk is detached. How to reproduce it (as minimally and precisely as possible): This seems quite hard to reproduce, but we do run into it quite a few times where the pod is stuck in , because the controller-manager is unable to detach disks from deleted nodes. Anything else we need to know?: Environment: Kubernetes version (use ): v1.15.11 Cloud provider or hardware configuration: azure (self-managed) OS (e.g: ): Ubuntu 18.04.3 LTS Kernel (e.g. ): 5.0.0-1027-azure Install tools: kubeadm Network plugin and version (if this is a network-related bug): Others: /sig cloud-provider /area provider/azure paging\n/sig azure\nThe label(s) cannot be applied, because the repository doesn't have them <detailsIn response to : /sig azure Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig cloud-provider\n/area provider/azure\nhi have you done sth manually on the node? if node is deleted finally, the disk would be in unattached state finally, then disk attach to another node would succeed finally.\nas discussed offline, the issue is probably due to vmss instance compute name change(incorrect format: , correct format: ), user should not change vmss instance compute name, otherwise it would break disk attach/detach, since compute name is used in vmss node name passing, code is here:\nHi We don't change the instance compute name - however what i do see in the azure metadata service is in the form - What you're referring to seems to be the computer name, and we do use the defaults as provided by azure. See The attach/detach operations works fairly well most of the time, and I think we're running into some edge case here. One thing i might have missed out is the fact that there seems to have been an issue with the VM at the time. So the VM was present, but azure platform would return 400 bad request for a detach operation which probably led to this .. But when the VM was eventually deleted by the autoscaler, the re-attach was never attempted by the reconciler it would seem. and the last message about that disk was\nshould be a valid compute name according to , do you know why we define the format like this?\nThe function is used to get the name from IMDS, and its format is .\nFrom the above errors, seems the request has been sent to VMSS but failed because it's not active. Do you know the VM state during that time?\nI am not sure of the state at the time, but i believe the VM was in some kind of failed state. The VM has since been deleted, but we had errors on the scaleset at the time. The events on the VMSS look like this. The VM was eventually deleted by the autoscaler, but the disk was never re-attached to a different node, and the pod remained in for about 2 days until controller-manager was restarted.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\nApparently we hit this issue today as a result of replacing the nodepool. We 2 new nodepools and deleted the old one, and now we have 718 of 1672 pods stuck with this error in one of our clusters. The volumes are failing to be attached to new pods, the VMSS for old nodepool is removed. In AKS portal disks are still reported to be attached to VM instances in the VMSS that was deleted. hundreds of pods are stuck in Init state won't likely go forward. /remove-lifecycle rotten\n/remove-lifecycle rotten\nI have seen similar issue when handling one customer's case, it's probably caused by broken vmss cluster, pls file a ticket to VMSS team to fix the issue.\nI happened to find this issue in one customer case, this issue is caused by obsolete cache in kube-controller-manager, should skip attach/detach retry when it's 400, below are the details, following two instances were deleted 3min ago before kube-controller-manager trying to attach/detach: workaround restart kube-controller-manager should work\n/assign"}
{"_id":"q-en-kubernetes-46cd7c4715ee42faad931721887eefd6a42fe48f3f45f3077b49b6fc07c07d89","text":"After has been implemented and enabled the 5k node kubemark run has extra ~10k lines of logs like this ones: Example kubemark 5k run: We should get rid of that, it doesn't bring any value only pollutes the log\n/sig scalability /cc"}
{"_id":"q-en-kubernetes-a5e229fd70ce573d53baf12b4a14c90677be4e835206788149f7a3ce3798d9b0","text":"Graduate seccomp to GA, as defined on the respective . This is an umbrella issue to track all outstanding work. [x] API Server changes (MERGED) [x] Kubelet changes (MERGED) [x] Mark flag as deprecated (MERGED) [x] Update PSP to support seccomp fields (MERGED) [x] Document Kubernetes seccomp support (MERGED)[x] Add seccomp least privilege for docker sandbox (MERGED) [x] Add seccomp least privilege for kuberuntime (MERGED) ) /area kubelet /area kubelet-api /sig node /area security /assign\n/cc\njust a heads up that we have concluded all the core items for seccomp to go GA. There are still 2 pending PRs that ideally would also go in 1.19, but we are waiting for them to be reviewed: and\nneeds to be rebased apparently and isn't just waiting in the merge pool as verify is not passing.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nI think the warning mechanism was blocked by a not existing API. This is now tracked in\n/remove-lifecycle stale\nUpdated the description to given that the warning mechanism is now set to that milestone.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\n/remove-lifecycle stale\n/priority backlog\n/triage accepted\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/lifecycle frozen What is the remaining work here?\nI guess we have to follow-up on and implement: I can take care of that if we agree that we want to ship that in 1.27. /cc"}
{"_id":"q-en-kubernetes-04a7465ce5cc37d6e8b4f5a3688fe336581dbf09f1490dc9271f31d40db1f63d","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: I created several Workloads which require up to ten ports for both UDP and TCP protocol. Now my logs get spammed with . What you expected to happen: should consider -combination instead of only . How to reproduce it (as minimally and precisely as possible): Deploy any service with a node-port with both protocols, e.g.: Anything else we need to know?: This was already acknowledged by here: However, as this thread addresses a different bug and seems kind of stuck, i'd like to suggest that this might be addressed in a new context. Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Deployed via Rancher v2.4.3\n/sig api-machinery\n/remove-sig api-machinery /sig network\n/triage unresolved Comment when the issue is assessed and confirmed. I am a bot run by vllry. ‍\nHow did you get into this situation? The allocator should not have granted the same node port twice?\nOh right, that issue. The only real way out is to de-allocate the NodePorts and then re-allocate them, I think\nAs this happens for all of my ports in the said Production Application (the one where my log gets spammed) as well as complete fresh installation (where i verified the bug), i think there's no \"way out\" except fixing the bug. As said, for me it's necessary to assign both protocols to the same port. This might be a niche-use-case, but checking the it's actually very likely for ports to allow tcp AND udp. I'm not sure about the history of this, but it might be that issue was the origin, which was adressed in PR (\"Allow services which use same port, different protocol to use the same nodePort for both\"). But this PR doesn't contain the adjustment of the said , so still considers the behaviour as invalid, while it's actually allowed and supported.\n/assign\nAhh, I thought you meant that you had accidentally been allocated the same nodePorts and I was very confused when you said \"all of them\". I have reproduced this issue, and I can see it by inspection. The obvious fix would be to collect ports split by protocol, but I think that's not right since we don't store it split. An even simpler fix would be for to build a . Then for each protocol, for each port, build that contains all the port numbers, but remove any duplicates ONLY IF they are duplicated across protocols. Duplicates in the same protocol should be retained (to report an error later). I don't have time to tackle this tonight (and write a test), but it seems like a good bug.\nperhaps I can fix this problem /assign\nthings may be a little more complicated. Right now the is not aware of the protocol of the port so I guess simply rewriting function to return is not enough, the port allocator will still return on same node port even with different protocol. perhaps we need to refactor the to make it protocol-aware.\nI don't think we need to do make the allocator be protocol aware. IIRC we allow you to use the same nodeport on different protocols ONLY within a single service. This means (I think) that we can do hyper-local deduplication as we collect the nodeports for each service. If we find the same nodeport number with different protocols, just remove the duplicates because it is allowed. I.e. a service that looks like: return - deduped. if ANY OTHER service tried to use it would be an error at the \"rebuilt\" allocator Likewise if a single Service claims to use the same nodeport number more than once on the same protocol, it should be an error: return - NOT deduped.\nobviously, we want a great unit test here :P\nah got it! thanks for your detailed explanation!\nHi! I have filed a PR to resolve this issue, could you take a look?"}
{"_id":"q-en-kubernetes-ab71f408e008080218422bbeafb97e78bd9dc60107a358858d77c0817cb34c3e","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: add a CRD resource with kind only a single character while cause kube-apiserver crashslice bounds out of range What you expected to happen: add validation to limit CRD kind definition or fix the bug of slice out of bounds How to reproduce it (as minimally and precisely as possible): apiVersion: kind: CustomResourceDefinition metadata: name: spec: group: version: v1 names: kind: R plural: rs singular: r scope: Cluster Anything else we need to know?: apiserverlogs: 10:05:32. 6 ] Shutting down OpenAPI controller E6606 10:05:32. 6 ] Observed a panic: runtime.boundsError{x:-1, y:1, signed:true, code:0x3(runtime error: slice bounds out of range [-1:]) goroutine 1670 [running]: (0x41458a0, Gxc00ed66d40) /workspace/anago-v1.17.G- +0xa3 (0x0, 0x0, 0x0) /workspace/anago-v1.17.G- +0x82 panic(0x41458a0, Oxc0Oed66d40) +0x1b2 (0x7493232, 0x1, 0x447d778, 0x1, 0x130, 0xc00ed13680) /workspace/anago-v1.17.0- +0x2b8 (\"builder).descriptionFor(€xc007b351f0, 0x0, 0x0, 0x4481982, 0x6, Gxc00ed06d00, 0x18) /workspace/anago-v1.17.0- +0x3c6 (\"builder).buildRoute(GxcGG7b351fG, 0xc00ed66d00, 0x18, 0x0, 0x0, 0x447f0F3, 0x4, 0x447f71b, 0x4, 0x4481982, ...) /workspace/anago-v1.17.0- .go :281 +0x197 (0xc66b7bcfd0, 0xc005d24a3c, 0x2, 0x101, 0x1, 0x2, Gxc60b76aec8) /workspace/anago-v1.17.0-rc .2.10+70132b0F130acc/src/k8s .io/kubernetes/output/dockerized/go/src/k8s .io/kubernetes/vendor/k8s .io/apiextensions-apiserver/pkg/controller/openapi/builder/builder .go :146 +0x49d (0xc66b7bcfd0, 0x0, Oxb, Gx447f1db, 0x4, 0x1) /workspace/anago-v1.17.0- +0x1a8 (\"Controller).Run(0xc0012c8050, Gxc0G1d61200, 0xc007a6d1e0, 0xc069cced80) /workspace/anago-v1.17.0- +0x2da created by /workspace/anago-v1.17.0- .go :236 +0x244 panic: runtime error: slice bounds out of range [-1:] [recovered] panic: runtime error: slice bounds out of range [-1:] Environment: Kubernetes version (use ): 1.18.0 Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ):Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig apps\n/assign\nThere is a typo in the subject."}
{"_id":"q-en-kubernetes-96914a35e7b0e014c5e344e430617ac0eb0a5cf91e6bc5fb60b57bece6b152e8","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI --Which jobs are failing: Which test(s) are failing: Since when has it been failing: Testgrid link: Reason for failure: Anything else we need to know: /sig testing /cc /priority critical-urgent /milestone v1.19\n/assign\nI see the PR merged, however, the job still failing in the recent run 22:31 PDT Failures for bootstrap-e2e-minion-group (if any): W0611 06:08:56.190] 2020/06/11 06:08:56 Step './cluster/log-dump/log- /workspace/artifacts' finished in 2m14.442417098s W0611 06:08:44.597] ERROR: () [/usr/bin/scp] exited with return code [1]. W0611 06:08:46.070] scp: No such file or directory W0611 06:08:46.070] scp: /var/log/node-problem- No such file or directory W0611 06:08:46.072] scp: No such file or directory W0611 06:08:46.072] scp: No such file or directory W0611 06:08:46.073] scp: No such file or directory W0611 06:08:46.074] scp: /var/log/node-problem- No such file or directory W0611 06:08:46.075] scp: No such file or directory W0611 06:08:46.075] scp: No such file or directory W0611 06:08:46.079] ERROR: () [/usr/bin/scp] exited with return code [1]. W0611 06:08:46.084] ERROR: () [/usr/bin/scp] exited with return code [1]. W0611 06:08:51.917] INSTANCEGROUPS=bootstrap-e2e-minion-group W0611 06:08:51.918] NODENAMES=bootstrap-e2e-minion-group-dq3m bootstrap-e2e-minion-group-gjk0 bootstrap-e2e-minion-group-qpxw I0611 06:08:54.017] Failures for bootstrap-e2e-minion-group (if any): W0611 06:08:56.190] 2020/06/11 06:08:56 Step './cluster/log-dump/log- /workspace/artifacts' finished in 2m14.442417098s W0611 06:08:56.191] 2020/06/11 06:08:56 Running: ./hack/e2e-internal/e2e-\nI see the issue closed by the PR however the job still failing since 06-10 16:31 PDT\nlooks like we just turned green :)\ncool! :)\nYes it's indeed ... thanks for the support!"}
{"_id":"q-en-kubernetes-0b006d7026e2b9b88292d9991e285eb8c6e152091f24ce82370f6211f56b3dec","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI --Which jobs are failing: https://k8s- Which test(s) are failing: sig-network] SCTP [Feature:SCTP] [LinuxOnly] should create a Pod with SCTP HostPort Since when has it been failing: Test [sig-network] SCTP [Feature:SCTP] [LinuxOnly] should create a Pod with SCTP HostPort has failed 3 times since it last passed at commit/. It started failing between commit/ (Tue Jun 09 13:25 PDT) and commit/ (Wed Jun 10 01:26 PDT): Testgrid link: https://k8s- Reason for failure: Anything else we need to know: I'm not sure if we already did it but there was a proposal to change from dockershim to containerd for the GCE jobs\nThere are no sig labels on this issue. Please add an appropriate label by using one of the following commands: - Please see the for a listing of the SIGs, working groups, and committees available. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign /cc\nGitHub didn't allow me to assign the following users: janosi. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign /cc Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nyes, that would break us... the test checks \"are we using kubenet\", but it didn't occur to me that we would need to deal with kubenet with CRI instead of kubenet with dockershim. It should be easy to make it check for both the dockershim rule and the CNI portmap rule though...\nyes, according to the \"Changes\" link on the dashboard, containerd replaced docker in GCE right before the failures.\nLet's see if I can make it :)\nSo actually the test is pretty generic (it just looks for and ) and it seems like it ought to pass correctly with CNI portmap too. I'm not sure what's failing. You may need to add some debugging...\nOK, so as I can see, containerd's CRI implementation just ignores all HostPort definitions that do not have TCP or UDP Protocol value. I.e. containerd will not send HostPort mapping definitions to the CNI if the protocol is not TCP or UDP.\nseems this is the reason why sctp was skipped\nYes, exactly. Introduced right after SCTP support was merged in k8s. Just found this in the package: Though not sure it \"knows\" the right runtime at runtime, because there is no user of it in :)\njust curious, why do we skip in containerd and allow it with dockershim?\nHmm, do you mean why do not we skip in case of containerd? The reason is that I did not remember that containerd PR about the SCTP support. Let's see if I can skip it.\nHm, that is not true. My search engine was wrong. After a restart I found usage of that in e2e_node\nHas a bug been filed upstream in containerd? Just skipping the tests seems like the wrong approach here.\nsorry, I meant why we are not allowing it in containerd, does this feature should work in all container runtimes? CRIO, docker and containerd? if affirmative, we can submit a PR to containerd/cri to allow SCTP, I can do it if you don't have time\nExcuse me, I misunderstood your question. Yep, why not. I mean, it could work in all container runtimes. As I understand the containerd implementation is based on a misunderstanding: that the host port mapping loads the SCTP kernel module and that would be bad for those apps that want to use user space SCTP. But as long as there is no SCTP socket created on a node the SCTP kernel module is not loaded.\nI can initiate a discussion with containerd about this.\nseems this dates back to the original SCTP KEP, seems like it should be a pretty simple fix (?) thanks for filing the bug and for the details there!\nCRI-O seems to have the same issue:\nJust confirming you want this filter to be gone now. Or do you need the filter to be manageable in some way.. filtered by a config switch?\nFor TCP and UDP HostPorts, the runtime is expected to do two things: a listening socket on that port so the kernel will think it's in use and not allocate it to anyone else iptables rules (or whatever) to route traffic from the host port to the container port For SCTP HostPorts, the runtime should do the second part (iptables rules), but should not do the first part (listening socket), for two reasons: Some users want to use \"userspace SCTP\", in which case they will be referring to SCTP at the Kubernetes level but need to keep the kernel module from ever being loaded. (This is discussed more .) If is not loaded when Kubernetes starts, we want to avoid having Kubernetes itself cause it to be auto-loaded, for security reasons. (More discussion of this .) So, runtimes should be filtering out SCTP for the opening-listening-sockets part, but not for the making-iptables-rules part. (Some implementations, notably CNI , don't actually do the opening-listening-sockets part, so they wouldn't be filtering at all.)\nExactly. The current state in the runtimes: Docker (dockershim, actually): OK. The logic was implemented together with the basic SCTP feature, so listening on the host port is filtered for SCTP, but the iptables rules are created Containerd (CRI, actually): the patch which was applied 2 years ago filters out every SCTP host port handling, because it filters SCTP ports from the list of port mappings that are passed to CNI (portmap) later. Actually, containerd (CRI) does not perform the \"listen\" part at all, so the patch is not needed at all CRI-O: the patch applied 2 years ago filters out everything. CRI-O imports the host port manager implementation from kubelet/dockershim. For this reason it has the logic that can filter out SCTP ports from the \"listen\" part, but still applies the iptables rules. I.e. the patch is not needed in CRI-O either.\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale The test is flaky now :thinking: https://k8s-\nI guess it is a timing issue? Do you think some extra polling rounds would help? (I did not disappear, but currently I am very loaded with my regular job...)\nI've submitted a PR to fix it and another one to test the sctp connectivity, please review them, I've pinged you there :smile:"}
{"_id":"q-en-kubernetes-fad8ee5585f13cbd059c3b23e9b0beba24a9e5cc8dc4cbda61f103e29d873595","text":"The purpose of the file for each Kubernetes repository is to provide a list of people who can assist the Kubernetes in the event that a security issue related to the repository is discovered or disclosed. As described in the file, those on the list should agree to our . Please update the  file for the apimachinery repository. After finding people who are willing to work in this capacity, you should add them to the list, then remove PSC members (except any PSC member will be working as a security contact for this repository). The list is GitHub usernames, optionally followed by an email address. If no email address is listed, the PSC will use the email address found on git commits made by the listed user. The file may already have people listed who are secuirty contacts. In that case, simply remove any PSC members who aren't also security contacts for the repo. See for more information /area security /committee product-security /kind cleanup /lifecycle frozen /priority important-soon /sig api-machinery"}
{"_id":"q-en-kubernetes-33911e735262f8481400ca2430d671c6536468b71d87ed87bbe4f11626c79fd6","text":"The purpose of the file for each Kubernetes repository is to provide a list of people who can assist the Kubernetes in the event that a security issue related to the repository is discovered or disclosed. As described in the file, those on the list should agree to our . Please update the  file for the sample-apiserver repository. After finding people who are willing to work in this capacity, you should add them to the list, then remove PSC members (except any PSC member will be working as a security contact for this repository). The list is GitHub usernames, optionally followed by an email address. If no email address is listed, the PSC will use the email address found on git commits made by the listed user. The file may already have people listed who are secuirty contacts. In that case, simply remove any PSC members who aren't also security contacts for the repo. See for more information /area security /committee product-security /kind cleanup /lifecycle frozen /priority important-soon /sig api-machinery"}
{"_id":"q-en-kubernetes-92c02c7b117b26d0de6fc38be2278efcef8362879c74a9a2126914da420991a4","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI --Which jobs are failing: ci-kubernetes-e2e-aks-engine-azure-master-windows-2004 Which test(s) are failing: [] Container Runtime blackbox test when running a container with a new image should be able to pull from private registry with secret [NodeConformance] Since when has it been failing: since enabling the Windows 2004 SAC release tests Testgrid link: Reason for failure: The manifest for is missing the new windows server version 2004. This has occurred before: I looks like we need someone with appropriate permissions to update the manifests once the images are ready. Anything else we need to know: There was some work done in to automate some of this. Is there something missing? There are a few other tests that are failing in this job but should be covered by updating /cc /sig windows\nThe test passed 2x consecutively. We can close this now. /close\nClosing this issue. <detailsIn response to : The test passed 2x consecutively. We can close this now. Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-d1e994e5749edcd07edbd0f19c7df9185dabd2833b25fd1731e4d47ccf8dce07","text":"<!-- Please only use this template for submitting reports about flaky tests or jobs (pass or fail with no underlying change in code) in Kubernetes CI --Which jobs are flaking: pull-kubernetes-e2e-gce Which test(s) are flaking: Discovery should validate PreferredVersion for each APIGroup Testgrid link: Reason for failure: The test does not handle API groups which are or removed dynamically during the run of the test. Just because an API group is present in at the beginning of the test does not mean or will be present by the end of the test. This is particularly true for the dynamic/custom-resource API groups registered/unregistered during the e2e test run /sig api-machinery /assign\nmitigation opened at\nSaw this today in It seems to happen relatively often Likely due to the CRD created in: Should the ValidatingAdmissionPolicy test CRD have its group name template changed to match the existing test exclusion for groups ending in /cc"}
{"_id":"q-en-kubernetes-32f73fe3230f64533d19b392edefbaf0b0e1a8feaecfa7ce5826559c53a1a616","text":"Which jobs are flaking: ci-kubernetes-node-kubelet-serial-hugepages Which test(s) are flaking: all Testgrid link: https://k8s- Reason for failure:\n/sig node\n/assign /cc\nGitHub didn't allow me to assign the following users: cynepco3hahue. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign /cc Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\ntestgrid alert:\n/cc\nI started to look into logs and the issue that one of the kubelet restarts fails on: I will try to reproduce the issue locally.\nany news? Have you managed to reproduce the issue locally?\nI already have a fix that is waiting for merge -"}
{"_id":"q-en-kubernetes-8f9ac5fdef0c838965d475f11d5aa85a5be83149ab6d29c24d50d1d19652ecb3","text":"What happened: This issue was originally reported via the Kubernetes HackerOne bug bounty by HackerOne user but we have decided that due to the high privileges required, it can be addressed publicly. A user able to create CRDs could create a malicious CRD such that listing CRs will cause enormous amounts of CPU usage on the API server. What you expected to happen: The function should not expand strings with more than two consecutive periods. Either it should collapse additional repeated periods or it should return an error. How to reproduce it (as minimally and precisely as possible): /kind bug /area apiserver /area security /committee product-security /priority important-longterm\nHow about we take it a step further by rejecting the jsonpath which contains when parsing it? as more than 2 periods seem to make no sense IMO\ngood idea. I have posted to do what you suggest.\n/sig api-machinery\n/cc"}
{"_id":"q-en-kubernetes-fb153de2f1893790dcc98459ecf834aed330320a9fa128246d064c46ffa677ee","text":"Looks like this test is never able to succeed because service does not exist: /kind failing-test /sig instrumentation\npossibly related to\n/assign\n/assign\n/assign\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle rotten\nwhere did we get with this?\nLooks like those tests didn't work for at least half a year and nobody cared... Should we delete this e2e test as well? If so I can help with that.\nYes, let's remove them. Would appreciate the help! /remove-lifecycle rotten\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\nStill failing: /remove-lifecycle stale"}
{"_id":"q-en-kubernetes-0a1bf2fc3bac35f872e7d40363318016d86b1c04208fb60bb87350c6a712505d","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: For each NLB created, K8s creates an ingress rule on the node group's security group for each port exposed for each CIDR attached to the VPC to allow health checks to succeed. AWS limits the number of rules per security group to 60 by default. This quota is adjustable but multiplied by the quota for security groups per network interface cannot exceed 1000. I have a VPC with 16 CIDR blocks attached to it and a quota for rules per security group set to 80. Each NLB created adds 18 new rules (16 for Health Check, 1 for Client, and 1 for MTU) to the node group's security groups. This high number of unnecessary rules created makes it impossible to create more than 4 NLBs per node group (if each one has a single port exposed). What you expected to happen: For the health check to succeed, the NLB must be able to ping the node port for the service on the cluster's node group. So the node group's security group just needs an inbound rule for each ENI of the NLB, making the number of rules necessary on the cluster's node group directly proportional to the number of AZs the NLB is configured. How to reproduce it (as minimally and precisely as possible): Create a VPC on AWS with 8 CIDRs attached to it; Create a K8s Cluster; Create a Network Load Balancer with 6 ports exposed; If the default quota of rules per security group has not been changed, the NLB creation will fail with RulesPerSecurityGroupLimitExceeded. Anything else we need to know?: Environment: Kubernetes version (use ):Cloud provider or hardware configuration: OS (e.g: ):Kernel (e.g. ):Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig cloud-provider\n/area provider/aws"}
{"_id":"q-en-kubernetes-78e05eb7f06c339373840cdecc454b25eff99bf89941b5ab73426b97281b26d9","text":"Running : cpu usage goes to 2 cores @ 100% What you expected to happen: Way lower cpu usage. Environment: Kubernetes version (use ):Cloud provider or hardware configuration: GKE OS (e.g: ):Kernel (e.g. ): Linux felix-Latitude-7490 5.3.0-62-generic ~18.04.1-Ubuntu SMP Wed Jun 24 16:17:03 UTC 2020 x8664 x8664 x86_64 GNU/Linux\nHey I just did not understood, the cpu goes to 200% in the node or in the place where kubectl is running? Thanks!\n/priority awaiting-more-evidence /sig cli\nOn my laptop running kubectl\nSame issue on mac: kubectl version Client Version: version.Info{Major:\"1\", Minor:\"17+\", GitVersion:\"v1.17.9-eks-\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2020-07-17T19:00:19Z\", GoVersion:\"go1.13.9\", Compiler:\"gc\", Platform:\"darwin/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"16+\", GitVersion:\"v1.16.13-eks-\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2020-07-17T18:48:53Z\", GoVersion:\"go1.13.9\", Compiler:\"gc\", Platform:\"linux/amd64\"} PID COMMAND %CPU TIME #TH #WQ #PORTS MEM PURG CMPRS PGRP PPID STATE BOOSTS %CPUME %CPUOTHRS UID FAULTS COW MSGSENT MSGRECV SYSBSD SYSMACH CSW PAGEINS IDLEW POWER INSTRS CYCLES USER #MREGS RPRVT VPRVT VSIZE KPRVT KSHRD kubectl 100.5 03:30.41 32/1 0 43 16M 0B 0B 1322 running *0[1] 0. 0. 274 36 16 + 117 + 3730 266+ 100.5 costel.lupoaie N/A N/A N/A N/A N/A N/A I noticed the problem after upgrade from 1.15. Maybe it't not kubectl itself but the way it communicates with external auth provider like IAM in my case.\noh I see. Thanks I've no access to cloud providers right now to test, but as you've provisioned in EKS using IAM and is using GKE it MIGHT be a thing. Will wait to see if someone else have access to cloud providers and can test this :)\nMaybe I am wrong and it's just some busy waiting. In my case I noticed it when draining nodes with many pods that cannot immediately be stopped because it would break the deployments maxUnavailable limits, so kubectl has to wait until it can stop all the pods on the draining node.\nIn my case also this was triggered when pods couldn't be evicted immediately because of a violation of pdb.\ndoes not lead to high cpu usage in my case, so I suspect it has nothing to do with IAM/GKE auth, but with evictions failing because of pdb (but I am not experienced enough to know kubectl works in the background).\nI can confirm 100% (not 200%) CPU core usage for each process. Currently I'm draining four nodes in parallel (4x ), all four are waiting due to PDB's, and all are at 100% CPU most of the time. I'll try with a newer kubectl next time.\nOK, so it seems to be related to PDB and Drain. I'll try to take a look into it, try to reproduce with the last supported clients. Thank you for reporting! /assign\nThanks for looking into it! And my apologies, I forgot to comment on this before: Our cluster is deployed on bare metal on premise, so it's probably not EKS og GKE related.\nIs that possible you try draining with \"disable-eviction\" just to make sure PDB or eviction control is causing this? Thanks!\nI found a newer kubectl lying around, so it's now v1.18.8 (1.17 didn't include it seems) I couldn't test on the same cluster, but I just drained a host from our staging env with a similar amount of pods to what we run in prod (~85 vs ~100), and I think you're right. CPU usage certainly didn't go above 20% at any point on any core, and that's basically background noise on this host.\nSo my theory here is that kubectl keeps waiting for the node to drain and this is causing a high CPU usage. I'm also creating a cluster here to reproduce the behavior, and see what part of the code is this being caused\nConfirmed here :) Even in last kubectl v1.19.2 the problem happens. It seems also it's putting some pressure in API Server because of the amount of nodes vs PDB. disable-eviction is a \"workaround\" not desired if using PDB /remove-priority awaiting-more-evidence /priority important-soon\nI'm hitting my head in the wall here trying to figure out why the goroutine causes the peak of CPU usage, so anyone with better skills then me in Go, feel free to join :) is the function taking the whole CPU apparently Will keep digging what part of this function is taking the whole CPU. When removing the and making this 'serial' it seems to keep the CPU in low usage but I don't know if this is the desired scenario, as draining a node could take much longer since other Pods may not be using PDB. EDIT: but it also seems that the drain does not goes further because I forgot to change the return/break stuff inside the loop. So I'm thinking tomorrow maybe create a new 'queue' of the 'failed because of PDB' Pods, and deal with them in a separate way."}
{"_id":"q-en-kubernetes-f814eb6fee82e38bddac144e6361725a95839de7b8ffd5deb6dd60487e14c610","text":"Using scheduler like or , it always get unwanted output like: It simply because when adding the flags in scheduler, it will try to look up the And this will invoke: Since flag already has to prevent running other code when having : We should not print the klog info when using or right?\n/sig scheduling\nthat's annoying. Can you help to fix it?\nOk I'll take care of this. /assign"}
{"_id":"q-en-kubernetes-8b2b2e2ffd22fe6758944952b9978a29f1bcbd284b0e2f61b9aae2c62c2ca39a","text":"Which test(s) are flaking: TestWatchHTTPTimeout Reason for failure: This test is flaking, but not enough to fail our presubmit CI (which currently retries up to 3 times and passes on a single green run). Observed in when trying to disable this flake toleration. Reproducer steps: test binary: with : This is reproducible on release-1.18 as well, though is happening an order of magnitude more frequently on master: /sig api-machinery\n/assign I would like to give it a shot.\nthanks for looking"}
{"_id":"q-en-kubernetes-71c52100bb44d787a3e8fee2a4cf4c2e3b2b14a4cab61aabfb0663937ac238a8","text":"There was an effort at one point to promote it to Beta This may have moved to Service Catalog repo So question, do we remove it from kubernetes/kubernetes ?\n/sig apps /sig service-catalog Thoughts please!\nI see removal of perma-alpha APIs as a natural next step after the \"\" work to put end dates on beta APIs\nThe PodPreset functionality is not used in Service Catalog, we have an issue to remove references to it in the svc-cat repo, see: In the separate issue, we will figure out how to implement the e2e functionality for injecting secrets into an application: IMO it's ok to remove it as right now there is no decision about how to use it in svc-cat and currently we do not have enough capacity to work on that. In Kyma project but we also . /cc\nMany of us have been waiting for this to become mature since along time. Removing PodPresets would be quite sad, as it would allow many people to stop having to write their own mutating webhooks for things PodPresets could do really well. Also I had in mind to use it for automatically mounting common volumes into pods and much more.\nMe too Doing my usual check of has-this-thing-got-to-beta-so-I-can-finally-use-it only to find PodPresets had entirely disappeared This feature was going to greatly simplify our manifests. We do have the mutating webhooks feature now, which we didn't when PodPresets were created. However PodPresets are a lot cleaner would have received a lot of use. Even chart designs could be greatly simplified with optional PodPreset manifests. So I was looking forward to getting them available for use. Things in alpha can never gain traction or significant interest because managed providers won't let you enable them. This is not really a case of 'perma-beta'. But, if they were not going anywhere then I can understand removing them.\nSpeaking of gaining traction and momentum, the 'Avoiding permanent beta' policy only has a nine-month countdown for beta features. This new policy clashes with the recent popularity of managed releases, which tend to lag upstream releases by 6-9 months (2-3 releases). E.g. the current . That is before users can start planning upgrades in the 3-6 months after that. So a brand-spanking new beta feature won't reach cloud provider users until 9-15 months after it reaches beta upstream. Then users will start trying it out, and if it works well and get some interest and momentum behind it, only to find it has been already removed from a subsequent releases because of the mandatory 9-month policy In practice AWS probably won't enable a beta feature that is already removed upstream, so we can basically expect beta features will never reach, or get a chance for real-world feedback from cloud provider users. It is going to be tough for when the real world can't test a release with a beta feature until after it has already hit mandatory removal\nFor long running alpha features I usually recommend announcing depreaction and removing after one release, instead of right away. For extending the beta policy please see:\nPlease don't remove this feature! We only just discovered it this week and it was positioned to solve a real issue with common volume specs, in the same way mentioned by It would be really sad to now see it removed when it was going to fix some integration issues for us!\nCould you elaborate on why PodPresets were never promoted to beta and why you want to remove them now? It seems like a useful feature.\nThis feature would have been very useful in several operating scenarios, such as setting proxy settings before a deployment.\nCan anyone comment on what alternatives are? Without this feature we'll have a lot of duplication. No one likes duplication. We can address the Environment variables via ConfigMap and the envFrom keyword, however I cannot see anything for volumes, imagePullSecret etc.\nIt seems the alternative is just to use mutating webhooks. And there is even a port of PodPreset to a mutating web hook by redhat:\n-1000 Removing PodPreset is terrible. Shame on you!\nIt doesn't look much maintained and active, sadly\nRelated:"}
{"_id":"q-en-kubernetes-24e50eba12cc5bb3b5fa8345e285846b14a09171148bf8d2bcc8a56f1bc6a4de","text":"Currently we unconditionally move pods out of unschedulable status when an event happens. However, we can optimize this by setting some conditions for each event to avoid unnecessarily doing that: 1) PV, PVC, StorageClass or CSINode add/update: move only pods with a PVC. 2) Service add/update: only if ServiceAffinity plugin is enabled in one of the profiles 3) Node add/update: test the pods against the , this include: NodeAffinity, NodeName NodePorts, NodeResources and TaintsTolerations filters. 4) Pod delete: keep it as is (spreading or affinity constraints must be tested, which is expensive to do) This is necessary to avoid wasting scheduling cycles. Related proposal: /sig scheduling /assign\nGitHub didn't allow me to assign the following users: adtac. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : Currently we unconditionally move pods out of unschedulable status when an event happens. However, we can optimize this by setting some conditions for each event to avoid unnecessarily doing that: 2) Service add/update: only if ServiceAffinity plugin is enabled in one of the profiles 3) Node add/update: test the pods against the , this include: NodeAffinity, NodeName NodePorts, NodeResources and TaintsTolerations filters. 4) Pod delete: keep it as is (spreading or affinity constraints must be tested, which is expensive to do) /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\n/cc I'm thinking of if we can amplify the scope: which is to say, not only enable/disable moving (all) pods, but also provides mechanics to let plugins notify the scheduler framework which particular pods should be moved. This would benefits in 's coscheduling implementation to move the preceding (N-1) pods back to activeQ when the Nth pod comes - which satisfies the \"minAvaialbe\" requirement. Moreover, this would help in getting rid of flushing pods back to activeQ .\nThe suggestion is not to enable/disable moving all pods in all events, this is only true for the Service events. For the other two, node and volumes, we will test each pod and move the ones that pass the condition. I agree that our proposal of integrating this with filters/permit plugins is more generic, but it adds a lot of complexity and potential for bugs, so we need to be careful with that I think.\nOne approach could be to start simple, validate and establish a reference point in terms of perf improvements, and then migrate to something more generic. The more generic approach will not be more performant than what is proposed in this issue.\nYes, filters/permit runs in the critical path, and move Pods isn't a lock-free operation. What I want to say is during the design/implementation of this proposal, keep it in mind that plugins may need to actively \"suggest\" moving particular Pods back to activeQ. Agree.\nRight, we already designed the interface for such a potential eventuality:\njust fixing double \"moving\" in title: /retitle Avoid moving pods out of unschedulable status unconditionally\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\n/assign /unassign\n/cc\nwhat is the current plan? Is it to do the entire proposal in \"Efficient Requeuing\" doc or just the minimal checks proposed in the issue?\nI plan to fix the problems mentioned in this issue, but not entirely following the logic mentioned here. For example, it's inappropriate to move a Pod with PVC once a CSINode happens - the Pod may fail due to a reason irrelevant to its PVC. What I'm proposing is more aligned with the Google doc, but there are 2 major distinctions: I'm not associating the interface to a particular extension point - the EventsToRegister is more a general hook instead of semantically associated with a particular scheduling phase. I'm not going to pass the changed event object to let users decide whether to move or not - that'd increase the latency, also impose an extra burden on plugin developers. Fixing this issue is the minimum goal I can promise in 1.21, but in the future, we may provide other interfaces like enable users to proactively move Pods inside scheduling queue.\nThat is true, but this is the case right now, not something new this issue is proposing to do. This issue is proposing to address the obvious moves that we can avoid with minimal complexity, which will also likely give us the most bang for the buck. Do you have a more detailed proposal for what you are planning to do?\nYes, the code is almost complete - PR1 () and PR2 (), I will send out another PR later to use ServiceAffinity as an example to show case how it works in details. Basically, the plugin can implement an optional interface called and return a number of ClusterEvents that a Pod failed by that plugin can be made re-schedulable, and also upon a Pod's scheduling failure, we store the \"failed plugin\" into , so that upon moving Pods in the internal queues, we're able to tell whether the Pod's failed plugins are associated/registered with the current event, and hence perform efficient enqueuing.\nYou can wait late today or tomorrow when the 3rd PoC PR will be ready, and it'd show you a whole picture.\nFYI: the 3rd PR to showcase how it works: PoC: ServiceAffinity enqueue .\nDo we have a benchmark we can use to prove that the changes are having a positive impact?\nAlso, are we planning to keep the existing checks that some events have?\nWill do that. Yes, they will be kept.\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/triage accepted\nI think we should still do 3: or even just NodeAffinity+TaintsTolerations The problem is that just one node is being and potentially we are retrying all pending pods. And this means that we are testing all Nodes for each of those Pods.\nI think we can do sth. like this: upon a Node Add/Update, for each unschedulablePodInfo, only verify if the Pod can pass NodeAffinity/NodePorr/etc. filtering logic when the Pod failed by NodeAffinity/NodePort/etc. plugin. In this way, we can save cycles for Pods not failed with NodeAffinity/NodePort.\nLatest status of this issue: Interface and implementation in framework side [x] Surface info of failed plugins during PerFilter, Filter and Permit [] [x] Store a cluster event to plugin map in SchedulerQueue [] Plugin implementation and perf evaluation.\n/help Some tips/notes for contributors who are interested: please dedicate some time to the item(s) you claimed as code freeze is around the corner, we'd like to finish the above items as early as possible works as a sample you can refer, basically you need to think about which events can make the unschedulabed Pods failed by the specified plugins schedulable, and specify the reasons. Also give a benchmark test result proving it. to design a desired benchmark test, you may need to try the \"churn operator\", which means you need to rebase locally (unless they get merged) if you meet timeout issue frequently during running a 5000Nodes test, you can work it around by rebasing it's a good practice to run the benchmark test multiple times and eliminate the outliers. Here is a for your information.\nI would like to take ‘ tainttoleration’ to learn something about scheduling plugins.\nThanks! I updated the tips to include a to automate running the benchmark tests multiple times.\nI claim \"nodeports\" and shall learn about it firstly.\nCan I try to deal with the \"nodeaffinity\" plugin\nI would like take \"podtopologyspread\" plugin.\nCan I take ?\nI can take plugin.\nCan I take ?\nCan i take ?\nI opened and will add testcase and test result tommorrow.\nI would like to take on\nI think I'm left with\nupdated with proper owners. Thanks! Please assign the reviewer to me when your PR is ready. /remove-help\nWei, there were followups from , are we tracking them somewhere?\nthe only followup is I will work on that. Let me know if any other followup I missed.\nsg, can you open an issue to track that unless you have an open PR.\nSure, create two issues tracking the follow-ups: [ ] sched: cleanup clusterEventReg from [] [x] sched: requeue optimization upon Node Add/Update []\nI submitted a PR ( ) for NodePorts, not yet tested. I shall test it this week.\nAfter discussion with , due to 1.21 code freeze (today), we decided to postpone merging most of the opening PRs here. Apologize for that - we'd better not rush to avoid regressions. So priority your time on the on-going PRs, we will still review them, just take your time :)\nUse to convert the json files to table.\nThat's a really useful tool thanks so much for writing that!\noh bummer. I would have liked at least serviceaffinity to go in.\nActually, to get rid of the reaction to service changes... all plugins would have to opt-out, so nvm.\nYes, to achieve that, every plugin has to implement EventsToRegister()... BTW: one way is work it around is to add some temporary logic on framework side - skip movingPods upon Service events if ServiceAffinity is not enabled. And remove this temporary logic when all plugins opt-in EventsToRegister() in the next release.\nThat wouldn't qualify for codefreeze though.\nNvm then :)\nWe should simply not register for service updates at all if ServiceAffinity is not configured, which is the default.\nProbably not a great idea to totally disable the registration, as an out-of-tree plugin may choose to respond to Service events.\nWe need to find a way for external plugins to register events because we will remove this registration once the plugin is deprecated.\nYes, also should take CRD into consideration. Probably dynamic informer.\nI still think we should disabled Service event handlers when the plugin is not registered. If an external plugin is using it, then they could enable the service affinity filter plugin with an empty affinity labels or patch event handlers until we figure out a way for external plugins to register extra events. btw, do you know of any external plugins that actually rely on service events?\nThat's fair, and I can work on this, adding some comments caveating this is a temporary workaround. I'm not aware so far.\ntbh, I would say plugins depending on services is an antipattern, given the existence of pod affinity. Probably some CRDs might make sense for specialized plugins.\nwe should add metrics to track the efficiency of the solution. The metric would track the number of pods that were not put back into the queue broken down by plugin and even type for example.\nSG. I compiled a list to include on-going and to-do items here .\nAre you still working on this?\ndid we have PRs for number 2 and number 3?\nI am asking about the two points from the four mentioned in the description of this issue:\nOops, misunderstanding, ignore me.\nnumber 2: mitigated the problem. In short, includes two parts: one part is to remove Service event from default list so that only Pods failed by a plugin (e.g., ServiceAffinity) which explicitly are interested in Service events can be moved; the other part is to implement EventsToRegister() in ServcieAfffinity plugin. But that doesn't prevent the internal processing cycles like going through the unschedulabe pods and verify whether each is failed by that plugin, so it's a mitigation. PR will fix this problem thoroughly. number 3:\nah, yeah, I knew we worked on that in the previous release, lol. Thanks."}
{"_id":"q-en-kubernetes-4f93c5799e56971be0d3796b0b592551cd30baaaf66c87c579655acba8af4b2a","text":"What happened: We're seeing legacy-cloud-providers/azure/clients (in our case, the synchronised copy used by latest cluster-autoscaler (1.19)) segfaulting under heavy pressure and ARM throttling: In that specific case the cluster-autoscaler triggered the issue, but it can likely affect other ARM clients using that code, including in-tree kubelet or controller-manager. What you expected to happen: How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): v1.18.8 Cloud provider or hardware configuration: Azure OS (e.g: ): Ubuntu 20.04.1 LTS Kernel (e.g. ): 5.4.0-1022-azure -Ubuntu SMP Install tools: Network plugin and version (if this is a network-related bug): Others: /kind bug /sig cloud-provider /area provider/azure\n/milestone v1.20"}
{"_id":"q-en-kubernetes-09f1be3358f1008ea8ae3f56d8aeca95a4589f9723026541a3a3d10cb2e0c00f","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: is created on a cgroup2 host What you expected to happen: Such directory should not be created How to reproduce it (as minimally and precisely as possible): Boot Ubuntu 20.04 with Install moby/moby with containerd/containerd and opencontainers/runc Install kubernetes-sigs/kind . This version uses as the base image. Run with Run Run to drop into the shell of the node The following directories are unexpectedly created. (No process seems to belong to these groups actually) Environment: (See above)\ncc\n/sig node"}
{"_id":"q-en-kubernetes-e7c476bc2d5f844c8e73dc71610efcbcfe941552138dfa628e2e9ebb146eabad","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Deploy a Windows cluster with containerd configured, SMB mount does not work by trying following project: It reports following error: What you expected to happen: How to reproduce it (as minimally and precisely as possible): Anything else we need to know?: Environment: Kubernetes version (use ): v1.18.6 Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others: /sig windows /sig storage\n/cc\nI was able to repro your issue using the azurefile-csi-driver for Windows. I also tried to set up a repo where I would mount an azure file share to a Windows node manually, create symlinks to the share and then schedule a pod to the node using a hostPath volume and the container was able to start up fine using contianerd. Here is the pod-spec i used I tried to mount the Azure file share with both New-PsDrive and New-SmbGlobalMapping and saw date being written the the in the azure file share in both instances. I used mklink /D to create the symlinks - and also tried creating a symlink chain because it looks like that is how they get set up when when using azurefile-csi-driver and csi-proxy Right now I think this issue may be specific to either how the csi-proxy is setting up the SMB share for Windows or how paths are getting passed around between the various components.\nthis is not related to csi-proxy, in-tree azure file does not use csi-proxy, and it does not work\nhostPath on containerd Windows does not work too:\nI captured containerd logs for this and see I spoke with and he thinks this may be related to a known issue where go-lang is getting confused on Windows if a symlink points directly to the root of a SBM share (which is does in this case). He suggested adding a trailing  to see if this fixes the issue. I will try that out now and report back.\nI tested some updates to CSI proxy to ensure the symlink to the sharepath had a trailing backslash and it looks like that resolved the issue. On my node running contianerd i have the following symlinks Previously the  on pvc--f195-4d9e-841b- was missing.\nThe core issue here is that containerd calls on the mount source path, and that function has a bug which prevents it from working when the path is a UNC share root. If the path has a trailing appended, then works properly. I have filed a bug on Go to fix this, and am working on a fix: golang/go\n/close\nClosing this issue. <detailsIn response to : /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThis issue is not resolved, right?\nI see this issue is resolved in csi-proxy v0.2.2, while it's not resolved for the in-tree azure file driver.\n/triage accepted\nAh, good point. yes, this is not fully resolved."}
{"_id":"q-en-kubernetes-b6c9f2814c257fcf076aee6432b3853318e13bb6778498e0a239839fa2da32e9","text":"Note: this issue should be viewed in the context of . If this happens, this is very likely a typo where someone used a parameter in a way they didn't mean to (for example, ). /sig scheduling /assign\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-testing, kubernetes/test-infra and/or . /lifecycle stale\n/remove-lifecycle stale\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\n/remove-lifecycle stale\n/good-first-issue\nThis request has been marked as suitable for new contributors. Please ensure the request meets the requirements listed . If this request no longer meets these requirements, the label can be removed by commenting with the command. <detailsIn response to : /good-first-issue Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nHi this is my first time contributing to kubernetes. can you provide me some guidance on how to to get started?\nHi, is this issue still valid to work on?\nyes, this is still valid. I recommend starting with the and reading from there.\nare you still working on this issue or should I work on it.\nI would like to work on this issue if you haven't started already.\nwould it be ok if I took a look at this?\n/assign\nHi Can you help me in why its not assigning the values to createNodes and createPods directly and using the opcode and countParam in performance-\nWe use a field to differentiate between different operations. The value of the field (e.g. \"createNodes\") determines the operation. The field defines the number of pods/nodes to create. What do you mean by \"assigning the values to createNodes and createPods directly\"?\nThanks for the information. By assigning, I just meant that whether we can assign createNodes directly like ( createNodes: $initNodes ).\nI don't think that's possible since there can potentially be other parameters besides the countParam. For example: Different operations could potentially have different parameters.\ncreateNodes and createPods aren't the only operations. Other operations (e.g. churn and barrier), including future ones, may not fit that format. It's also easier to parse and more explicit this way.\nThanks , Having a doubt more, in performance- under the name:SchedulingPodAntiAffinity, ! why are we giving the same values in 5000Nodes as given in 500Nodes, or this is an error.\nthat sounds like a typo indeed\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\nHello Are you still going to work on this issue?\nHi I have worked on the issue but not working for a while as was unsure about the exact problem.\n, no problem if you are willing to work on this issue, shall unassign myself\n/unassign\nThanks :)\n/assign\n/remove-lifecycle stale\nhey! I am beginner and I interested in scheduling and I know basics of scheduling , so how can I start to contribute on this issue.\nSorry, PR for this issue has already created.\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:"}
{"_id":"q-en-kubernetes-4700b6431d53fed3ac6cc4bc51c5b0e7fdf43657f232453edec23d6ad11df27a","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI --Which jobs are failing: /test pull-kubernetes-node-crio1-18-e2e Which test(s) are failing: pull-kubernetes-node-crio1-18-e2e Reason for failure: Just before node e2e tests are started, we verify that the runtime (containerd or docker) is properly running or not on the target system. This check doesn't include crio, so the targeted node is not reported ready if it's using crio as the runtime.\n/sig testing\n/sig node /remove-sig testing This check is owned by SIG node\nThe code has also not allowed this sort of dependency (beyond some legacy dockershim specific bits), but I guess it's somewhat more understandable to check in ."}
{"_id":"q-en-kubernetes-7e392226d1eb576bb46e52e7a6bfd6dd4b8cfb2ebe6708d3d3dea7c04613110b","text":"Warning: The v0.1 and v1beta1 metadata server endpoints are deprecated and scheduled for shutdown on September 30, 2020. This test (which runs in most e2e jobs that run on GCE): Schedules a container that runs this: Resulting in a number of legacy metadata endpoints getting hit. To see if this test is going to start failing across all release branches on Sep 30th, we could try disabling the endpoints ahead of time via: In fact and support for expecting different results based on whether is set. My concern is the test expects 200 for some root endpoints, and 403 for disabled endpoints. However the e-mail I received as a GCP customer tells me any legacy endpoints may return 404. I'm inclined to suggest we remove the legacy endpoints from the check, bump the image version, and cherry-pick use of the new image version back to all release-branches. WDYT since you were involved with the earlier PRs /sig auth /sig cloud-provider /provider gcp /sig testing /priority important-soon (I'm tempted to critical-urgent this but it's not drop-everything yet) /milestone v1.20\n/kind cleanup\nHeyho I wanted to check how it is going? The code freeze is starting 12.11.2020 (about 3-4 weeks from now) and while there is still plenty of time, we want to ensure that each relevant issue has a chance to be considered. As the issue is tagged for 1.20, is it still planned for this release?\nBug Triage here, we're pretty late into code freeze and it doesn't look like there has been much activity recently. I am going to clear the milestone on this issue. /milestone clear\nany update here?\nTry to cleanup the legacy endpoints in\n/milestone v1.21\n/reopen This needs to stay open until the new image has been verified to work and has been promoted (ref: )\nReopened this issue. <detailsIn response to : /reopen This needs to stay open until the new image has been verified to work and has been promoted (ref: ) Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/triage accepted /assign"}
{"_id":"q-en-kubernetes-adc864120b4bf3f0c4719014a19f7db55b0b10439d7006f29b684827d46fc0cb","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI --Which jobs are failing: ci-kubernetes-node-kubelet-serial node-kubelet-serial job Which test(s) are failing: Hard to tell, but everything at one point or another. Since when has it been failing: A month or two, by recollection. would indicate several years... ci-kubernetes-node-kubelet-serial | failing_days | 992 Testgrid link: Reason for failure: Hard to tell, some is timeout, but I think timeouts due to test failures. Anything else we need to know: Nothing consistent about the failures that I can tell besides timeouts. This is the only place these serial tests are running.\n/sig node\n/cc\nI will take a look at what happens with the serial lane.\nSo the kubelet failed to restart because of I think it can be related to the topology manager scope PR, I will dig into the code tomorrow. FYI\nIt actually looks like it was introduced by: With the same issue reported here: And Alexey has a fix for it here (I haven't reviewed it):\nThanks for the update.\nSome of these failures are definitely due to , which is causing tests to fail since pods aren't terminating in time. See e.g.\n/assign\nUnder the latest run, I saw that the kubelet failed to restart because of The problem that one of the topology manager tests failed in the middle and did not remove the state file, I can not guarantee that it a single problem, but let's start from it."}
{"_id":"q-en-kubernetes-74931a9f052513bf144ba059c4e4fc0b79a9524532a0c3ceb7b1249903c7dd42","text":"<!-- Please only use this template for submitting reports about flaky tests or jobs (pass or fail with no underlying change in code) in Kubernetes CI --Which jobs are flaking: Which test(s) are flaking: Testgrid link: Reason for failure: Timeout while - trying to gave a 404 or 502 Anything else we need to know: 4 flakes in the past week: Triage: /sig network <!-- Please see the deflaking doc () for more guidance! --\n/triage unresolved Comment when the issue is assessed and confirmed. I am a bot run by vllry. ‍\n/assign\nHi and any update on this ?\n/assign\nMinhan -- can you take a look?\n/triage accepted\nLooks like the LB did not stabilize before timeout. Still looking. We are considering disabling some of the tests here\nThis test is still flaking a bit -- recent failures from the past 2 days:\nIt looks like this test is still flaking quite frequently: Any ideas on how to move forward?\nI will start disabling the tests."}
{"_id":"q-en-kubernetes-7b671bef88144f0cebade813d78d73aec8a03a4e2cf1c4e228cccedb3bfc2e2a","text":"<!-- Please only use this template for submitting reports about flaky tests or jobs (pass or fail with no underlying change in code) in Kubernetes CI --Which jobs are flaking: bazel-test-master (periodic-kubernetes-bazel-test-master ) Which test(s) are flaking: gotest: Testgrid link: Reason for failure: Anything else we need to know: links to appreciated links to specific failures in spyglass appreciated <!-- Please see the deflaking doc () for more guidance! --cc\n/sig api-machinery\nThis is continuing to be flaky with more recent failure:\n/assign Would you mind taking a look here? thanks! /triage accepted\n- can we somehow uncover when this started happening? This test is very old and it seems it started happening very recently, so it would be helpful to see when this started happening...\nTriage shows flakes starting ~Sept 30: Before Sept 30 it seems the job was flaking much less frequently.\nMight be related to I will take a look.\nIt seems it's not related to that PR (neither to the previous: ) As there were already failures like that on Sep 24th (before this one was merged). I will take a deeper look.\nOK - it's not related to any changes mentioned before. I can reproduce this particular failure by simply adding somewhere in this block: So it's probably connected to cpu-starvation at the test level and the solution is to switch to less aggressive timeouts. Will send out a PR.\nActually - it appeared to be slightly more subtle then that. But I already fully understand the problem. It's purely test issue, but require a bit more changes (and I won't get to it today/tomorrow). I will send a fix for that early next week.\n- if that changes anything from your tracking perspective ^^\nNo board changes yet - the next stage is \"Observing\" once this issue gets closed :) If you have time - I'd love to pick your brain/understand the error more thoroughly though! Hoping you drop a good explanation in your PR next week\nThanks for sharing those insights, on your discovery of the solution to this bug.\nIt seems the PR didn't address that fully. Looking.\nout for review"}
{"_id":"q-en-kubernetes-9d9bedd3f0610afcad243ef7558bc22ee3b89ec8b1f120499439bd302201b62b","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI --Which jobs are failing: gci-gce-serial-kube-dns-nodecache Which test(s) are failing: N/A Since when has it been failing: 10/13 Testgrid link: https://k8s- Reason for failure: Cluster failed to start: Anything else we need to know: /cc\nSo, is a failing job, we can see that there is a error around binding to a specific IP address. In passing jobs, the initial core file isn't printed out () so not sure how to do an apples to apples comparison of what is broken here yet.\n/kind bug I have been looking at this offline with and The only difference between a failing and a passing job is the arguments to the bind plugin. Failing Job: Link - Image - Config: Passing Job: Link - Image: Config:\nIssue is reproducible when spinning up a cluster via kubetest using:\nchanges to lines 907-910 in hack/local-up- and lines 2467 to 2469 in configure- seem like the issue.\nWhat we see is substitution works correctly for line , but for line , it replaces \"PILLARLOCALDNS\" with empty string.\n/assign Can you investigate?\nYeah lookin into it\nRegex is matching the entire line :). Should be easy fix . Will verify in hack local up and issue PR shortly\n/sig network /triage accepted"}
{"_id":"q-en-kubernetes-b8ed1dbe8b468fe9f76018304d329e1539d0aa653ea1e07f2f9f7ec2da0a06e7","text":"What would you like to be : Add integration tests for Why is this needed: This is part of the kubernetes/enhancements /sig cli /priority important-soon\n/assign\n/triage accepted"}
{"_id":"q-en-kubernetes-b19831576787b8f74d1ac763ee7547be79eedfdf3aa07547bdd6452e8f905b8a","text":"What would you like to be : e2e test for service ExternalIPs Why is this needed: There is no test covering Service ExternalIPs Since we can not assume that the e2e binary will have connectivity against the nodes and that we can add routes to the environment, we are going to test it from a pod. I've verified that this works locally with the following commands: The test should leverage current framework, maybe something as A good example is here, maybe adding a ?\n/sig network /help wanted /triage accepted\n/help wanted /good-first-issue\n/assign\nI'd like to pick this up!\nOk,I'm not started :) /unassign /assign\nThere are two different externalIP codepaths, one for the case where the claimed external IP exists as an IP on the default interface, and one for where it does not. I suppose we could just temporarily add an IP to the default interface...\ngood idea Dan Please remember, it has to work for IPv6 and for IPv4 ... anyway, we can address these things during the review\nThis has been partially implemented with . Covering the code path for when an IP is does not exist on a node. I'll assign this back to you so you can split this issue or close it in favor of another issue. Thanks! /unassign /assign\nwhat is the different codepath for ExternalIPs that already exist on the host? ~My reading is that the only difference is that it opens a local socket on that port, can you expand on this?~ Found it\nso, it seems that this work in some environments because they are not using the bridge plugin why is not allowed that pods can reach ExternalIPs?\nWow, this is ancient and needs some fixing. This is not about preventing pod traffic from reaching those IPs. I had to really stare at it, and I am not 100% in my interpretation. First, we recognize such traffic from \"off machine\" (!physdev-is-in && !src=LOCAL, which we should probably switch to localDetector) and steer it to the Service. Second, we recognize ANY such traffic that is destined for a LOCAL address and shortcut that to the Service, Implicitly, traffic destined for this IP where the IP is not local goes to the system's default routing logic. The reliance on the bridge plugin is awful and archaic and I will buy a beverage of your choice for whomever gets rid of that. I think localDetector can cover this, but not 100% sure\nbut the ExternalIP belongs to a service and we(kube-proxy) \"knows\" where to send it (to the service) , that is the part I'm not sure to understand, because whatever the user is doing with that ExternalIP it has to send it back to the cluster to reach the service. However, there is always context that is important and maybe I'm missing something. Assuming we want to handle ALL the ExternalIP traffic with kube-proxy, the logic will be much simpler: traffic to an externalIP will be ALWAYS sent to the Service (DNAT) if traffic is NOT LOCAL, we also SNAT\n/remove-good-first-issue /remove-help-wanted"}
{"_id":"q-en-kubernetes-bf3f2554164b636c15af30366e6e5bdc5625dbcdc6f39ed32305e9b53962aee7","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Using example for from documentation failed with: What you expected to happen: Example from documentation works. How to reproduce it (as minimally and precisely as possible): How should configuration look like for ? Environment: Kubernetes version (use ):OS (e.g: ):Kernel (e.g. ):\n/sig scheduling\n/triage accepted oh, I missed a dash. It should be: Would you mind sending a PR to kubernetes/website?"}
{"_id":"q-en-kubernetes-fe836db0f396b7544c45d10c1bdc83b467bb243decb6d399f76c072a922c10ac","text":"What happened: Process running as non-root will fail to work as it cannot access /etc/hosts inside the container when umask of kubelet is configured as 027. It is because the ioutil.WriteFile() used to create /etc/hosts is impacted by umask. Even though kubelet specify the permission of 0644 explicitly, with umask of 0027, it will be configured as 0640. As a result, application running inside the pods with non-root user, failed to access it. What you expected to happen: kubelet should pre-configure required umask by itself, instead of depending on the environment. At least we should document the required umask. How to reproduce it (as minimally and precisely as possible): Lauch kubelet with umask of 027 Anything else we need to know?: It's the same problem figured out. Environment: Kubernetes version (use ): 1.19 Cloud provider or hardware configuration: N/A OS (e.g: ): Ubuntu 18.04.4 LTS Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\n/area kubelet\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\n/lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle rotten\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-contributor-experience at . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-cef38c378015a3c37c6ee1632d82c03243825979d637fc81c55115a5c8a6f392","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI --Which jobs are failing: Which test(s) are failing: Since when has it been failing: Testgrid link: Reason for failure: Anything else we need to know: Introduced in /cc /sig node /assign /milestone v1.20\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-44740d4070d0501af97dd465c140fe134e86e61a92e57be660e314c999c2fc8e","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: When I register a device plugin and set the option in true, then when I create a pod and specify this resource as 0 in yaml, an error will be reported. Examples are as follows. The request to register the device pluginA pod yaml with the resource 0kubelet return error What you expected to happen: I think when the kubelet calls , the resource value should be checked. If it is 0, the step should be skipped to avoid errors. How to reproduce it (as minimally and precisely as possible): Register a device plugin which set option is true Create a pod with the above resource 0 Anything else we need to know?: In fact, when the amount of requested resources is 0, devicemanager will get the number of allocated devices as nil, resulting in the internal cache not storing the application resource information corresponding to the container. But if is set to true at the same time, then devicemanager will check podDevices cache. The two processing steps contradict each other so it will return an error.\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node"}
{"_id":"q-en-kubernetes-43b17effa5a0f40b960beadbb50098bc7d24e26a5775c33f71d691800c42fb26","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Below k8s UT fail when run with master-golang. TestCheckpointStateRestore/RestorecheckpointwithinvalidJSON error:=== RUN TestCheckpointStateRestore/RestorecheckpointwithinvalidJSON I1124 05:38:14. ] [cpumanager] initializing new in-memory state store unexpected error while creatng checkpointState: could not restore state from checkpoint: json: unexpected end of JSON input, please drain this node and delete the CPU manager checkpoint file \"/tmp/cpumanagerstatetest479660663/cpumanagercheckpointtest\" before restarting Kubelet --- FAIL: TestCheckpointStateRestore/RestorecheckpointwithinvalidJSON (0.00s) TestStrategicMergePatchInvalid error:=== RUN TestStrategicMergePatchInvalid expected \"invalid character 'b' looking for beginning of value\", got \"json: invalid character 'b' looking for beginning of value\" --- FAIL: TestStrategicMergePatchInvalid (0.00s) TestJSONPatch error:=== RUN TestJSONPatch invalid-syntax: expected error invalid character 'i' looking for beginning of value, but got json: invalid character 'i' looking for beginning of value --- FAIL: TestJSONPatch (0.00s) TestV4ErrorDecoder expandless error:=== RUN TestV4ErrorDecoder wrong error for message \"{\": want=\"error stream protocol error: unexpected end of JSON input in \"{\"\", got=\"error stream protocol error: json: unexpected end of JSON input in \"{\"\" --- FAIL: TestV4ErrorDecoder (0.00s What you expected to happen: The tests should pass with no error. How to reproduce it (as minimally and precisely as possible): a stable version of golang installed master go from upstream git clone cd go/src cd ../.. cp -r ./go /usr/local export PATH=/usr/local/go/bin:$PATH kubernetes repo and run the unit tests. () Anything else we need to know?: These failures are occurring due to mismatch of error messages between upstream go and the k8s tests. There is an addition \"json:\" prefix to SyntaxError messages after this below commit: Environment: Kubernetes version (use ): latest source code from upstream Cloud provider or hardware configuration: OS (e.g: ): \"CentOS Linux 8 (Core)\" Kernel (e.g. ): Linux rajalakshmi-workspace 4.18.0-193.19.1.el82.ppc64le SMP Mon Sep 14 14:48:04 UTC 2020 ppc64le ppc64le ppc64le GNU/Linux Install tools: [root go]# go version go version devel + Fri Nov 20 00:54:43 2020 +0000 linux/ppc64le [root go]# Network plugin and version (if this is a network-related bug): Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig testing\nwill fix it in Go 1.16 as release-blocker.\nI would fix the test here to be more tolerant. exact string matches on error messages aren't really an API… I would not expect that to block the golang release\nAgree too, we can think about having a substring match to avoid these issues instead!\n/assign Lets see if we can fix this in the k8s tests"}
{"_id":"q-en-kubernetes-f9766970ccd4222b6bcd610bf151cff19f58f52961a6419348e3fed09954e4dc","text":"<!-- Please only use this template for submitting reports about flaky tests or jobs (pass or fail with no underlying change in code) in Kubernetes CI --Which jobs are flaking: bazel-test-master Which test(s) are flaking: //pkg/kubelet/cm/devicemanager/godefaulttest:run2of_2 Testgrid link: Reason for failure: Anything else we need to know: links to appreciated Triage shows nothings (seems to not have updated data), but there are multiple failures in the job bazel-test-master. links to specific failures in spyglass appreciated <!-- Please see the deflaking doc () for more guidance! --\n/sig release\nThe label(s) cannot be applied, because the repository doesn't have them <detailsIn response to : /sig sig-release Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nDigging further as the step has another recent test fail ... So on the two runs reported by has 2 test failures has 1 test failure And the most recent failure in this area is a TestAdmit/match test failure but on a different test.\n/sig node\n/assign /triage accepted I'll take a look at them and fix."}
{"_id":"q-en-kubernetes-ccdb74dbd9ebc4bd41922c370ab977e3c1e001639e0c07715bc14f84fc8b75bc","text":"What happened: I was running the subpath testsuite against Azurefile. This test failed and can be easily reproduced: Basically what this test does is: a PVC use azurefile a pod that has two containers. One of it mounts the volume itself. The second container mounts the subpath of the volume. the subpath dir from the container that has the whole volume the pod, the pod will stuck in terminating status and cannot be cleaned up other than What you expected to happen: The pod can be deleted successfully. How to reproduce it (as minimally and precisely as possible): the pvc the pod the subpath dir from the parent pod the pod Anything else we need to know?: The volume is being formatted as cifs type Environment: Kubernetes version (use ): v1.18.6 Cloud provider or hardware configuration: Azurefile-csi-driver 0.9.0 OS (e.g: ): Ubuntu 20.04 Kernel (e.g. ): Linux nodepool-- 5.4.0-1016-azure -Ubuntu SMP Thu Jun 4 22:49:15 UTC 2020 x8664 x8664 x86_64 GNU/Linux Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig storage /cc /cc\nlog After debugging with and offline, the issue is that: After the dir from the subpath was removed, the subpath mountpoint on the host machine becomes into a weird state where you can still find the dir by doing a readdir to its parent. But doing or something like that will result in a \"No such file or directory\". When you do a ls on the path:\n/triage accepted\nSo the issue from code is that It will fail the path exists check while the actual mountpoint exists. So when the code logic runs to here: It will fail to remove the container dir because it somehow detect there is still file there. To fix this as a workaround, user can manually , and then the kubelet can take care of this. To fix it ultimately, we need to see if this is an issue specific to cifs filesystem? Or we should consider do an umount regardless if the path exists.\n/assign\n/assign\nTried the same scenario using NFS. It works as expected using NFS.\nYeah, that's why I am wondering if this is something special with CIFS?\nLooks like it has something to do with the azurefile-csi storage class mount options. I was getting the below error when I was trying to do ls on the subpath. So I realized there might be some issue with the file permissions. root ls ls: cannot open directory '.': No such file or directory When I removed the dirmode and filemode properties from the storageclass and performed the same steps as mentioned in the issue, the pod got deleted successfully.(Although it took a lot of time 10-15 mins for the pod to get deleted)\nFYI, I am planning to use this PR to fix the unmount issue in Azure File & SMB CSI driver: Not sure how to handle gracefully in k8s subPath code since that's common code.\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nSo this PR does not fix the Azurefile subpath test issue?\nno, that PR only fixed the unmount issue inside CSI driver. While for the subPath issue, we should fix it in k8s upstream code as you mentioned above:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:\nback to original discussion, shall we use instead of inside func? That will solve the issue of smb file PathExists check issue.\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Reopen this issue or PR with Mark this issue or PR as fresh with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /close [1]:\nClosing this issue. <detailsIn response to : The Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closedReopen this issue or PR with Mark this issue or PR as fresh with Offer to help out with [Issue Triage][1] Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Reopen this issue or PR with Mark this issue or PR as fresh with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /close [1]:\nClosing this issue. <detailsIn response to : The Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closedReopen this issue or PR with Mark this issue or PR as fresh with Offer to help out with [Issue Triage][1] Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nthis subPath issue was fixed in , ,"}
{"_id":"q-en-kubernetes-181aae799cc8ee94e98a4ca4ae02cf540a79a602ea6edbc2b4ddfec20f7725f2","text":"What happened: Adding a VolumeMount to a container with a multi-part subpath results in only the leaf directory having 777 permissions. If you define e.g. then within the volume, and will have permissions, whereas will have permissions. This means that if you want to remove and eventually from within a pod, you will need to use a privileged pod. If the sub-directories already exist, their permissions are unchanged, and the order in which these subpaths are created appears to be the order in which they appear within the spec -- if we see first, then will be created with 755 permissions and will be created with 777 permissions, even if there's another volumeMount for further down. What you expected to happen: Sub-directories created to support a volumeMount should all have the same permissions as the leaf directory. How to reproduce it (as minimally and precisely as possible): the yaml below. The first example has volume mounts so that is created before ; the second example has volume mounts in reversed order, so is encountered before , which results in being unwritable. <details<summaryThis pod will complete</summary</details<details<summaryThis pod will fail</summary</detailsAnything else we need to know?: I believe this is the return of Environment: Reproduces on minikube and a (OpenShift) development cluster I have access to Kubernetes version (use ): minikube: OpenShift cluster: output: Cloud provider or hardware configuration: Minikube: v1.15.1; docker driver; hostpath volumes (all default) OpenShift cluster: AWS, EBS volumes OS (e.g: ): Minikube: Fedora 32 OpenShift cluster: RHEL 8.2 Kernel (e.g. ): Minikube: OpenShift cluster: Install tools: minikube v1.15.1 / Not sure Network plugin and version (if this is a network-related bug): N/A Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig storage\n/assign\nGitHub didn't allow me to assign the following users: mengjiao-liu. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign"}
{"_id":"q-en-kubernetes-374a80502f6b8859498039c88f587dc8f22c2abb9974a05e714dfd17aa2252f6","text":"In reference to conversation on #sig-network-policy-api.  What would you like to be : Add a test to the netpol validation framework to validate a scenario where traffic between two pods has egress and ingress rules defined. Why is this needed: To ensure consistent behavior between different implementations of networkPolicy. For example, describes different behavior between Calico networkPolicy and Azure networkPolicy\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig sig-network\nThe label(s) cannot be applied, because the repository doesn't have them <detailsIn response to : /sig sig-network Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig network\nthanks for offering to hack on this\n/sig network\ncc\nI reviewed the current networkpolicy tests, and I believe this case is already captured in the currents tests.\nThanks for digging into this ... so Your case, yesterday was more then just combining egress and ingress. Your case , was one where Egress blocked a traffic that was allowed by other rules. Which is quite interesting and confuses alot of people. ... It highlights the 'podselector centric' nature of policies.... I think you want to do , right? In that case, Egress rule is specifically created that disallows but Ingress rules explicitly allow And the result is that Egress wins, and traffic is blocked ...\nThis is an interesting test case, I'd be happy to hack on this and look at adding a table-driven netpol test case but don't want to step on your toes if you're already looking into ! Definitely see the value of illuminating this corner case with a clear test scenario!\nThx for the feedback. You are right that it's still a different case. : I'm looking at what I can throw together today. Stay tuned, and would love your input.\nI created the initial version of the test.  The code (and comments) are likely a bit verbose at this point. Should I open a PR against this repo now, or do you want me to make changes to that code prior to the PR?\nLooks awesome, I'd go ahead with the PR!\ncan you reference this issue in your new PR ?\n/assign"}
{"_id":"q-en-kubernetes-b7e5bceb9e558da9f3a941788a29989a1d233b0b46f0b77ff1f2e63e703dc964","text":"We could either: where appropriate - least amount of code change but cumbersome OS version in the tag instead of the repository /cc\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig windows\n/kind bug"}
{"_id":"q-en-kubernetes-e13a3765f3b220183700c97220ce5674ec114ad3ff68bc827eafde8e290c4a4e","text":"[ ] two weeks soak start date : testgrid-link [ ] two weeks soak end date : [ ] test promotion pr : kubernetes/kubernetes#? According to this APIsnoop query, there are still some remaining StatefulsetScale endpoints that are untested. with this query, you can filter untested endpoints by their category and eligibility for conformance. e.g below shows a query to find all conformance eligible untested, stable, apps endpoints sql-mode select distinct endpoint, right(useragent,73) AS useragent from testing.auditevent -- where useragent ilike '%subresource%' where endpoint ilike '%AppsV1NamespacedStatefulSetScale%' and releasedate::BIGINT round(((EXTRACT(EPOCH FROM NOW()))::numeric)1000,0) - and useragent like 'e2e%' order by endpoint limit 30; example endpoint | useragent -----------------------------------------|--------------------------------------------------------------------------- patchAppsV1NamespacedStatefulSetScale | [StatefulSetBasic] should have a working scale subresource [Conformance] readAppsV1NamespacedStatefulSetScale | [StatefulSetBasic] should have a working scale subresource [Conformance] replaceAppsV1NamespacedStatefulSetScale | [StatefulSetBasic] should have a working scale subresource [Conformance] (3 rows) If a test with these calls gets merged, test coverage will go up by 1 point* This test is also created with the goal of conformance promotion. /sig testing /sig architecture /area conformance\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-444fa5b2edd377b73e5bc84b27cebb7f3765c55163f585dc8d441d2a1c256789","text":"What happened: with --cpu-manager-policy=static kubelet will panic when pause container lose during kubelet restarting What you expected to happen: kubelet can start normally How to reproduce it (as minimally and precisely as possible): 1,config kubelet with --cpu-manager-policy=static 2, Run a large number of pods on the test node 3,stop kubelet use command: 4,kill and remove all the pause container like this: ! 5,start kubelet use command: and then kubelet maybe panic, log is : Anything else we need to know?: When we actually encountered this problem, only one pause container was lost. In order to reproduce this bug, I deleted all pause containers. Environment: Kubernetes version (use ): 1.20.1 Cloud provider or hardware configuration: x8664 OS (e.g: ):18.04.5 LTS (Bionic Beaver) Kernel (e.g. ): Linux ecs-linux-0008-43ee 4.15.0-123-generic -Ubuntu SMP Wed Oct 21 09:40:11 UTC 2020 x8664 x8664 x8664 GNU/Linux Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig node /assign\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\nStale issues rot after 30d of inactivity. Mark the issue as fresh with . Rotten issues close after an additional 30d of inactivity. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle rotten\nRotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . Send feedback to sig-contributor-experience at . /close\nClosing this issue. <detailsIn response to : Rotten issues close after 30d of inactivity. Reopen the issue with . Mark the issue as fresh with . /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI think it makes sense to reopen this issue as the referenced PR is still open.\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/remove-lifecycle rotten\n/triage accepted\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nI am seeing this issue with a customer also. What is the timeline for this PR to be reviewed and merged?"}
{"_id":"q-en-kubernetes-ee0826757468ea334f95c77509b257c3d392b6713b4311ba7afa125d642dcec9","text":"The based container images for are shipped with the wrong architecture within the manifest if not amd64: Refers to We updated the images a few minutes ago to buster-v1.3.0, which are now ready to be used as base for the multi-architecture images. If the images contain the right architectures, then it may magically happen that the kube-proxy images are fixed, too. This is because we already use the build syntax for it: /cc\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig release"}
{"_id":"q-en-kubernetes-3444fabb883927bdace65eb5d9db61f45e39011ade9a032d43fca15b1006c909","text":"What happened: The current in-tree PV topology is still using beta version label \"failure-\". This was intentionally not upgrade because of CSI migration. However, since the CSI migration timeline is likely to slip, and with the fix of , it should be safe to update the in-tree PVC to GA version () So that we can unblock removing the beta label from node. What you expected to happen: Replace all beta topology label to GA version once merged. /sig storage /cc\n/assign\nGitHub didn't allow me to assign the following users: kassarl. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\ngce-pd topology update already merged:\naws-ebs topology update is here:\n/triage accepted\ncinder topology fix is here: Waiting for topology label update.\nOpenStack Cinder:\n/reopen Has vsphere made the fix?\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nNot yet. is going to make the fix."}
{"_id":"q-en-kubernetes-c41f05d8910be5009b755790528ec94d72d551b527b45c1ed91bde3c53341f9a","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: When reviewing Windows kubelet logs noticed errors like: What you expected to happen: The stats should be returned How to reproduce it (as minimally and precisely as possible): This occurs when the pods has just started or during termination. To simulate the issue: Then deploy pod: Anything else we need to know?: The container is in a state but is has not started running so the call to get the stats fails: This is handled in other places like moby: and Environment: Kubernetes version (use ): 1.18+ Cloud provider or hardware configuration: aks-engine OS (e.g: ): Windows Kernel (e.g. ): 2019 Install tools: Network plugin and version (if this is a network-related bug): Others: Dockershim /sig windows /priority critical-urgent /assign\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\ncc:"}
{"_id":"q-en-kubernetes-bfd32e3b4446984b607eec8f1f52f6ed2d0a69f7505b06c358a314b66387192a","text":"What happened: What you expected to happen: How to reproduce it (as minimally and precisely as possible): Run the netpol test suite. Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/triage accepted\n/sig network"}
{"_id":"q-en-kubernetes-ea38d7e555d00e2e1e1f80a069d48e0be42991c74b8337dc8123f04000087f50","text":"MetricsGrabber tests failing when scheduler & controller-manager not visible. In managed offerings, like GKE, scheduler and controller-manager are not visible causing BeforeSuite of MetricsGrabber to fail. Test is failing then control plane pods can't be found (as on GKE): HasControlPlanePods was in /assign\n/sig instrumentation\nJust removing this check allows tests to run without problems (some skipped, one passing).\n/triage accepted"}
{"_id":"q-en-kubernetes-4a22b7be5e953790da465f200505d8aba99618aabc30b0801e44f633857f7538","text":"[x] two weeks soak end date : 26 Feb 2021 According to this APIsnoop query, there are still some remaining RESOURCENAME endpoints which are untested. with this query you can filter untested endpoints by their category and eligiblity for conformance. e.g below shows a query to find all conformance eligible untested,stable,core endpoints Create a ReplicaSet with a static label Read the ReplicaSetScale Update the ReplicaSetScale Get the ReplicaSetScale to ensure it is updated Patch the ReplicaSetScale Read the ReplicaSetScale to ensure it is Patch Delete the Namespace and ReplicaSet Due to the complexity of setting up the resources for APIService we have used the current e2e test as template. It has been extended in a for review. example endpoint | useragent ----------------------------------------|\npatchAppsV1NamespacedReplicaSetScale | (linux/amd64) kubernetes/$Format -- [sig-apps] ReplicaSet Replicaset should have a working scale subresource readAppsV1NamespacedReplicaSetScale | (linux/amd64) kubernetes/$Format -- [sig-apps] ReplicaSet Replicaset should have a working scale subresource replaceAppsV1NamespacedReplicaSetScale | (linux/amd64) kubernetes/$Format -- [sig-apps] ReplicaSet Replicaset should have a working scale subresource (3 rows) If a test with these calls gets merged, test coverage will go up by 3 points This test is also created with the goal of conformance promotion. /sig testing /sig architecture\n/triage accepted\nPresented in Conformance meeting of 9 Feb 2021 and it was agreed that that method see to achieve the purpose and a PR can be created. A similar approach would be taken for the DepoymentScale endpoints.\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-ea5dfa8ca4c60a50617e7c454844a28b8df064eecf8f8671a6154c2ab71f66c0","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: What you expected to happen: The test should pass How to reproduce it (as minimally and precisely as possible): I think the key here is being on a linux host with cgroupsv2 enabled. Anything else we need to know?: Environment: Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others: /sig node\n/assign\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-613d127cd5afdadcbee3c136bb9a085f17e19f478307e1e6bd53399b9c176078","text":"file was found 'empty'. It blocked running etcd in a fail loop with Before this issue: Data disk (etcd) run out of space due to . .txt be not corrupted. It should be updated: only when needed (infrequently) safely (rename swap or writing on pre-allocated space) migrator script that runs etcd in case of the file not being found. Use whole space of etcd data disk and restart migrator+etcd when concurrent process is disk hungry. Requires changes of this logic: This logic (file write) should not trigger if the minor version matches: The write itself: should be defensive: Please notice that truncates file before writingKubernetes version: 1.18 Cloud provider or hardware configuration: GCE\n/sig api-machinery\n/assign /triage accepted\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\n/remove-lifecycle stale /lifecycle frozen"}
{"_id":"q-en-kubernetes-9e77c1eae275c1db9b153ebc2059acc2c6dc209d44d586853b1d6ddfc8aedcc0","text":"was removed in 1.20 in favor of a plugin. However, this is not included in release artifacts. The binary is included in release artifacts, and ideally easily installable via kubectl plugin managers like . /sig cli /assign\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThis question came up in a docs review of a page referencing () Once this is resolved, please ensure the website is updated to document installation/use of this tool.\n/sig release"}
{"_id":"q-en-kubernetes-fa126d79950d466047e6ed1ac3816f4c853174133d960101916fac2981f766f8","text":"on the new node fails sometimes. The following error appears in the log: on the new node succeeds every time Join a control plane node The node has over 40 GB of free space on its root partition. I some more logging to the following functions: , , , , , , It appears that the ephemeral storage is detected a bit too late: /sig node storage Kubernetes version (use ): 1.20.4, commit Kernel (e.g. ): 4.9.252-1.ph2\n/assign I will try to reproduce it today.\nLocalStorageCapacityIsolation is a feature gate for kubelet. Can you check if you disable it? ReadyCondition will check ephemeral-storage as required capacities if this feature gate is enabled in kubelet configuration. Would you check it first?\n[x] 1.8 alpha ephemeral-storage [x] 1.10 Promote LocalStorageCapacityIsolation feature to beta [x] 1.15 KEPs Quotas for ephemeral-storage for LocalStorageCapacityIsolationFSQuotaMonitoring default to false is Alpha since 1.15 [x] Failed to promote LocalStorageCapacityIsolationFSQuotaMonitoring feature to beta [x] 1.20 Honor disabled LocalStorageCapacityIsolation in scheduling I am no sure whether there is a plan to GA LocalStorageCapacityIsolation?\nI tried to fix it in . Need some testing later.\nIt appears that the pod fails before sees the storage capacity:\nWould you try this ? I build it with on PR branch.\nI tried to apply the PR to 1.20.4. After four successful s, one failed. I got the following error:\n! It means that during the node resource updated the first time, the cadvisor is not available yet. If we wait here, it would be a problem. I would think more of how to fix it.\nHi, Same problem here. I'm using K8s v1.20.4 with CentOS7 (2 cores, 4GB and 20GB storage) Cluster creation with kubeadm init always works, but when I try to add several master nodes with kubeadm join it takes a lot of time waiting for etcd pod to be created. In most of cases, etcd pod never runs. The rest of components (APIserver, controller, scheduler and proxy) exists and works (except APIServer which reboots trying to connect the missing etcd) Checking the kubelet logs in the missing etcd hosts I have found this: Mar 10 14:23:27 kmaster-02 kubelet[9448]: W0310 14:23:27. 9448 ] Failed to admit pod etcd-kmaster-02_kube-system() - Unexpected error while attempting to recover from admission failure: preemption: error finding a set of pods to preempt: no set of running pods found to reclaim resources: [(res: ephemeral-storage, q: ), ] That node have 17GB storage free. If kubelet is restarted, the etcd (pod and container in docker ps -a) appears. If I could help doing something please tell me. I would like to help\nFor anyone who is experiencing this issue and still wants to use kubeadm until it is fixed: You can remove the ephemeral-storage request from the etcd static pod and the join will work. Just make sure that your nodes actually have more than 100MB free storage left. You can reinsert the storage request by patching the etcd manifests after the nodes are up. To patch etcd during a create the following file (the filename matters, the file location doesn't) and use to join additional control plane nodes.\nThanks for the workaround I'm encountering this same issue and the workaround worked for me. Edit: the patch I tried doesn't work though (not sure if this is really a problem). I issued the command: and the response shows that the resource can't be patched this way: There may be another way but I'm still fairly new to Kubernetes so it is not obvious to me at the moment. Anyhow, thanks for the suggestion!\nI updated Add a start-delay of 10 seconds for a condition that the capacity of node ephemeral-storage is 0. Not sure if this will fix it.\nwe are starting to see reports about this every week.\ncc\ncc\nFrom it sounds like the issue is kubeadm etcd ephemeral-storage requests are too big for some environments? So far this is sounding like a kubeadm issue, not a kubernetes issue.\nit seems to be something else. i doubt that users are trying machines that don't satisfy 100Mi. for example, the report in the OP points out that the machine has 40GB.\nI can confirm that the VMs on which I installed k8s control plane components had sufficient disk space when I encountered the problem. I'm running Kubernetes v1.20.4 and kubeadm v1.20.4.\nI think the root cause here sounds right. There is kind of race here. adding delay will help, but it does not have 100% guarantee. Not sure whether it is easy to add logic to make sure capacity is reported correctly before starting pods..\nDid something change in 1.20 regarding the ordering of cadvisor and node startup?\nIf the reason would be a disk problem(or something we can figure out) and the delay can solve it, I think we should add a warning besides the delay.\nWe need to figure out what triggers the race condition. This only happens on some machines. I have not seen this myself.\nWe're able to repro this nearly 100% of the time with Cluster API Provider Azure + k8s 1.20.x + Ubuntu 18.04 + default template, which makes k8s 1.20 pretty much unusable at this point. Trying to see if this repros with other non-Azure cluster-api providers with kubeadm 1.20.\nDoes anyone know why this only repros on and not ? The ephemeral storage resource requirement here is used in common by both and : In practice I can workaround this issue in cluster-api + Azure by adding a sleep command to my spec, but it requires a significant number to make work ~100% of the time (like 90 seconds). Does the flow take significantly longer before it gets to the part where it submits an etcd pod to the apiserver? Is there any other explanation for this being a -only symptom?\ni can see this failing for any static pod with ephemeral storage requests. that is, unless there is proof that this is somehow specific to the kubeadm static pod for etcd that kubeadm deploys.\nI agree that the specific kubeadm-delivered etcd pod is not interesting. (In fact it's very permissive, it's only asking for 100Mi.) What definitely seems to be happening (in Azure) is that kubelet takes a long time to know about its underlying node's ephemeral-storage availability. Kubelet starts: ~2 mins later, it complains that it can't schedule etcd: Again, why does this not repro on nodes created by ?\nit also seems to be an even more scoped problem because it's not reproducible for me locally (i create a number of clusters every day, usually), the kubeadm CI or in other high level projects that use kubeadm (e.g. kind, minikube, etc). we should try to find a common reason / condition from all the reporting users that are seeing this.\nExperiencing same problem on openSuSE Kubic/Tumbletweed snapshot with kubeadm/kubelet 1.20.2, kernel 5.11.15 running on Xen, 2 CPU, 4GB ram 30GB disk (BRTFS)\nProblem consistently appear/disappear when switching between snapshots (Working) and , so a package update in that period has probably triggered this. Following files contains changes between those two snapshots.\nare you suggesting a root cause in underlying Linux storage driver surface area that is adding delay/uncertainty to kubelet's ability to introspect available ephemeral-storage on the node host OS?\nA bit of luck combined with 6th sense i got it narrowed down to systemd update here Looking at the changelog, one item sticks out: Anyone else enabled cgroupsv2 on i.e. the CentOS cases (it needs to be enabled explicitly for that AFAIK)?\nACK on your latest respond regarding unified cgroup hierarchy. Is there a quick, cross-distro way to verify that any folks who are repro'ing have a similar systemd config?\nThe precence of file/dir /sys/fs/cgroup/cgroup.controllers should mean you are using cgroupsv2 ( Ref: ) I can confirm that the file is present on the kubic snapshot which is not working. It is not present on the one that does work. ( I'm using cri-o as runtime which should support cgroupsv2 ) cgroup2 support was introduced for K8S in 1.19 ( ).\nThanks. So tl;dr your working theory is that cgroupsv2 + kubelet ephemeral-storage admission has some pathologies. I'll post quickly with what my repro containerd + Ubuntu 18.04-LTS OS surface area looks like.\nI ran into this issue consistently on Ubuntu 20.04.2 LTS using cri-o but is not present on my systems.\nI can also confirm that cgroupsv2 is not a factor in my repro'ing environment:\nAlso, I'm not using cri-o, containerd here:\nOk - so much for that theory. But still: What I did to narrow it to systemd-update was taking the working kubic snapshot () and run zypper update systemd which updates systemd (along with systemd-sysvinit and udev packages to same version). That consistently made it fail. So I think a pointer to the root-cause might still be hiding somewhere around this changelog - even if it is not cgroupsv2. That part was just a theory (seemed like a \"big\" thing), but the update of systemd package is fairly hard evidence.\nI have identified where these particular symptoms (probably) originate, and cursory testing suggests something like this eliminates the issue: I'll be advocating for a few mins in today's sig-node discussion if any folks following this thread want to chime in. cc\n/sig cluster-lifecycle /triage accepted\nI've also observed that on some occasions the node allocatable storage is in a continual state of reporting \"0\". Restarting kubelet immediately fixed that. More debugging to do....\ndid you manage to reproduce this with an arbitrary static pod that is not the etcd one?\nI was not able to reproduce in a non-kubeadm test, but I suspect that the way kubelet bootstrap is implemented there (and possibly environmental reasons) the static pod handling was always happening after this code executed: (the above is in the containerManager method) kubeadm w/ ephemeral-storage requirements in etcd is the most effective repro at this time\nthe writing of the etcd pod manifest in during is after the kubelet has already started and i wonder if that is contributing to the observed failures. probably not given this is a cadvisor related problem. to verify that i would write an arbitrary static pod manifest with the same request in : before (which also starts the kubelet) right after the kubelet has started if is causing such racy problems i don't think it should be graduated to GA. though, it has been Beta for ~12 releases (on by default), so it's already used everywhere.\nBased on your assessment do you think we should remove the ephemeral-storage requirement from default kubeadm etcd static pod? I can make that change.\nbased on the investigations here (thanks all), it feels like is a feature that may or may not work (for static pods at least). it doesn't feel right to mitigate a kubelet bug on the side of the deployer, but given how many reports there are about this i think it should be done at this point. happy to LGTM approve your PRs. note it must also be backported to all versions in support (back to 1.19). suggesting the following release note: we should keep this issue open, too.\nSounds good. I'll submit a PR soon with the kubeadm change, and I'll continue to drive toward a root cause solution on the kubelet side. I'll bring this up at sig-node tomorrow. Thanks!\nis a possible duplicate of this issue - seeing a CrashLoopBackOff failure for kube-apiserver under a similar scenario to those mentioned here, e.g. failure to join a third control plane node with mentions of in the kubelet logs.\nAn ugly as sin workaroud: (please don't hate me :D )\nIt seems like this could be closed with and related cherry picks now being part of the newest releases? Bumping to 1.21.2 allowed me to remove the hack for etcd and several test deployments have steered clear entirely of the original issue. Thanks\ngreat news. I'd prefer to pin this issue to this PR: That PR will deliver more race condition resilience for all static pod + ephemeral-storage requirement scenarios.\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:"}
{"_id":"q-en-kubernetes-a06ff54733d6ac38b4f9939973a6ee6122f6dc6b2f472cc33182e26504501d9f","text":"Currently whenever a vSphere volume is provisioned, we are looking at all nodes in the cluster and then fetching their hosts and then their datastores and then ensuring that selected datastore in present on all the nodes. This is done every time a new PVC is provisioned. It can be pretty expensive in large clusters and number of involved API calls can be big. /sig storage /area vsphere cc\nThe label(s) cannot be applied, because the repository doesn't have them <detailsIn response to : Currently whenever a vSphere volume is provisioned, we are looking at all nodes in the cluster and then fetching their hosts and then their datastores and then ensuring that selected datastore in present on all the nodes. This is done every time a new PVC is provisioned. /area vsphere Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/triage accepted\n/assign"}
{"_id":"q-en-kubernetes-947136ea8248fb75219d4a2780e225b9730b8db4ff2d00fa7ad31a0a1295b1db","text":"Seen while running on If hack/update- fails, the output should make it clear Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others: /area code-organization /sig architecture /cc /assign\nMaybe we can add a to check the exit code.\n/cc\nsomething like that seems like a good idea\nusually when we want to do stuff like that we instead record the exit code for specific steps with something like so we can check after and inform the user of the specific failure (and suggest what to do), then\n/assign"}
{"_id":"q-en-kubernetes-008ca2b69bf6f84855aa7fc74e613d0733365e6cd47dcc25c8f18519fbaaf463","text":"Scheduler throughput drops significantly in http://perf- ! After , seems to be the curlspirit.\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/priority critical-urgent /sig scheduling /sig scalability"}
{"_id":"q-en-kubernetes-c392df55ab7dc72e13e65c63fc1f61f730cca43a7c22b90fcf3fe28c2fb22e91","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --Serializing a v2beta2 HorizontalPodAutoscaler with (and ) produces yaml that cannot be applied due to a ValidationError on the field which is under the optional field . A valid HPA serialized with should be able to be applied to the cluster with no validation errors, without filling out the optional field. Other valid objects (such as Namespaces and Deployments) can be applied directly, without filling out their optional fields. Run this to generate the serialized yaml (same outcome with both and functions) Running the go code should yield this yaml: Writing the above yaml to a file and applying it (with ) will result in: Kubernetes version (use ): Server: v1.20.2, Client: v1.20.4 go version: go1.16 darwin/amd64 has: cc cc\n/sig api-machinery /sig autoscaling /sig cli\n/assign /triage accepted\nmarking conditions as +optional in the API type would resolve the client-side apply error\nIt would remove the validation error, but I'd be concerned about any side effects or unwanted outcomes when moving that field from required to optional. A somewhat analogous field is required. But if you anticipate this won't change much else, that will work.\nClient-side validation is non-authoritative, and problematic for the reasons described in this issue ... the client is disallowing a request the server would actually have no problems with. No field the server would permit to be empty (e.g. succeeds with ) should be marked as required for this reason. fwiw, openapi schema generation marks fields tagged as as optional:\nIssues go stale after 90d of inactivity. Mark the issue as fresh with . Stale issues rot after an additional 30d of inactivity and eventually close. If this issue is safe to close now please do so with . Send feedback to sig-contributor-experience at . /lifecycle stale\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:\n/remove-lifecycle rotten"}
{"_id":"q-en-kubernetes-401bd803bf545c47b9f079f4e133a5da93cd748cff1afbf545e3f20a5f25f893","text":"<!-- Please only use this template for submitting reports about flaky tests or jobs (pass or fail with no underlying change in code) in Kubernetes CI --ci-kubernetes-e2e-ubuntu-gce [sig-storage] Subpath Container restart should verify that container can restart successfully after configmaps modified links to appreciated links to specific failures in spyglass appreciated Triage shows that this has been flaking quite frequently: It has flaked in 5 of the last 10 runs: /sig storage /priority important-soon <!-- Please see the deflaking doc () for more guidance! --\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nNoticed some more flakes on this one\nYep, looks like this is still flaking quite frequently: Going to re-ping SIG Storage.\n/assign\nTest is flaking maybe half of the time. ~This test is failing in gce-ubuntu-master-default - not sure if it's failing in any other dashboards, nor whether this is run anywhere else.~ edit: This is failing on Ubuntu only. Not sure if this is limited to master branch. Fix to the original issue was in since 1.19. This is a subpath test which a configmap a container that mounts the configmap a liveness probe and waits for the container to restart. liveness probe and waits for the container to stop restarting. The test failed at the last step - the container never stopped restarting. Timeout is 2min. The liveness probe continuously checks whether a \"probeFile\" exists in (an emptyDir), and if it doesn't exist, whether the configmap file contains the modified value. The probeFile essentially acts as a condition which guards the probe from checking the configmap file. The test fails the liveness probe by removing the probeFile without updating the configmap just yet. It then fixes the liveness probe by updating the configmap to the modified value. So if the container keeps restarting, the liveness probe could've been failing continuously within 2min, due to one of the following: (? not sure about this one) Command can't be executed (for example if container can't start) configmap file was never updated. Within 2min, the pod only restarted 5 times. The liveness probe's period is 2s, initial delay is 1s, and failure threshold is 1, so a restart due to liveness probe failure is triggered every 3s. The image is busybox. The command is just , so it's unlikely for the container to have crashed. There were no liveness probe unhealthy events after the test fixes the probe, and yet it continued to restart. No more events after probe is fixed.\nCould we check if the configmap value was updated for sure to rule out that scenario ? I have not tried to run an end to end test locally :) so am not sure how to debug it\nWill do! I ran the test in a GKE Ubuntu cluster running 1.19.6 and couldn't reproduce the failure. I'm now trying to run a cluster on GCE with Ubuntu nodes running k8s master, but haven't been successful after trying a number of techniques. If anyone happens to know how to do this let me know; I'll keep asking around otherwise.\nI got a repro on a GKE cluster backed by COS. After the test failed the container ends up in Running state eventually, and the configmap file outputs the expected modified value. The fix is to increase the test timeout. PR coming up."}
{"_id":"q-en-kubernetes-32aefc1c9667deeac25c7fa9b9550d239ed1fdf9d6a45f9e42a1f0f724022acf","text":"found this in while finding beta API dependencies for 1.22. This will block from running at all in 1.22 I think this is for this deployment: , and in ingress-gce I found\n/sig network /triage accepted I pinged some GCE sig-network folks in the PR\ngiven that uses extensions/v1beta1 Deployments, that particular manifest seems unused\nonce we land the fix in , anyone know how to update the image we use in e2e tests?\n/assign The update to use v1 APIs is merged to in What version should be tagged we can pick up in Does need to be picked back to one or more release branches?\nIt would be helpful to get a version containing the fix tagged, so the manifest can be updated in to get e2e signal even before opens for 1.22.\nAck. We are planning to cut a release soon. Once the release is cut, I will update the k/k to use that\nwhat's the ETA for cutting ingress-gce release? I think it might be related to failing pull-kubernetes-e2e-gce-100-performance run in"}
{"_id":"q-en-kubernetes-bd154303d20599beecef17a46b6ca985c48134dc369a3996d772f4c6675030ec","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --In , client-go don't share transport when c.Dial is not nil, but kubelet custorm Dial by default, After PR is integrated, the connections between kubelet and kube-apiserver in our cluster with 4000 nodes increases by five times. kubelet share transport by default, one kubelet only keep one connection to kube-apiserver. start kubelet with rotate-certificates=false Kubernetes version (use ): v1.19.4 Cloud provider or hardware configuration: OS (e.g: ):EulerOS 2.9 Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig api-machinery /sig node\n/triage accepted /assign\nhow many connections between kubelet and kube apiserver on one node after PR in your case? it's mentioned it increases by five times, does that mean before the PR, there is one connection, after the PR, there are 5 or 6 connections?\nbefore the PR, there is one connection, after the PR, there 3 connections, there are some customization capabilities in our product, and we'll have another two connections. In our kubelet startup parameters, rotate-certificates=false\nSome increase was expected, since multiple REST clients constructed from a config with a custom dialer cannot safely share a transport, and client-go constructs a REST client for each API group/version accessed. Something like would be required to rework how client-go constructs clients to start sharing REST clients above the transport level between API groups/versions. I think this is working as intended until client-go client construction is reworked.\ndo you know where in the code the 5 connections are created? are they all safe to share the transport? also are the 5 connections all active all the time? if not, IdleConnTimeout should cause the connection to be closed.\nI test the code on master branch, if I change rotateCertificates to false, there are 6 connections between kubelet and kube-apiserver ! !\ntransport will share when config.Transport not nil if rotateCertificates is true, config.Transport will been set in kubeletcertificate.UpdateTransport but if rotateCertificates is false, config.Transport will not been set.\n/area kubelet Any update on this issue?\nI will work on this issue\nHi if rotate-certificates=false kubelet customizes the dialer to provide a closeAllConns to close the connection when the connection is dead but not been closed. In , there is a discuss about another solution after client-go has HTTP 2.0 health check, so we don't need this change. So we can solve the current problem by reverting . I have submitted to revert , and this change requires your final confirmation. Thanks you.\nare we guaranteed kubelets always speak to kube-apiserver over http/2? if they use a load-balancer that is not http/2 capable, isn't the \"close all connections\" fix still required as a backstop?\nI thought http1.1 wasn't affected because it will always try to connect a new connection\nI'm not confident in that. The original issue dated back to before client-go defaulted to http/2, and we definitely saw issues with stuck TCP connections then.\nyeah, I could reproduce it with http1.1, however, it can be solved using the transport CloseIdleConnections method please take a look , the last commit needs more work, but I think that the 2 first commits exposeing the CloseIdleConnections() are useful\nok, this seems to work , but it will be nice to have more feedback would you be able to test and see if this solve your issue?\nHi aojea, I have test on v1.21.1 before merge , there are 6 connections to kube-apiserver from kubelet ! after merge , there is only 1 connection to kube-apiserver from kubelet ! so, can solve my issue. thank you very much."}
{"_id":"q-en-kubernetes-aba51dd96df40d0beb046926abc87ad433d09ca6fbfd961a87056932acd3b90a","text":"is wrong as it says default label can be specified, but KEP and other lines of changelog mentions annotation. Changelog is correct and consistent about default container feature for kubectl. I'm still not able to get this feature to work actually for . Kubernetes version (use ): v1.21.0\n/triage accepted thanks for reporting early!\n/remove-kind bug /kind documentation\n/sig release"}
{"_id":"q-en-kubernetes-5dbeddbf5de175404f5522b2077e28a64b79b608e6265257d0db899db5393953","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --panic as below when an Endpoint doesn't have zone. should print zone as if an Endpoint's zone is unset. not set \"\" for Nodes. Pods and export them as a service. kubectl to describe the corresponding EndpointSlice. Kubernetes version (use ): v1.21.0 Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig network /sig cli /assign"}
{"_id":"q-en-kubernetes-bbd0dea25958876674448dd36e75dcb0eab526dc9c6df3a993f93a241b35e43d","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --The default fluent-bit configuration for GCE Windows nodes produces unparsed JSON logs and does not forward container logs. Container log output is produced. fluent-bit on a given windows image. a GCE Windows e2e cluster and view the stackdriver output for pods. Kubernetes version (use ): 1.20+ Cloud provider or hardware configuration: GKE/GCE OS (e.g: ): Windows Kernel (e.g. ): Any version Install tools: fluent-bit Network plugin and version (if this is a network-related bug): Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig windows"}
{"_id":"q-en-kubernetes-0d31b529ad18ab8ac0c277cb97efa356e067ca8611b5a6622e5e9dbc9e5cd06a","text":"I am aware that the following is a corner case that does not occur in day-to-day business, but it took me a long while to debug, so I thought I would raise this issue to save others from the same pain. Deploying the following job with empty imagePullSecrets leads to it being stuck in forever, and status looking the following way: The status should look something like: This would allow my autoscaler to kick in and add the corresponding nodes. Run Kubernetes version (use ):Cloud provider or hardware configuration: AWS EKS OS (e.g: ): MacOSX 11.3 Kernel (e.g. ): Darwin Kernel Version 20.4.0 Install tools: brew\n/sig scheduling (I assume)\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /sig scheduling (I assume) Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign Scheduler log\nShould we instead forbid empty items in ? Is there a use case for using as ?\nFor me personally that would be totally fine I had no use case for it. It was more produced by chance as I wasn't aware that I had to handle the case separately. An error which led to an early failure would have been fine by me and solved all my problems in tracking this down.\n/assign as\nForbidding it is the appropriate solution. However, we can't do it for backwards compatibility of the API. /triage accepted"}
{"_id":"q-en-kubernetes-42ade061097ae5f80973c5fd4b117fd96ad5db21a1537dde34d7fd47084d9373","text":"This issue is a bucket placeholder for collaborating on the \"Known Issues\" additions for the 1.22 Release Notes(). If you know of issues or API changes that are going out in 1.22, please comment here so that we can coordinate incorporating information about these changes in the Release Notes. /assign /sig release /milestone v1.22\nGitHub didn't allow me to assign the following users: simcard0000. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : This issue is a bucket placeholder for collaborating on the \"Known Issues\" additions for the 1.22 Release Notes(). If you know of issues or API changes that are going out in 1.19, please comment here so that we can coordinate incorporating information about these changes in the Release Notes. /sig release /milestone v1.22 Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/triage accepted\n1.19 or 1.22 ?\n/assign\nFor the 1.21 example in , should be changed to\nPlease add the following to the Known Issues section of the 1.22 Release Notes: CC\nCan you also mention the CSIMigrationVSphere feature gate in the note as well?\nwhat should I mention about the CSIMigrationVSphere feature gate? If someone uses vSphere CSI Driver, he/she should not upgrade to v1.22 whether CSIMigrationVSphere is on or off. If someone does not use vSphere CSI Driver, then we also don't need to worry about this feature gate.\nIf the feature gate is enabled, they should also hold off on upgrading to 1.22\nThe main concern as far as k8s release is - it is intree to CSI migration of vSphere volumes that is broken in 1.22. So I think our release note should say something like:"}
{"_id":"q-en-kubernetes-5d3ee62bd648d090ba5f60cffe2913263c9a6e8cf4fbf17ccb827dfa4692ba2c","text":"Expose totalAnnotationSizeLimitB in the validation package: This value is useful for validating annotations outside of the validation package, such as in SSA: Otherwise, size validation on annotations outside of the package would diverge and cause issues later. /kind cleanup /wg api-expression cc\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign"}
{"_id":"q-en-kubernetes-7b39b114e3f8b25c65861c9c397d758891e478151ec492017dfbfbc72cee16e3","text":"[X] Test PR: kuberenetes/kubernetes [x] Two weeks soak end date: 1 July 2021 [x] Test promotion PR: kubernetes/kubernetes According to this APIsnoop query, there are still some remaining StatefulSet endpoints which are untested. Create a watch to track deployment events. Create a deployment with a static label. Confirm that the pods are running. Get the deployment status. Parse the response and confirm that the deployment status conditions can be listed. Update the deployment status. Confirm via the watch that the status has been updated. Patch the deployment status. Confirm via the watch that the status has been patched. Using an existing as a template for a new for deployment lifecycle test. This query shows the endpoints hit within a short period of running the e2e test If a test with these calls gets merged, test coverage will go up by 1 point This test is also created with the goal of conformance promotion. /sig apps /sig testing /sig architecture /area conformance\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-0bd6fcf7065619d9a1d0932d28d8889dbc56a60ae38b5c44380960e04fbf80d7","text":"As part of the Kubernetes upgrade workflow we have in place, we are upgrading kubernetes from 1.15.3 -1.18.15 (both versions and all versions in the middle are custom patched for different behavior of the and . None of them having anything to do with kube-proxy's IPVS), and as part of this upgrade we are also migrating the from to mode. We have calico 3.12 running on the cluster as part of the CNI stack. Once the pods are migrated from to mode by applying the new DaemonSet manifest, the cluster has a series of stale rules for chain left around that breaks the runtime state of the pods. have cluster running in HA mode (3 Node cluster) where all nodes of the cluster are running as leaders and take the workload as well. There is no non-leader Pods. have customized the kubernetes and calico infra to enable the usage of Link local IP address in range for internal usecases Most of the IPs and the names have been mangled. Stale IP table rules to be cleaned up once the pods and endpoints are stable with all endpoints in Ready address mode. down the pods on a given namespace to 0 (leave the service and Endpoints around) the control plane to 1.18.15 (take the intermediate hop path. 1.15.3 -1.16.7 -1.17.5 -1.18.15) apply the kube-proxy manifest provided above. up the pods scaled down in Step 1. Current workaround we have been putting it use the as an init container and run the and then bring up the in IPVS mode and that seem to take care of the stale entries (but this seem to be too much of a big hammer approach of cleaning up) Kubernetes version (use ):Cloud provider or hardware configuration:OS (e.g: ):Kernel (e.g. ):Install tools: kubeadmNetwork plugin and version (if this is a network-related bug):Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThe label(s) cannot be applied. These labels are supported: <detailsIn response to : /label area/kube-proxy Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /sig sig/network Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig network\nI don't think that the live migration between kube-proxy implementations iptables<-ipvs is supported, at least I never heard about that and seems tricky. you use to have all the historical context, what do you think?\nWe tried implementing a cleanup mode but it was problematic and removed. (see ) You can run but that's pretty coarse. Maybe we should add a value to that flag to say which mode to try to cleanup? PRs welcome. The simplest answer is to reboot.\nLet's make this a docs issues - we should write this down\ncan we merge the above so that we can close this issue?\nnot just yet :-) wouldn't signal registration and calling clean-up would at least clean things up on graceful shutdown? Granted rules will not be there until kube-proxy is restarted."}
{"_id":"q-en-kubernetes-8aeed6724c27cddf2cc70c9bec327ee20c01cdda59dcd5cd082a6d29e4dc5236","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --kubelet panics on creation of inline ephemeral volume with FSType specified. creation of inline ephemeral volume succeeds with FSType specified. no existing isolated test case. Happens during e2e testing of pmem-csi. Kubernetes version (use ): The pmem-csi CI system where this was detected, runs same tests on k8s systems with 1.19 and 1.20. The problem occurs in testing on 1.20 system, problem does not occur on 1.19 system.Cloud provider or hardware configuration: pmem-csi tests run on cluster formed by few qemu-based systems running Fedora-33 and emulated pmem storage. The issue has been seen in e2e testing in , where a new test was recently introduced that creates a pod with an inline ephemeral volume. If the related VolumeSource has filesystem type set to empty \"\", or FSType is omitted in spec, then the pod creation works OK. In that case, pmem-csi chooses it's default fstype which is . Example of working definition of VolumeSource: If the filesystem type in the VolumeSource is set to or then kubelet fails to create the volume and pod. A panic trace is written in the node kubelet log and the creation will be retried few times, On pmem-csi side: the 1st NodePublishVolume is received, namespace and volume are created, formatted with mkfs, mounted and NodePublishVolume returns OK. Same request is received few times with increasing time interval. e2e testing code which tries to create the pod, will fail with:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\n/sig node storage\n... and ownership changed for those volumes."}
{"_id":"q-en-kubernetes-64e2dd7a9a7961a18efa8850e8dfe40f31f4c0585f7d06bddc3f08c803cd6a78","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --When I ran , it failed with the following error: I expected the build to successfully complete, or to be warned that I am missing a required build dependency. If you don't have docker buildx installed, you will hit this error. However, that error is masked when you pass the full command line: Appears to have been introduced in Kubernetes version (use ): HEAD (1.22.x) Cloud provider or hardware configuration: N/A, local build OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig release\nThat's not the same as in the script, which should be setting which ungates this command. (we depended on this env for much longer in the project to use ) Stable docker versions should have buildx, at least behind this env. unfortunately the project has never validated a minimum docker version for building.\nbuildx as a build requirement was actually introduced at least as back as 1.21 as well, for some of the core images. we've just standardized on it more since then (including the linked commit), so if we do add some validation we should probably backport it. I think docker buildx has been available since 19.03?, but it appears your install is too old to have the flag in buildx. we could also try switching to , I'm not sure which version introduced .\nNo difference with the flag. But I am not running upstream Docker, I'm running Docker out of the Debian repos.\ndiscussed on slack: seems debian packages don't have the buildx plugin , while docker-ce / docker desktop do. we should at least be checking for the command and throwing a clear error like we do with e.g. tar.\nFiled to discuss the expectation users might have re: the docker PPAs vs the Debian repos and buildx availability.\n/\nalso\nNote: reverting fixes the issue, so I have been building locally with that as a carry patch as a workaround.\ncc apparently there are docker packages that don't ship this, which maybe changes the equations a bit, see above.\n/assign\ntl;dr from isn't actually part of moby, but Docker Inc. ships it as part of their Docker packages in their PPA. It's a separate tool with a separate code repo.\nwould it be eligible for another package? otherwise we probably need to provide support for installing it at least to some extent, e.g. if you are on mac with ancient bash we point to installing with homebrew in the failure message that detects this (and we should also be detecting that the command is missing).\nYou can build it from source and install it as a docker plugin by following the instructions at I doubt it'll be part of any of the FOSS distro repos soon (although I'm happy to be proven wrong on that).\nwe don't have the yard stick that something we use should be part of FOSS distro repos. Just that it needs to be easily installable.\nSeems like you can also use the release artifacts from github as well, you don't need their ppa and you won't need to build it either.\nHere is an image I created to solve not having buildx in the official image\n/assign\nhas been updated to use the official binary buildx image\nAL2 also does not come with buildx built in and I landed in with the same issue."}
{"_id":"q-en-kubernetes-bb9d05ab47b1643a3c91e70feca768ba7aea05c3a73104d8523869b02a612c9e","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --pod with nodeSeletor edit pod, delete the nodeSeletor field report error with: ! The errors message should displayed as: Because the nodeSeletor is been deleted with this edit. Kubernetes version (use ):Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig api\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /sig api Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig api-machinery\n/remove-sig api-machinery /sig apps\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:\n/remove-lifecycle stale\n/remove-lifecycle rotten"}
{"_id":"q-en-kubernetes-47e3fff376584abbbc5a14b242e4a53be291e4510620d46d1cf2cd7fab556e9c","text":"/kind failing-test <!-- If this is a flake, please add: /kind flake --/sig node /cc\nThe failure should be caused by\n/assign\n/triage accepted /assign"}
{"_id":"q-en-kubernetes-d927c7177ed39805b33b660a9c2d4d7bf5946fae73d5160a4555c76e83f9522e","text":"I am getting a crash when trying to sort pod with kubectl on Not crashing and at least an error message <details<summaryCrash log</summary</details- Kubernetes version (use ): Cloud provider or hardware configuration: tried on kind latest and openshift 4.7 OS (e.g: ): debian sid Kernel (e.g. ): 5.11.x Install tools: NA Network plugin and version (if this is a network-related bug): NA Others: kubectl compiled from tip branch (as of this bug report) with GO version 1.16.5\n/sig cli\nHey I'm not able to reproduce the crash, although I do get an error on running However, the following works: Could you try this and check if this crashes as well?\nIt definitively crashes it, did you test it in a repo where there is actual pod in there? that error would not show if there is nothing.\nYep Hmm, interesting. Will try to reproduce it again\nSorry, not sure how to remove the label but this is expected. When sorting if the type is not expected it will output errors using over which will logs to the FATAL, ERROR, WARNING, and INFO logs, including a stack trace of all running goroutines, then calls os.Exit(255). Arguments are handled in the manner of fmt.Printf; a newline is appended if missing.\nThose labels are not set on the issue: <detailsIn response to : /remove-kind bug Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nYou need to have multiple pods to reproduce this error. It won't happen with just one pod. For example: The error says the type can't be sorted, so that is the problem: As said, this error is generated intentionally. Here is the code: The problem stems from the fact that  has no way to return an error, so it logs a fatal error which os.Exits. It seems like the problem is that the error message is too verbose. One solution could be to change the code to only print the error message and exit, instead of using which is what is outputting the stack trace. It probably doesn't make sense to dump a stack trace for something that is a user error."}
{"_id":"q-en-kubernetes-862f32665f7ff5a74e5631fecd3a9519d01b2fb3056b9a5567a6d9f9cb201478","text":"<!-- Please only use this template for submitting reports about continuously failing tests or jobs in Kubernetes CI -- (containerd) (crio) [sig-node] ResourceMetricsAPI [NodeFeature:ResourceMetrics] when querying /resource/metrics should report resource usage through the resource metrics api Ever since this commit was merged (containerd) (crio) We need to see what the commit is breaking this particular NodeFeatures test on both crio and containerd.\n/sig node\n/cc\n/cc\n/triage accepted\nah, let's use the RT's issue /close\nClosing this issue. <detailsIn response to : ah, let's use the RT's issue Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-e603fe5ce8ece9dbd9f0832cb091bd8f6895f7a263b51dd8de4e2124e6f25cdb","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --kubelet failed to start when specifying ,with following error: Try to start kubelet with : Kubernetes version (use ): master branch Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig node\nSubmitted PR to fix this issue.\n/triage accepted /priority important-soon"}
{"_id":"q-en-kubernetes-804d53620cf3fadbbe0ac63047ffbcfe1b763effbb880645a77f54d2588b9780","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --I found that for with returns a very small value. The value should be almost the same as the one of . My minimal : Result: While the top three lines () are good, the float values in the bottom three lines () might be broken. Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig api-machinery\n/assign\n/cc /triage accepted"}
{"_id":"q-en-kubernetes-cae18774614f470926558066cfaad9f76aa35af725b3532d8618aeed63a7d6a4","text":"While investigating some slowness in mounting/unmounting operations of PVs (in this case EFS volumes) for pods, we observed errors like these: fortunately it does recover quickly on Kubelet’s next retry: Looking through the Kubelet code I found out that the issue is not related to the umount calls themselves, but rather a race condition inside where we call umount. Because of the way golang’s function (called inside ) is implemented, it for the child process to finish after it is . Since there isn't any locking on the child process across those function calls, it's possible that the child process exits event before the Wait() call starts. In such case the wait call returns with a error (). This error is more likely to happen especially if the gap between the Start() and Wait() calls is more (which could for e,g happen if the goroutine had CPU cycles snatched away from it for some time in between those calls). A similar issue also exists in the . While this is a self-recovering issue (because of Kubelet retries), it does seem to increase the pod startup/cleanup latency upto a 1s. Open to suggestions for a fix. The simplest one occurring to me is to treat the error as a special one on which we don't actually return an error for the mount/unmount calls. /sig storage /sig scalability /assign\nAnother thing is when analyzing a distribution of such errors across different nodes, we observed a skew. The top 10% of the nodes were seeing more than 90% of these errors. And these are nodes that are running hotter (w.r.t the pod churn) compared to others: ! Given this, I'm wondering if those errors are higher on those nodes just because of statistical significance or other causes like more GC pauses happening on those nodes.\n/triage accepted\n^ I've opened this PR to fix the Mount/Unmount functions in the file. In general we might want to make a similar fix to other invocations of exec.Run() and exec.CombinedOutput() within our codebase. But that'll be a bigger change. Happy to open another issue for it.\nSorry for opening the PR against wrong repo, here's the right one -"}
{"_id":"q-en-kubernetes-6addd3c1e7199aac87ba907f15b2227475fea0adf298db12db53b8969dda42ad","text":"This shipped in 1.22 as deprecated in 1.22. As a hacky fix to unbreak tests on 1.23 dev, I bumped the deprecation policy one release: We will need to revert that before 1.23 ships and figure out a better solution. /kind cleanup /milestone v1.23 /sig node\n/triage accepted\nI want to try it. Can I take this issue?\nI have explained how to fix this on PR:\n/reopen reverted in as fix did not work\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/cc\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nHi, Im the bug triage shadow. Since we're in 1.23 milestone code freeze, could we move this issue to milestone 1.24?\nthis is critical for 1.23 and I think it just got forgotten... we need to get this fixed before test freeze\n/priority important-soon"}
{"_id":"q-en-kubernetes-07ff28dc2e90ab46673a9e73d897dd1721db789904ba063b78fff5090ea32daf","text":"When calling GetAllocatableDevices, the API returns all the registered devices, regardless of their health status. Thing is, the - and in general doesn't make much sense to report as allocatable Unhealthy devices, as is also documented in the (see last paragraph) The API should not report unhealthy device. This means the reported device set can potentially change across multiple invocations of the API, as . a kube = 1.21 environment, enable the feature gate any device plugin - including sample device plugin sure 1+ devices are reported unhealthy the podresources GetAllocatableResources API, observe the unhealthy device is reported in the return value. N/A Kubernetes version (use ): 1.21 and beyond Cloud provider or hardware configuration: not relevant OS (e.g: ): any Kernel (e.g. ): not relevant Install tools: not relevant Network plugin and version (if this is a network-related bug): not relevant Others: not relevant\n/sig node\n/triage accepted"}
{"_id":"q-en-kubernetes-986ebc5f1ee3014f101ca7c08d0a452a9f1e89012ba9ad9c3742f7a63611d2f5","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --Following link on sig-network page () is broken Link should be working Kubernetes version (use ):1.21 Cloud provider or hardware configuration:None OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/sig-network\n/sig network\n/triage accepted"}
{"_id":"q-en-kubernetes-667c03a1dc8f1ee7f1158d9381efb0236f75ce957c008d6835a5d4ca1b7b96b9","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --As described in issue , we found kubernetes have been running normally, but the control-plane nodes's metrics value of startTime is similar to the timestamp. Get the correct metrics of node As described in issue anything else we need to provide? As described in issue Kubernetes version (use ): Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\n/remove-sig node /sig instrumentation /assign\nI think the issue is about Summary API having start time for node metric, but this timestamp is incorrect. This is node issue, not instrumentation. /sig node /remove-sig instrumentation /unassign\nWhen discussing adding node start time metric to /metrics/resource endpoint I remember that SIG Node response to this was that there is no good way to define node start time. I don't think we will be able to provide correct value here. As summary API is technically deprecated, it makes no sense to fix it. We should warn users against using it and wait for removal of Summary API.\nOkay, let's wait for the release of metrics-server v0.6.0\nDo you know if this is fixed in metrics server?\n/triage needs-information Can you please add info on name of timeserias and other steps to repro ti understand exactly what is needed here. Thank you!\nin metrics server v0.6.0, we will use /metrics/resource Kubelet endpoint instead summary API\nas issue described , the result is: for example : the startup time of the node test-us-west-1-control-plane-577lp is \"startTime\": \"2021-08-03T19:33:50Z\",, and timestamp reported by metrics is \"2021-08-03T19:34:36Z\", but as described, output from control plane node"}
{"_id":"q-en-kubernetes-e7c38fb3a79882bd75a0e0f99bc9ad04f971dd72570ffcd22429185f154951f1","text":"I'm seeing a panic in kube apiserver (see stack trace below). It appears to be linked to the migration away from Hashicorp's LRU implementation to in The Hashicorp implementation is thread-safe, however the forked copy in utils is not. See the comment here Looking at the utils PR, I see there's actually a wrapper around the thread unsafe LRU cache, and that's being used here: The LRU isn't a lot of code and I think I've convinced myself the locking works fine, so not sure why the element Value is coming back as nil instead of *entry. Stack trace: Not to panic Not sure, but I see it in our CI fairly often Kubernetes version (use ): 1.22Cloud provider or hardware configuration: OS (e.g: ): Kernel (e.g. ): Install tools: Network plugin and version (if this is a network-related bug): Others:\n/cc\n/sig api-machinery\nif Add(key, nil) was called, that could set the value to nil in an existing object... does that happen?\nThe nil value we're hitting is on the linked list element, which is managed by the LRU code, not the user-facing value given in -- that gets saved The panic we hit is , where it tries to re-cast the LL element value back to the lru entry type. I don't see how that can happen.\nThe actual cache is perfectly happy with nil values:\nFrom is mutating since it moves an element to the front when it's accessed. Thanks !\nwould you consider sending a PR to fix it? /assign /triage accepted\nis the PR to fix"}
{"_id":"q-en-kubernetes-777ad7d72db143a151a24bf09007bd12775400d0d01897525cdcd0ecea974fac","text":"SSH'ed into a GKE node running on GCP and was welcomed by following MOTD message: a) Noticed download link is not working, I suspect this storage bucket is no longer used (?) b) Link to get Kuberntes source version does not work: This should be following (without ) Working links :) This is defined in the \"configure-\" file bellow: As this seems GCP specific (IIUC), I think filtering \"-gke.501\" will fix part a) For b) I am not sure what the correct download link should be Kubernetes version: Kubernetes v1.18.20-gke.501 Cloud provider: GCP OS: COS\nPull request to remove \"-gke.\" from link to Github to make it valid. Regarding the wrong download link to GKE source code release, I noticed that: When running on GKE the link has (GCS storage bucket name) while in OOS Kubernetes it is . (So GKE is patched here) I think is depreciated and replaced by but in either case link don' work. It's unclear how one can download the source code.\n/sig cloud-provider release for triagingnot sure if the link is maintained by release team /area provider/gcp\nWas able to confirm the right download location ( GCS bucket) to download GKE source release. Submitted separate PR to fix the URL.\n/assign /triage accepted\nThis is an related to the GKE product. I've filed this to the public :\n/close\nClosing this issue. <detailsIn response to : /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-0c07e91eb7d66ee92c3b6d05f523cece58039dffcf1123c44ba752c30922deab","text":"N/A [sig-node] NodeProblemDetector: N/A https://prow- Not cgroupv2 compatible when access cpu and memory data\n/sig node\nHi, can i work on this ?\nSure. Thanks\n/assign\n/triage accepted\n/priority important-soon\n/cc\nare you still working on this? Were you able to setup environment to repro the test failure? Let us know if you need any help. Thanks!\nSorry for the delay, I followed the wrong guide and ended up getting stuck. But now I have managed to set it up. Please give me few more days to work on it.\nThanks Feel free to reach out here or on slack if you need any help, thanks for working on this!\n/retitle Make nodeproblemdetector test cgroupv2 compatible\nFailing job:\nTestgrid:"}
{"_id":"q-en-kubernetes-a25bba4518f771786855331651a60fed00feaa6f22555c429465b3e593778196","text":"<!-- Please only use this template for submitting reports about flaky tests or jobs (pass or fail with no underlying change in code) in Kubernetes CI --ci-cos-cgroupv2-containerd-node-e2e <!-- You can identify the SIG from the \"prowjobconfigurl\" on the testgrid dashboard for a test --<!-- Please see the deflaking doc () for more guidance! --/sig node\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : #### Which jobs are flaking: links to appreciated links to specific failures in spyglass appreciated </details\nOn cgroupv2, it looks like like the number of for cgroup is sometimes higher than is currently expected in the test. We currently expect this however, in a few runs of the test we've seen We may need to adjust this value similar as was done in for\n/cc\nyes, good idea. I think we need to adjust the PageFaults test as well. I've opened a PR:\n/assign /triage accepted /area test /priority important-soon"}
{"_id":"q-en-kubernetes-4b0aaebc10595b2310d879ccfd264c1dbafd916f75ddeabf532337c80b30d324","text":"pull-kubernetes-unit TestUpdatePodWithTerminatedPod 10/8 (when it was introduced in ) No response No response Running with shows immediate failures: /sig node\n/cc\n/assign I will try to fix this. At present, I found that adding before the status check will definitely trigger the error\nLooks like it stopped flaking? Was this fixed? /triage accepted\nIf the test runs fast enough, it won't failed but it's not guaranteed. fixed in\nIt's still flaking\n/assign\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nthis started flaking more on 3/21 - adding to the milestone for investigation to make sure something didn't regress here\ncc\n/unassign /assign"}
{"_id":"q-en-kubernetes-b4bf9b19d24a261d47cc6eedfe299569e554e3becd9dcdf9c665d53af5c9e2eb","text":"As per the PRs kubernetes/kubernetes kubernetes/kubernetes, the defaulting behaviour of the pod exec parameters had been changed however in the code there is a comment where it is mentioned it defaults to True The comment about the default value must be removed and the openapi spec must be fixed Present in the file as comment and also in the OpenApiSpec swagger If this issue seems appropriate can I take it up please <details</details<details</details<details</details<details</details<details</details<details</details\n/sig docs\n/sig node\n/assign\n/assign\n/help /triage accepted\nThis request has been marked as needing help from a contributor. Please ensure that the issue body includes answers to the following questions: Why are we solving this issue? To address this issue, are there any code changes? If there are code changes, what needs to be done in the code and what places can the assignee treat as reference points? Does this issue have zero to low barrier of entry? How can the assignee reach out to you for help? For more details on the requirements of such an issue, please see and ensure that they are met. If this request no longer meets these requirements, the label can be removed by commenting with the command. <detailsIn response to : /help /triage accepted </details\n/kind documentation"}
{"_id":"q-en-kubernetes-768e64f47ada6db7cc30fd697f7a40f044b4c151072bc930e0f353b7e0a147a6","text":"runs in a container. Talos sets OOM score to -450 for the kubelet container and passes in the kubelet config. End result is that OOM score adj is -999 though. should have configured OOM score . Source code analysis. First, OOM score is set correctly from the config (flags): But then, in container manager, OOM score is reset to -999 via the following: This function should skip action if the PID is not running on the host: But the function looks to be confusing: When kubelet runs in a container with PID namespace shared with the host, kubelet's PID namespace will be equal to PID 1 namespace. If kubelet runs in a container with unshared PID namespace, kubelet's PID is 1 and its PID namespace is identical to PID 1 namespace. So how this function can return ever? Moreover, function is always setting OOM score to a hardcoded constant -999 ignoring configuation (). No response <details, reproduced with 1.22. </details<details</details<details</details<details</details<details</details<details</details\n/sig node\n/triage accepted\n/assign\nI don't think this method needs to change... I have make a draft PR for the default value to use the flag if present, but I would like more time to add some testing. Maybe you can have a look: ?\nThanks, this PR looks good to me, as it's going to set the OOM score to the configured value. And in our case kubelet runs in a container (not on the host, with proper OOM score via OCI spec), so this code should be even skipped, not sure how this can be addressed.\nneither do I...\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale"}
{"_id":"q-en-kubernetes-921d22c2f559c36be3e8ae85e52ba24120da7ee8caf115af1ba9cbca6aaf812b","text":"This is an umbrella issue to track changes in the . It is required to update the guide based on the suggestion we've received so far after officially migrating , , and (ongoing migration). The guide can be improved by adding more relevant details about structured logging migration, examples, common mistakes, and easy navigation in the guide using appropriate referencing. We've observed common mistakes made by contributors while doing the migration. To avoid those \"common mistakes\", more examples are needed in the structured logging migration guide: approach could be manually looking at various review comments that we've received so far while performing migration of , , and and figure out common mistakes done by contributors. Add examples to avoid those mistakes, and enhance the guidelines present in addition to this, people involved in the previous migration process, either as approver/reviewer or as contributors, can suggest changes in this umbrella issue or open a pull request towards the necessary changes referencing this issue. As we advance, we need to make the migration guide suitable for both \"contributors\" and \"users\" of Kubernetes. Since slowly we are progressing and covering more components of Kubernetes, it is required to improve the guide so that both contributors and Kubernetes users can effectively use it.\nSuggestions in form of review comments from previous migrations are here:\n/sig instrumentation /wg structured-logging /area logging /priority important-longterm /kind documentation /cc\nDo you intend to update the documentation yourself, or are you looking for others to help here?\nYes, 1) I'm looking for contributors and reviewers to suggest more changes (I may not be able to cover all the required changes) and I'll add those changes in the table above. 2) Contributors can submit PRs for these changes in the migration guide, I can definitely submit PRs if we see that the issue is not progressing enough\nThanks. I'm going to assign this to you for now then. /assign shivanshu1333\n/assign\n/triage accepted\n/sig docs\n/assign\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale"}
{"_id":"q-en-kubernetes-35499a3ca46347fc92f5eaa71524164ce200edabc0d3f386dc26b07d9278777e","text":"I was trying a scenario where liveness/readiness probe command run inside the container for long time('sleep 10 && exit 10') with shorter timeoutSecod value say 2 Sec. As expected container restarted because of timeout but when i check the events in pod describe it stated only unhealthy and in events actual reason for probe failure is not printed \"liveness probe failed:<'no reason printed'\". Without clear logs or pod events description it is very difficult to debug and understand what is the failure. Below is the example manifest file to reproduce the issue. apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-exec spec: containers: name: liveness image: args: /bin/sh -c touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; sleep 600 readinessProbe: exec: command: [/bin/sh, -c , 'sleep 10 && exit10'] initialDelaySeconds: 5 periodSeconds: 5 timeoutSeconds: 2 livenessProbe: exec: command: [/bin/sh, -c ,'sleep 5 && exit10'] initialDelaySeconds: 5 periodSeconds: 5 timeoutSeconds: 2 It would be better if actual reason or something for probe failure printed in pod describe events incase of liveness/readiness probe gets timeout. Example like below in pod describe events: Warning Unhealthy 3s (x2 over 8s) kubelet Liveness probe failed: \"probe failed due to timeout \" Warning Unhealthy 3s (x2 over 8s) kubelet Readiness probe failed: \"probe failed due to timeout \" ! It would be better if actual reason or something for probe failure printed in pod describe events incase of liveness/readiness probe gets timeout. Example like below in pod describe events: Warning Unhealthy 3s (x2 over 8s) kubelet Liveness probe failed: \"probe failed due to timeout \" Warning Unhealthy 3s (x2 over 8s) kubelet Readiness probe failed: \"probe failed due to timeout \" Below is the example manifest file to reproduce the issue. apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-exec spec: containers: name: liveness image: args: /bin/sh -c touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; sleep 600 readinessProbe: exec: command: [/bin/sh, -c , 'sleep 10 && exit10'] initialDelaySeconds: 5 periodSeconds: 5 timeoutSeconds: 2 livenessProbe: exec: command: [/bin/sh, -c ,'sleep 5 && exit10'] initialDelaySeconds: 5 periodSeconds: 5 timeoutSeconds: 2 Create a pod using above manifest file and do pod describe (kubectl describe pod <pod-name ) and check for the events. ! No response kubectl version Client Version: version.Info{Major:\"1\", Minor:\"21\", GitVersion:\"v1.21.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2021-08-31T06:59:53Z\", GoVersion:\"go1.16.4\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"21\", GitVersion:\"v1.21.1\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2021-08-31T06:52:44Z\", GoVersion:\"go1.16.4\", Compiler:\"gc\", Platform:\"linux/amd64\"} openstack <details</details<details</detailscontainerd No response\nI tried to fix one similar issue in\n/sig node\n/triage accepted\ncan I work on this issue? Thanks very much. /assign\nWe've experienced this issue with clusters on v1.20. The for that version mention the following change: Since v1.20 is seemingly when exec probe timeouts started being enforced, would it be possible to have this fix for this issue backported to all minor versions since then? Thanks!\nbump on that question\nWhat do you think of? Thanks very much."}
{"_id":"q-en-kubernetes-d3c8c14a47bead0a2aabecf2aacf5b825b61c49fe97ab53351d341210fbbc8d4","text":"kubeadm-kinder-latest ci-kubernetes-e2e-kubeadm-kinder-latest.Overall E2EKubeadm suite.[sig-cluster-lifecycle] [area-kubeadm] proxy addon kube-proxy ServiceAccount should be bound to the system:node-proxier cluster role 2021-11-09 23:32 CET () E2EKubeadm suite: [sig-cluster-lifecycle] [area-kubeadm] proxy addon kube-proxy ServiceAccount should be bound to the system:node-proxier cluster role This issue seems to be linked to FYI: /cc /sig cluster-lifecycle /area kubeadm\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nfix PR is here, needs LGTM:\nAlso failing on Thanks !"}
{"_id":"q-en-kubernetes-84ac6b29d3552ee4565d85bb68dd153f357ab1763ab9daf494e08879761ee84b","text":"ci-kubernetes-e2e-aks-engine-azure-master-windows-containerd-serial-slow Kubernetes e2e suite: [sig-windows] [Feature:Windows] Memory Limits [Serial] [Slow] Allocatable node memory should be equal to a calculated allocatable memory value expandmore this month No response_ /sig windows\n/assign\n/triage accepted /priority important-soon"}
{"_id":"q-en-kubernetes-70cc8a837504b60a5c6c5f5140cd4a6de426afd5bd1cc3a810b5342e232bd03d","text":"integration-master ci-kubernetes-integration-master.OverallChanges 11/19/2021 23:54 CET see this as example to get information about the failure No response No response_\ngreat catch /sig api-machinery /priority critical-urgent it seems all tests fail with how can this fail consistently only in periodics and not in presubmits? is an external dependency?\nI would say this is the origin Server Side Strict Field Validation \npresubmits only run short tests, periodics run all long ones... looking if there's a long test that needs fixup\nfix in"}
{"_id":"q-en-kubernetes-f6b8358021d22b6a22cf92b93e9370729451ca3669c4f6afe9d84808b437dd55","text":"ci-kubernetes-node-kubelet-serial-containerd: ci-kubernetes-node-kubelet-serial: (This is just an example, most of the tests appear to be failing) /sig node /kind failing-test /milestone v1.23 since test is failing in a release-blocking job /priority critical-urgent /assign /cc I believe this stems from\n/triage accepted These failures look weird - I don't understand why dockershim is sometimes failing after restarting the kubelet. I guess we should double check that timings match the old GPU test (I think I might've tightened them a bit, bc they were passing when I ran the device suite on the same nodes). What's weirder is that they passed on my PR and all of a sudden broke after merging\nDiscussed at today - the easiest solution now that 1.23 has been branched is to punt these tests out of the serial lane so we can get it green again for 1.23, since any changes at this point must be backported and we would prefer that this is not release-blocking.\n/milestone clear should address this; we won't get a fresh serial run for another ~6 hours so let's confirm next business day that this is temporarily fixed.\n/remove-priority critical-urgent /priority important-longterm\ncan we keep it in the milestone until we get a verified green run?\nCI is green today after skipping DevicePluginProbe tests .\n/assign Maybe related to\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:\nLooks like serial has been stable for a while, closing: /close\nClosing this issue. <detailsIn response to : Looks like serial has been stable for a while, closing: Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/reopen it is referred to here: Tests are skipped today\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\ndid you have a chance to look at this? I remember you were close to understanding on what is going wrong here\noh welp, missed that\nI was (past tense) but now I need to go back and look at my notes. If someone else wants to take over, be my guest, I've been quite busy with other tasks lately. Otherwise I'll get back to this!\nSo! I'm able to resume working in this area. So I'm testing last k/k master branch tip and running with : and the tests seems to be running and passing quite reliably! (5 times in a row seems like a promising start :) ) To move forward I suggest to: a explicit s in . This will make more evident we have a pending issue in our camp, that's the most egregious exception we have and it's confusing the skip in the config: - this will enable us to do focused work in this area a PR to remove the skip and iterate over it running CI until it's good enough I can take care of all the above. If CI is still failing, we need to explore possible missing/wrong configuration, but then other e2enode tests should be failing, not just this one.\nI agree, this plan sounds good. Thank you!\nxref:\nstarted to fail also locally - we have baremetal lanes in RH - in the project CI they skip the relevant test. Investigating a possible common root cause\nafter a very long troubleshooting and quite few attempts, is ready for review"}
{"_id":"q-en-kubernetes-bc1f56f76d806f780c64417014c4b072addf1f4df6d254dd55319774f349a24c","text":"Flaked a couple of times on serial lane with the same error. Test is relatively new. This run only has the single test flake: /sig node /kind flake /cc\n/assign\nNew recent failures: /triage accepted"}
{"_id":"q-en-kubernetes-828b2f2bc9a1c73b1b130d20e1091ab566168df681bda6f542cc66f3f230828b","text":"/kind failing-test <!-- If this is a flake, please add: /kind flake --<!-- Please assign a SIG using: /sig SIG-NAME --\n/sig cli\nThere may have two issues to fix: doesn't support run with assertion must check whether schema is nil first\n/assign\n/triage accepted"}
{"_id":"q-en-kubernetes-b00d4aab62f403b854c1a0aa3c8ac8b12f2203f1df9b02fc2cc1a8c8fa361c58","text":"gce-master-scale-correctness Kubernetes e2e suite.[sig-network] Networking Granular Checks: Services should update nodePort: udp [Slow] ci-kubernetes-e2e-gce-scale-correctness.Overall kubetest.Test Kubernetes e2e suite.[sig-storage] CSI Volumes [Driver: csi-hostpath] [Testpattern: Dynamic PV (filesystem volmode)] volumeMode should not mount / map unused volumes in a pod [LinuxOnly] Kubernetes e2e suite.[sig-storage] CSI Volumes [Driver: csi-hostpath] [Testpattern: Generic Ephemeral-volume (default fs) (late-binding)] ephemeral should support two pods which have the same volume definition Kubernetes e2e suite.[sig-network] Services should be able to update service type to NodePort listening on same port number but different protocols 12/1/2021 17:36:39 ET No response No response /sig storage /sig network\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/milestone v1.23 /priority important-soon\n/assign aojea\n/assign bowei\nheh, what are the chances, two different test services took the same port The test was expecting to not receive any answer but the second service replies and makes the test fail link"}
{"_id":"q-en-kubernetes-4a058229eb86de6555a991f0b180b07c42b35b29d04a1ae7b1d2febd668ac9b4","text":"During a Kubernetes deployment the connectivity to web services from Windows nodes is flapping. There are intermittent traffic delays and some errors observed on web services during the deployment. Post deployment the flapping is no longer observed. No flapping in connectivity during the deployment. Create a deployment with multiple services and monitor connectivity. No response <details20 </details<detailsAKS </details<details</details<details</details<details</details<details</details\n/sig windows\n/triage accepted /priority important-soon"}
{"_id":"q-en-kubernetes-3e7e95e6f142ca39e6e3ca0b78c3932bd6cbdfdc898eed9d3ffda27d28f904c7","text":"We should make the change in PR API reference links to should use HTTPS, not HTTP Visit h and look at the IANA hyperlinks Once this fix is made, a PR to is needed to update the docs (by autogenerating them). This typically would be deferred until the next Kubernetes release. v1.23 not applicable not applicable not applicable not applicable not applicable\n/sig docs\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI opened this issue so that someone else can open a new PR that is equivalent to .\n/retitle Update links to IANA for Service Name and Transport Protocol Port Number Registry\nI would like to work on this issue. May I assign it to myself?\nAnyone - including you - is welcome to work on this issue! So long as you've you can open a pull request to propose a fix.\nThanks\n/assign\nI have made the PR. I hope this works. If not then will be happy to do the necessary changes.\nHad to close the previous PR because of issues, here is the new one I made."}
{"_id":"q-en-kubernetes-2214f549829b0bccc5ce423eaca68e7f13803c72fd2591e1ab24d03ad7fdf316","text":"Condition PIDPressure is not set to true when the system thread number exceeds even it's reaching the limit. Condition PIDPressure is set to true correctly See cause below The bug is due to , which truncates any process number greater than because of type conversion. See kernel code . Reading it from should be working. v1.23.1 <detailsNA </details<details</details<details</details<details</details<details</details\n/sig node\n/triage accepted /priority important-longterm"}
{"_id":"q-en-kubernetes-9f83a27f9592105435bbac386998288b990d6f5d165b5fe2e78e2c4c68ba6eee","text":"There is no receive or close operation on channel of struct () . Thus, a goroutine leak happens when is executed (), such as when running . no goroutine leak run TestRunPositiveRegister() No response <details97d40890d00acf721ecabb8c9a6fec3b3234b74b and following versions </details<details</details<details</details<details</details<details</details<details</details\n/area kubelet /sig node /assign I'd try to fix it\n/sig node\nThe author has submitted a PR for this issue:\n/triage accepted /assign /priority important-longterm\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:\nThe issue has been marked as an important bug and triaged. Such issues are automatically marked as frozen when hitting the rotten state to avoid missing important bugs. Please send feedback to sig-contributor-experience at . /lifecycle frozen"}
{"_id":"q-en-kubernetes-c3c001164d06967bffc88ae4a064b83e4730508e6212066ef72f1419123a3c10","text":"I deployed a pod with and on a namesapace with a set with and elements and I get the error: The pod to be deployed Create a new namespace Create (using a manifest file named resource-quota-) and verify a named compute-resources Content of the resource-quota- file Create (using a manifest file named ) a pod named high-priority with the and for both and . Content of the file The pod is deployed correctly if I delete the <details</details<detailsaws </details<details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig Scheduling\n/wg API Expression\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /wg API Expression Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/wg <API Expression\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /wg <API ExpressionInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/wg \"API Expression\"\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /wg \"API Expression\" Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/wg APIExpression\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /wg APIExpression Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/wg wg-api-expression\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /wg wg-api-expression Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nhey, v1.18.20 is out of support now, you should try with a newer version. Besides, I tried with an old version v1.18.19, but worked as well.\nAre u using some mutating webhook which inject some sidecar into the pod without setting resources fields?\nIt's probably that, but IDK which WebHook is doing that. The feature you suggested would really help.\nhi the fix pr has been merged, maybe u could use it to locate which container is causing the problem.\nThanks"}
{"_id":"q-en-kubernetes-a8d1b8b007f3b41febd8e28d8c3f6d762909a93ce545d0ad364f8e401a78d99f","text":"In the implement of GCE mode WaitForAttach: when the runtime.GOOS = \"windows\" and it meets err , it should instead of . the code should be like this : None The modification is in order to match the code cosistant pattern. Any GCE Any Any Any Any\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig storage\n/sig cloud-provider\n/assign\n/assign"}
{"_id":"q-en-kubernetes-4fde01bf9e733eb02396ee79e06a0d3d7fd635df01f60f51582a52b831148901","text":"/kind failing-test <!-- If this is a flake, please add: /kind flake --/kind flake <!-- Please assign a SIG using: /sig SIG-NAME --/sig api-machinery\n/assign I want to take a try\n2 but actual 0 till timeout resource services, expected 2, actual 0 error waiting or ResourceQuota status: timed out waiting for the condition time later, actual state became 2 Seems polls a ResourceQuota status with timeout 10s may be not enough, we should enlarge it into 300s, that is poll for 300s timeout instead of 10s.\n/triage accepted"}
{"_id":"q-en-kubernetes-638e1ca866d667abba257ab7bbd4b65d0d17bd2e9e09c09309c35a895edf6db9","text":"There are pre-defined tests patterns that specify \"CSI Ephemeral\" in combination with a non-default fstype, for example: This code then should set but currently doesn't: Not sure whether these patterns are used in Kubernetes. I ran into this with PMEM-CSI, which uses them. The effect is that all test cases using those patterns use the default fstype instead of ext4 or XFS. /sig storage Kubernetes 1.24 and several older versions. n/a No response No response No response No response_\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/assign /remove-lifecycle stale"}
{"_id":"q-en-kubernetes-9fbcd5201ae07f3b7a89bd2301992ba7ab5d6d8edc83db6ef44c2beaac86b1f5","text":"Integration test timed out due to the following error The error appeared 748 times in the log. Looks like lots of churning happens during the apf configuration bootstrapping phase. Is there a race condition between when the type is being registered and the API calls? Stack trace: Integration job link: PR: I searched for this error in CI and this PR, looks like we only see this error in this PR so far. So I am hoping it has not introduced any flake. (maybe we need to let more time to pass to start seeing flakes) We should not see this error I saw it only in the PR mentioned.\n/sig api-machinery cc\nI see\nThe fact that it happens to a non-mandatory object implies that the object was received from the server, so the type must be in the server. The APF config-consuming controller does this:\nLook for the output from this line:\nAnother PR with this failure:\nthe panic means the request scope FieldManager is nil that is initialized here: OpenAPIModels is set here: called from here: and openAPIModels is constructed here: it looks like it is possible for openapiconfig to be unset, which causes the openapi models to be nil, which causes apply to panic. I don't know if it is only the integration tests that are subject to that configuration. cc /priority important-soon (since this is now being exercised in bring-up of integration test servers)\nif APF is creating/bootstrapping the objects without using apply, and then updating status using apply, that would explain how the object exists but apply fails\nI suspect that APF is not successfully updating status at all in the integration tests with this server configuration\ndo you want me to revert , while we work on a solution?\nI don't know how bad the test impact is. Work with / to determine if we should revert or fix forward\n: I do not follow why you talked about server-side stuff then suggested the problem is in the client (a controller in the kube-apiserver) not supplying a fieldManager. The client does supply the fieldManager, for both creates and updates. See the following lines:\nDifferent field manager. is set server-side in the endpoint handler based on the server configuration.\nSo if the problem is server-side, why did you suggest the problem is in the client?\nRegarding the client, I looked in and found this line: actually several of the same form, all with the same fieldManager.\nI don't think I did suggest that :) all of the setup and calls in are server-side\n: please help me understand . That is about the client side. Are you saying that a client that creates an object using and later updates the object using --- with a non-empty fieldManager in both --- will run into this server-side scenario? If so, is that because this client behavior is wrong or because there is a bug server side?\nThat was in response to , trying to figure out how the object was created but could not be updated. Yes, when the server is not configured to enable openapi and therefore does not have apply enabled. Because the server is misconfigured (which, from what I can tell, is limited to some of the server instances started for integration tests)\nLooking at and it is not obvious to me whether or how openapi could be disabled. Would that be by turning off the feature flag? FYI, the inclusion of \"V3\" in the flag makes me unsure what the server does about other versions. : I think you are saying that it is possible to disable openapi and that this in turn disables server-side apply. That seems kind of surprising to me. Does this mean that it is simply incorrect to run a client that does server-side apply with a server that has openapi disabled? Is this intended? Is it documented anywhere?\nIt's not intended as an externalized configuration option. integration tests have the ability to fiddle with many of the config struct options in the server that a kube-apiserver binary invocation could not. They intentionally use this to do things like inject specialized authorizers / authenticators for testing purposes, but also can unintentionally result in a test server that doesn't match a real API server invocation. I think this is an integration test server configuration problem, which should be fixed.\nis correct that we have a misconfiguration in the integration tests and SSA is only enabled on a subset of integration tests. I'd like to understand the test failure conditions a bit better though. The can be consistently reproduced on a non SSA targeted test. (Eg: selected from one of the test failures). However, I'm seeing the panic message while the test is still green and passes locally. it's a bit bizarre what contributes to the flakiness of the test because the FieldManager panic can be reproduced 100% of the time.\nThe panic occurs when APF is updating status (which it just switched to do using SSA). The APF status update failure does not block the API requests the test cares about, so the test passes. I could see extra CPU usage contributing to flakiness.\nThanks for the explanation. Created to fix the integration test configurations.\n/triage accepted\nThe integration test job started timing out dramatically last night (50% of runs) once merged. Since that was the only change made to integration tests recently, it seems related. I think we should revert (and therefore ) until the test impact is understood\nrevert of -\ndid this get fully reverted? we're still seeing panics in integration tests over this:\nbumping priority and milestoning until we know we're not currently triggering panics\nlooks like it got reintroduced in , revert open in\nFor integration tests, maybe we can cheat and cache the OpenAPI data, so that it's not recomputed for each test?\n:-/ maybe... but I really don't want to have non-real init paths in integration if at all possible\nI wish we could go a little further in lazily initializing openapi (especially v2), so let's see what we can do here (after code-freeze probably)\nif SSA is actively using the openapi, it being lazy won't help, right?\ndoes the revert mean that we can remove milestone v1.24 from this issue?\nYes, cleared\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:\n/remove-lifecycle rotten\nTaking a fresh look at this PR and wondering if it has been mostly fixed already? From comments my understanding is that this panic occurred due to two conditions: running as part of the integration test suite is asked to service an SSA request while are nil. The apiservice has logic to panic in this case. From the code path commented is nil when the server is created with is nil. APF worker (running alongside an integration test) issues an SSA request to the test apiserver. Since the apiserver is misconfigured it panics. This panic could happen sporadically because sometimes the tests finish before the APF worker is able to issue the request. As a brief check to see if this might be fixed, I ran the test mentioned was failing from SSA. I logged out the 's created and found that they all had managed fields. I saw this as an indication SSA was now functioning properly. Indeed, to start the test server the function calls . This function has a code path which unconditionally leads to the apiserver's being set. calls before running the returned apiserver: calls before using the returned config to create the returned apiserver: uses which sets : Thus, any test using should no longer be affected by this panic. The question then becomes: do all integration tests start their servers in this way? It's hard to say. I found a PR (, ) which refactors many occurrences of the old stanza: Into something using the new method: ` But it appears is still being used in a few integration tests: Should these occurrences be refactored to use too?\nThere was a ton of changes I did around that. But the last PR (already removing the old is already out for review): - FYI Thinking about it now, I guess I agree that the problem might have been 10 different way kube-apiserver was set up through different tests. I just don't know why it was causing flakes, not failing consistently...\nThere were definitely test setups that were not enabling openapi in the recent past. Two possibilities for flakes: Enabling openapi in all integration test setups was expensive and caused spurious timeouts The excessive numbers of background panics handled and logged in integration tests when the API server tried to handle SSA requests bogged down the CI machine (I think something like 40,000 panics with full stack traces each were getting logged)\ni don't think that's the case. With the recent changes that I was doing as part of I think OpenAPI is already enabled for all tests having kube-apiserver (almost all) and that seems to work. But you remainded me that back then we also had an issue that we were running many more tests simulanuously, which could have been contributing to this problem significantly. The second reason is definitely possible contributor too. I guess my concern was that if some requests were panicing due to lack of OpenAPI, this means that these aren't really needed. But that actually might make sense - the fact that we weren't able to upgrade status of PriorityLevels/FlowSchemas shouldn't prevent non P&F-tests from succeeding, and that worked in P&F tests...\nI think the panics were happening in async controllers, and no functionality blocked by the controllers intersected with the surface being tested, it just (maybe) caused log/perf issues\nI believe has resolved the panic issue, I was able to verify with which changes the priority & fairness controller to use server side apply. Before merged, I analyzed the integration log: Integration test log file from an unrelated pr does not contain any trace of the panic as expected: On the other hand, the log file from integration job for (before merged) had plenty of these panics even though the integration job turned green: I wanted to check which tests produced these panics, so I did a breakdown by package Then I narrowed it down to the individual test(s): Before merged, the above tests used , and either is or is which results in to be nil. But a quick code search revealed that many more integration tests used , so why were the other tests not affected by the panic? I inspected the log of a test that was not affected by the panic, and grepped for , and found the following: In the above case, the apf controller had not advanced far enough, the controller was active for about , may be these tests finish quickly and signals the server to shutdown before the apf controller can send an APPLY command. On the other hand, log from a test that was affected by the panic shows that the apf controller was active for more than , and has lot more activities: I am assuming that integration tests have at least log level 4. Also, the size of the log file differed significantly, without panics around , and with panics over and changed all these tests to use and now integration log for shows no panic.\nGreat! (This also proves why eliminating gazilion different things of initializing stuff in favor of one makes sense). - FYI ^^ - let's merge the PR then - I just one question there, so let's wait for Jordan to respond it.\nCan we just force the openapi/fieldmanager to be set at start-time now that SSA is GA, it should basically always be present, rather than fail when SSA is unavailable (that's probably wrong by the way). If that triggers any test flakiness, then we would solve these. We can wait for to be solved since it will almost certainly have a huge impact on this.\nHas this been done Jeff?"}
{"_id":"q-en-kubernetes-9fac73cbeefff3e50660c328aaeeb2b47c423f7885694acc8690c878b3ad99f5","text":"Unit test in package flakes there is a race condition between the go routine that executes the request handler and the timeout go routine that waits for the handler to finish /sig api-machinery\n/assign\nI have a PR in the works\nshould fix it, I can easily reproduce it with the PR, the flake is not reproducible any more cc\n/triage accepted"}
{"_id":"q-en-kubernetes-56e7478c97a10d92fc7484acd2b672ce77d361e72c8e8784f90182dd99033ab6","text":"In </span /kind failing-test <!-- If this is a flake, please add: /kind flake --/sig node I believe the issue here is that the test is trying to get 5 restarts, but if we don't start the loop until after a few failures have occurred we start getting caught up in exponential backoff. The 6th backoff iteration is a 2m30s wait (i believe), so we exceed the default timeout of 5m for the test. We may need to change the test so that we are watching for pod status updates and catch the first one. HOWEVER - there's also some head of line blocking in pod status manager updates that I suspect might result in updates getting delayed, leading to timeout. Either way, the test has too little margin with exponential failure backoff to be reliable, and needs to be improved.\n/triage accepted /assign"}
{"_id":"q-en-kubernetes-f05405644445a0e792013d4eae28a3cb46c227199ee42d853fdd090273fe2800","text":"Since scheduler Policy API was total removed, we don't support any more. Remove unsupported configuration.\n/sig scheduling /assign\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig scheduling /cc\n/sig scheduling (trying to tell the bots to remove )\nThe label(s) cannot be applied. These labels are supported: <detailsIn response to : /remove-needs sig /remove-label needs-sig Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nIt's true scheduler doesn't quite use this field anymore, but I'm wondering if some other project like ClusterAutoscaler or simulator still consumes this via .\nThere's no place we can config this field now by configure file(before we can apply the change by policy api, but now it's removed). So it seems like unless we support this feature with CC, or this field is useless. To other projects, it's the same. When they are using legacy policy api, they can configure this field, but once they synced with the latest kubernetes, with the policy api removed, they can not configure this anymore. In this situation, we should add support in CC. I don't quite understand the relationship between k/k with k/autoscaling or others, I think they sync with k/k periodic.\nAs I mentioned above, it's still accessible via a framework runtime programmatically, and thus may be still useful. For example, upon a scheduling failure, a ClusterAutoscaler's implementation may choose to dryrun through all filter plugins to ensure a new provisioned node can fit the pod, instead of resolving the failure while failing at another latter filter plugin.\nOh, I haven't thought of configuration via programmatically, that saved my concern. But a little strange.\nThat's exactly what I meant. Given it's a public option in framework, and there is no alternative if we disabled it, it'd be good to check with potential consumers. For now I can think of CA ( and k-sigs/scheduler-simulator ( If they can confirm it's not needed in their usecases, I'm fine with removing .\nTo k/autoscaler, at least these codes are still there together with the legacy policy api. , looking forward to the response.\nis not used in sigs/kube-scheduler-simulator. ‍\nthis isn't used by cluster autoscaler, we can remove it."}
{"_id":"q-en-kubernetes-4ee63e795a95b2ddd1db27a0ea604546840d3c5aaaaad1cbf8e9f94cee6783df","text":"In the K8s Conformance meeting of 6 April the testing of Node endpoints and their eligibility for conformance testing was discussed. This issue is created to get more community input on the topic. Should node endpoints be tested? Can we accualy test these endpoints with the current test framework /sig testing /sig architecture /area conformance\n/triage accepted /cc\nI have lost track of this issue where it have already been discussed. Let's continue to discuss further here. Apologies for the confusion /cc\nClayton We have historically focused conformance on the perceived end user experience deploying applications to a conformant Kubernetes cluster as being consistently successful. From that perspective, while these are actions that are indeed key to conformant behavior on the majority of distributions, the first two are not invoked by end users, only by kubelets and potentially administrators in rare cases. I would expect to be able to create a node object, delete it, mutate its status, and delete a portion of them on a conformant Kubernetes cluster as an admin - but I would not expect a similar end user to do so. If we can make a case that a conformant Kubernetes distribution would reasonably not allow any of these actions via the API, then we could make a case to omit them from conformance. I can see a small path to adding these to conformance under a \"they may be necessary as administrative action on the vast majority of kubernetes clusters\", but I can't necessarily bump that to 100% of Kube clusters, especially if we put the application deployer as the center of our story (an application deployer does not need to delete node collections in order to succeed at running an application on conformant kubernetes)\nAddressed the question in SIG Node\nCreated to make the remaining node endpoints ineligible. readCoreV1NodeStatus replaceCoreV1NodeStatus deleteCoreV1CollectionNode"}
{"_id":"q-en-kubernetes-de92c939e0a1e60c55a5cf3ecf8c3bdeb2213fc2caf98746cabbe1f6c74c2fbb","text":"ci-kubernetes-e2e-gce-master-new-gci-kubectl-skew kubetest.SkewTest 2022-04-14 15:13:42 +0000 UTC The error message is It's complaining about \"file contents\" and not \"url contents\". And because this is a skew test I am assuming this is due to The fix might be to cherry pick this to 1.23. No response /sig cli\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI've opened the cherry pick at if we decide that's the right route. The other options I can think of are to revert this change on master or to add the old environment variable back in as a file path. /assign /priority critical-urgent\nTracked the root of this to This should start passing once that merges. The cherry pick will also fix. We really need to get skew tests running on PRs."}
{"_id":"q-en-kubernetes-4a798c609678596178892c7255d5cb622f9a5ebde3e41ac4403c4fdb42531beb","text":"Please publish 3.5.4-0 image and update to to benefit from fix for , while still keeping v3.5.3 critical correctness fixes (, ). For recent PRs for update to etcd v3.5.3 see and\n/sig release /sig api-machinery /kind bug\n/cc /triage accepted\nHey, Can I work on this ? I'm a beginner, but all the good first issues are mostly assigned. Can you guide me on doing this ?\nI am working on this. /assign\nHi I would like to work on Make sense to backport in v1.22, v1.23 & v1.24? Thanks!!\nhi, it makes sense to me, but requires approval from dependency maintainers. i can only approve the kubeadm changes.\nwas this backported ?"}
{"_id":"q-en-kubernetes-6bd60201c464597f57742f0cdd678358885c62b89b4b459f1bb929b8b5a182ee","text":"/kind failing-test <!-- If this is a flake, please add: /kind flake --/kind flake the test case was add in /cc smarterclayton <!-- Please assign a SIG using: /sig SIG-NAME --/sig node\n/assign\nbumping priority, seeing this in PR presubmit runs\n/triage accepted"}
{"_id":"q-en-kubernetes-108be3290e39da28e1ba1a62ee2acdaabaf58121c5b483bdc1d93743d8cdb6ec","text":"Follow-up cleanup work for: Most of the information about this issue can be found in , where the work merged with a to change the function signature of to return a func() as well as an error. This issue will help to track that. All TODOs introduced in are resolved. N/A N/A <detailsN/A </details<detailsN/A </details<details</details<details</details<details</details<details</details\n/sig api-machinery\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /priority low Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/cc /triage accepted\n/priority important-longterm\n/assign\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale"}
{"_id":"q-en-kubernetes-11931a81f49292e30fa187df18acf1a79a4af7c51f2db8ca1fc930b4aeb6f468","text":"Un-deperecate the kubelet flag, as this option is typically set per-node similar to (which is not deprecated). Or more generally, consider not deprecating fields available in config that make sense to set per-node. I noticed this while trying to ensure KIND moves towards not setting any deprecated flags, this one is tricky because it's not a fixed value. For a lot of kubernetes users, kubeadm is used for bootstrapping, and kubeadm supports supplying additional kubelet flags when configuring each node, but the kubelet config is cluster-wide, created during setting up the first node and uploaded to a configmap then used for all nodes. In the future kubeadm may have the ability to do per-node kubelet config, but that effort will probably be involved and isn't staffed at the moment. technically you could run kubeadm in phases and then modify the config, but this is more complex to implement. It's possible this won't be necessary if kubeadm patches can be extended to kubelet which may be a smaller feature change for kubeadm, but personally I think being able to configure simple per-node strings with a flag override is desirable for implementing clusters, and we've had the flag for many years with no trouble ... The actual deprecation seems to be due to a blanket deprecation on things possible to set in config, and not targeted at this flag in particular. See discussion in\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\ntechnically KIND doesn't strictly need this flag, FWIW, but we were asked by users to set it to make detection easier and have values more similar to \"real\" clusters. other projects in Kubernetes are also doing this, e.g. some of of the cluster api-provider implementations:\nI think for 1.25 in kubeadm we can enable patching of kubeletconfuguration. This will allow us to use the current global kubletconfiguration for all nodes, but patches can act as the node instance specific config. Flags would still take precedence over patches and are the simpler UX. But if one day all flags are removed the above solution will still support the instance specific use case. I don't know whether some flags should be de-deprecated, but IIRC the plan to migrate all flags to kubeletconfiguration and not add more flags was still desirable by sig node.\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\nThe Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle rotten [1]:\n/remove-lifecycle rotten\n/assign It seems to be reasonable. I opened to undeprecate it."}
{"_id":"q-en-kubernetes-72d7af02b7cb24adccad9225734c2fb8351d64a440dcf1418ee834dbb0676072","text":"release-master-informing capg-conformance-main-ci-artifacts Kubernetes e2e suite.BeforeSuite 05-25 06:37 IST Failure Error waiting for all pods to be running and ready: 3 / 13 pods in namespace kube-system are NOT in RUNNING and READY state in 10m0s No response /sig network /sig node cc\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/remove sig-network /remove sig-node /unassign This seems a deployment problem ... on calico to be precise first point of contact for initial triage should be the job/sig owners ... this job also seems to be failing for a long time /sig cluster-lifecycle sig-cluster-lifecycle-cluster-api-alerts\n/remove-sig network /remove-sig node\ncc\nwill take a look /assign\nCheck the Calico version, there was a similar report that was fixed bumping the version. As a side not, I'm seeing a lot of problems related to Lots of projects/test/code were assuming the secret was created and is causing a lot of noise in different CIs\nHey thanks for the PR! The test still shows in the - will it take some time for the changes you made in the PR to reflect?\nhello! those are not failures, they are flakys :/ afaik\nHey folks, this test has now started failing again on the dashboard. Should we open a new issue for it or is it okay to track it here itself? !\nchecking that\nthe capg conformance failed but now have flakys only\nThanks a ton"}
{"_id":"q-en-kubernetes-3a31751d8f7078739317089d27ce716cd397688065176cd536a90311245cb433","text":"[X] test approval issue : [X] test pr : [x] two weeks soak end date : 15 Jul 2022 [x] test promotion pr : According to this APIsnoop query, there are still a single remaining Event endpoint which is untested. replaceCoreV1NamespacedEvent Using a number of existing e2e test practices a new has been created for this pod status endpoint. The e2e logs for this test are listed below. This query shows the event endpoints hit within a short period of running the e2e test. If a test with these calls gets merged, test coverage will go up by 1 point. This test is also created with the goal of conformance promotion. /sig testing /sig architecture /area conformance\n/triage accepted\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-fa1e3592e03be9bd9db82b235e463f224a9a94bfb5a37ca2bda074ac06ed4ee4","text":"Creating a destroying services test is a required test case in , but it's missing in the upstream e2e test. Since creating a destroying services test should work in both windows and linux cluster, it's worthy adding it to create a service and corresponding pod delete the service re-create the service load balancing rules for pods are recovered e2e tests for sig-network would have coverage for recover load balancing rules for pods after service is deleted and re-created. Reproducer is above No response <detailsmain </details<detailsany </details<details</details<details</details<details</details<details</details\n/sig network testing\n/triage accepted /assign"}
{"_id":"q-en-kubernetes-90016efc10ba60fd2e2129cd26baae3371a7d9a65592210e7cc6dfa34ade5cb3","text":"in this test the rootfs is limited to 5M and the pod is evicted by creating a 10M file with; But this fails if the fs doesn't support the fallocate syscall. The K8s e2e must not itself require support for features that K8s does not require. I am unsure why my overlay fs in containers doesn't support fallocate(2) while the fs on the node (ext4) does. I have queried on the slack. I build my Linux kernel myself but I can't find anything related. If the command is changed to; the test works on an fs that doesn't support fallocate(2). It may seem more \"primitive\" but it doesn't require a feature that K8s doesn't. <detailsNone </detailsNo response No response\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig testing\nInteresting /Triage accepted"}
{"_id":"q-en-kubernetes-7679fec6fc3307560ba7099cdd14fd03a7bc8aa630e2c9e270959a4136246b1d","text":"The project uses a lot of CRDs - in some cases almost 2,000. I do mean here, not custom resources. We've noticed that API server memory consumption balloons when many CRDs are installed - for example with ~1,900 CRDs we see the API server using up to ~8Gi RSS. A little profiling shows that around 1.5Gi of that memory is being used by the etcd3 client's logger. With Kubernetes built from with the CRDs at loaded I'm seeing the API server using up to ~7.5GiB RSS. I see it dip to ~5.5GiB, presumably when garbage collection runs. You can see taking up a ton of memory below. Zap is the logger used by the etcd client. Note that this is a sample, so it's actually using more than the indicated ~800MiB. ! I'm seeing the API server using up to ~6GiB RSS. I see it dip to ~4.5GiB, presumably due to GC runs. You can see no longer appears below. ! I would not expect a logger to be responsible for such a significant amount of the API server's memory consumption. You can get the above profile by: out . to build a kind node. to start a control plane for the above build. to allow anonymous pprof access the CRDs from I used to see the RSS but I imagine would show it too. See for details on how I disabled the logger to get the \"without logger\" numbers. No response <details</details<detailsN/A </details<details</details<details</details<details</details<details</details\n/sig api-machinery\nSince you looked at the code, and I'd rather not spend time doing that. Any chance you can tell if it's because it's being mis-used?\nThis is a scale dimension where the apiserver has historically . With the OpenAPI discovery performance being improved it looks like this might be a hint at what the next bottleneck might be? Any idea how many etcd clients were instantiated? I suspect it's quite a lot.\nSadly I didn't yet look deeply past turning it off. I was slightly optimistic that we might just be able to disable it without any meaningful impact. What I do know is that the offending code seems to be on the etcd3 client end - i.e. it defaults to creating its own logger when one isn't supplied. You can see that code at\nI don't - will try to find out.\nThe zap logs might tell you if you have the data. If my memory of the code is right, you should get 1 etcd client per CRD right now (plus some for builtin types).\n/triage accepted\nBased on my read of the code, you're right. I see roughly: All of /apis is handled by a crdHandler When a request comes in it gets an instance of the crd (from an informer) Gets serving info from a map of CRDs - if there is none it... Calls customresource.NewStorage which creates a genericregistry.Store Calls its CompleteWithOptions method That calls its “Decorator”, which is a genericregistry.StorageWithCacher That calls generic.NewRawStorage which just wraps factory.Create factory.Create calls newETCD3Storage which calls newETCD3Client I'm experiencing technical difficulties getting lsof running in my kind container but /proc/pid/net confirms we have a connection to etcd per CRD: Note that 1,937 - 60 = 1,877 - the number of CRDs we have loaded (the 1,878 above includes the header line)."}
{"_id":"q-en-kubernetes-64820056d60c9d73c0e10b019a42542bb69a2317edd6420a8df0a40f5860c9c1","text":"The gets very long testing mulitple functions in such as , and . These checks should be performed in dedicated Test functions using table for different scenarios. Target functions and are being as part of The remaining checks should be migrated to these functions. Also, a test function should be to migrate the remaining tests.\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig apps\n/assign\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\nAny progress on this? I'm working on and I noticed that has relatively low coverage so I was going to work on getting the coverage up.\nI created that is somewhat related to this.\nno progess. go ahead please\n/assign"}
{"_id":"q-en-kubernetes-647cba4e10e47735b73ba312460d98b00a45cf0c5a34d59d8529290e678b0deb","text":"When enable apiserver tracing, the webhook request spans only show spanName as /, mostly because use http request operations as default span name, which is hard to understand. ! Following the open-telemetry standard of spanName, it's recommended to use as spanName. ! SpanName of webhook request should be , like ! APIServerTracing feature gates of apiserver a pod tracing dashboard No response <details18 </details<detailsNone </details<details</details<details</details<details</details<details</details\nI make a fix for this.\n/sig api-machinery instrumentation\nDo you know if this is fixed in later versions of opentelemetry?\n/triage accepted /assign\nLooks like might be related\nnot yet, I think we can fix this in otel upstream, but it's hard to backport to otel , which is the version of otel used in Kubernetes\nGood point. It think fixing it here is a good idea in that case.\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:\n/remove-lifecycle stale\ncould I take on this issue?\nyes. Feel free to work on this.\n! I test locally by 1.29 kubernetes and start a sample webhook, looks like the client span has already conforms to semantic conventions , which should be in the form of \"HTTP {METHODNAME}\"\n! For apiserver as a server, now the spanName is still KubernetesAPI, which I think we could enhance it into \"HTTP get_account/{accountId}\" format\nAfter digging for a while, I found one solution to implement this, I draft this idea in , could you help review about whether this is a correct way to implement it?"}
{"_id":"q-en-kubernetes-1a1f837590f81f9849ac7605d8bd9eccee30ea8328192a45383fa25504691e30","text":"This one involves systemd, DNS, musl, and Kubelet v1.25 Starting in v1.25.0, pods running on Kubelets on certain nodes with a FQDN hostname (more on that later) will not be able to resolve DNS queries if they're using musl (e.g. alpine:3.16.2, busybox, etc). Our story starts with systemd. Back in 2020, systemd a \"search .\" in stub resolvers if it detected the hostname to be a FQDN (to solve a separate problem). Check for a static hostname with . In , a change was made to Kubelet to propagate that \"search .\" into pods (to solve a separate DNS problem). That change landed in and appears in . Unfortunatey, musl based DNS resolvers struggle with this \"search .\" EDIT: details in musl-based DNS resolvers (depite their other quirks) should continue to function To see if you have a suitable repro node, check your (likely a symlink to . Next, start a container (any CNI) that makes some use of musl resolvers. Confirm that starting in Kubelet v1.25.0, the container now includes the \"search .\" if it was present on the host. Try using musl-based DNS resolution. Note that uses its own DNS resolver, its not a reproducer. Note also that pod-to-pod traffic and outbound masquerade are fine too. This is not a masquerade problem. Edit the in the container to remove the \".\" and your musl problems instantly disappear. Add the dot and they're back. I've been spelunking for a while. I can confirm this is not related to a specific CNI - flannel, Calico, and Cilium all reproduce the issue. Not networking related OpenShift is not affected. In 2021, RedHat folks noticed DNS \"breaks\" and reverted the systemd change on their OS cc <details</details<detailsThis isn't cloud provider specific. However, the most likely cases where folks use FQDN's are on-prem imo. So you're likely to see reports originating from there. </details<details</details<detailsThis was found while preparing Typhoon's v1.25.0 release of Kubernetes v1.25.0 </details<detailscontainerd 1.6.6 </details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\nCould either revert (which propagates \"search .\" into containers) or find a different solution to the empty \"search \" problem it was trying to solve (does OpenShift even need that fix if they already ? cc A workaournd to problem is to configure hosts to strip \"search .\" from ala\nhas a great packet capture and this has been reported to musl\n/sig network /cc\nimpressive investigation in the linked issue , that seems to conclude that is a muslc bug, since it happens outside of Kubernetes too. Just reverting the PR in Kubernetes doesn't looks right, it fixed a known bug ... and this is really a muslc bug .... maybe always removing the single dot from ? I think that for Kubernetes we should understand first what does and means and its implications, maybe some developer can help? you know someone?\nIts discussed by the musl folks what means. It does not sound trivial to implement on the musl side, its a new case they did not anticipate supporting. Unfortunately this is subtle, and folks using alpine or related ecosystem things will get bitten and eventually end up here. I agree its a musl issue, but Kubelet now propagating triggers this I think its worth noting that did not set out to propagate , it was merely trying to prevent an empty . There may be better ways to achive that.\nIIUC means \"try the input name without any suffix\". Resolvers will do this ANYWAY, but it's implementation-defined whether they do it first or last or in parallel, so allowing lets node owners be explicit (though, curiously, it seems at least glibc may try the same name AGAIN at the end of the search list). This is pretty clearly a bug in MUSL, but I wonder if, instead of , we should have treated as rather than or the errant ? It nullifies the control that provides, but it wouldn't tickle musl like this.\n/cc I foresee a new wave of bug reports caused because of this muslc bug , better to keep CoreDNS friends in the loop\nThe patch for this got merged for the upcoming 1.25.1 and 1.26."}
{"_id":"q-en-kubernetes-9749eb152fe9034471d7606d65c1d581b51dd64e96b9b33d9a702d9b22cce3f4","text":"adds new version for in 1.26. For 1.26, we are setting the storage version to because vN-1 understands that level. In 1.27, we should change the storage version from to .\n/sig api-machinery\n/cc\n/cc\n/triage accepted\nThe Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs. This bot triages issues and PRs according to the following rules: After 90d of inactivity, is applied After 30d of inactivity since was applied, is applied After 30d of inactivity since was applied, the issue is closed You can: Mark this issue or PR as fresh with Mark this issue or PR as rotten with Close this issue or PR with Offer to help out with [Issue Triage][1] Please send feedback to sig-contributor-experience at . /lifecycle stale [1]:"}
{"_id":"q-en-kubernetes-3e83118de03d0fe14a8145255f936afcefd6219ebb776fa0fe02b7cedc1c13ef","text":"Upgraded from Kubernetes v1.25.0 to v1.25.1. Standard metrics-server deployment now causes the following message to appear in the kube-apiserver logs on a recurring basis: Adding a quick hack debug Infof print shows: It appears that the apiserver is sending an If-None-Match cache control header when refreshing the openapi spec, and the metrics-server service is correctly responding with a 304 Not Modified. No errors logged; proper refresh of OpenAPI specs Deploy metrics-server, or any other service that uses apiserver aggregation and honors caching headers such as , to a Kubernetes 1.25.1 cluster. made bad assumptions; not all HTTP 3XX responses are redirects. It should probably only reject the response if it is 300-399 AND the header is set. <details</detailsnone <details</detailsk3s containerd flannel\nTaking a stab at some tags here... /cc /cc /cc /area apiserver /sig api-machinery\nthanks for the report, looks like this should be /priority critical-urgent /kind regression cc since this requires a respin of a final 1.22 patch release :-/\nFix open in , will get cherry-picks open to start greening up CI\n/triage accepted\nI need to make a redirect to the external URL but it's not allowed :("}
{"_id":"q-en-kubernetes-2fe80e49aa5ae53d41f8add8e37445a19adaf3497eeef8427dfc711a0635077f","text":"When a Pod admission fails due to node-pressure condition then the Pod's looks like this: \"Pod The node had condition: [DiskPressure].\" The \"Pod The node...\" reads awkward. A message like: \"The node had condition: [DiskPressure].\". a job with a pod on a single node cluster, when the node is under disk pressure (then it is tainted with ). The Pod gets stuck in the phase. the node by kubectl command line: Pod is scheduled but admission by Kubelet fails as the taint is re- shortly after its manual removal. Then, the status of the Pod contains the awkward message. No response <details</details<detailsReproducible on kind </details<details</details<details</details<details</details<details</details\n/sig node /triage accepted /priority backlog\nThe first part () is returned here: . I wonder where the is being appended to the beginning. It's my first try to contribute, so I am a little bit lost.\nI think this is a candidate place which might be responsible for the prefix (but haven't checked): I would suggest you to try to reproduce the issue locally first then experiment with modifying the code.\n/assign\nthanks, I've managed to reproduce and used Delve to make sure who's responsible. I opened the changing the message. If it's accepted it will change to ."}
{"_id":"q-en-kubernetes-ce0e421c34eed98860db911d5d334cd66db99e26a231af53f976763de79a714f","text":"A common pattern in Kubernetes (both internally and in third party usage) is something like: (this is from the ). This is commonly called at a fairly high rate; in our case we would actually end up seeing this account for significant portions of CPU/memory to compute the and run on it. In our controller, as an optimization, we replaced usage of and just used our own simpler label matching code, leading to a 70% speed up in our function (which does more than just match) and 96% allocation size improvement; see I believe similar could be done in Kubernetes directly by having a specialized Selector implementation for simple labels. This would transparently given substantial performance benefits to anyone using the library\n/sig api-machinery\nis the change here a new kind of selector matching or a tweak to something existing? It's hard to tell if this is proposing a new design to partition types of selectors that needs a kep versus a perf tweak to existing code that is straight-forward.\nThe proposal is an internal performance improvement to make a specialized implementation of the selector matching internally. So two concrete implementations for 1 shared interface. is a PoC of this. I think it can be extended (by me, or others if someone wants to) to complete it\nWe discussed this one during our bug triage. There seemed to be the feeling that this is large enough to need a KEP? Would you be up for that? Additionally you could bring it up in our SIG meeting to socialize / discuss it more if you wanted.\nThe patch shared in helps to explain, it's a lot smaller than the description made it sound. This is a special case for equality-only selectors. Probably still worth bringing up at a sig-meeting, but if if the patch is a good representation, it is more bounded than expected from the issue description.\n/triage accepted"}
{"_id":"q-en-kubernetes-b4de8d19fa7577b91e9d20fe50de242313fdcfa4e902ab568769fed1654a59ef","text":"Created limit range object with this parameters then tried to create this pod got message: Error from server (Invalid): error when creating \"pod-to-be-\": Pod \"nginx-to-be-removed-4\" is invalid: spec.containers[0].resources.requests: Invalid value: \"700m\": must be less than or equal to cpu limit Pod should be created with Requests & Limits of cpu equal to 700m (since it is in range of min/max cpu limitrange object) create limitrange object and pod as in description No response <details</details<detailson premise/AKS </details<details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\nYou have a default CPU limit of 10m, and no explicit CPU limit on the container, but a request larger than the limit. The request must always be less than or equal to the limit. Your pod needs to also set a higher limit, or the limit range needs to not set a default limit.\n/remove-kind bug /kind support\n/cc\nbut this is not obvious behavior since there is min & max set in limit range. This is limits (not min/max) that should be overwritten by pod. limits should be set to requests in this case imho\nlimit range limits what the user can set, it doesn't perform any mutations other than the defaulting and doesn't claim to. Min/max enforce that pods cannot set higher / lower. The default ensures that if the pod doesn't set a limit it will have one. personally I would remove the default limit, especially at 10m, this will aggressively limit pods. it would be, if the pod set a limit that would be a substantial breaking change, I don't think we could do that short of a v2 API and I'm not sure it would be correct behavior even then. we could make it opt-in, but then, why not just set them to match on the pod to begin with?"}
{"_id":"q-en-kubernetes-1a929f3f32506a77af05bbb3d419a7027c81550ce76cd4ecbbbd64d2b06017fd","text":"As per, , wire up metrics/slis for ccm\n/assign /sig instrumentation /sig cloud-provider\nGitHub didn't allow me to assign the following users: Richabanker. Note that only , repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time. For more information please see <detailsIn response to : /assign /sig instrumentation /sig cloud-provider Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\n/triage accepted\n/remove-triage accepted\n/triage accepted"}
{"_id":"q-en-kubernetes-3adf443a08b875ded3a1c0a6a2e6a06ab5e3f98761de8f84e4c945877614c7fc","text":"After merged, build logs for e2e tests have started showing stack traces that look like: It seems like the used in this is nil I have a suspicion this is happening for storage tests that skip No new stack traces in build logs Run e2e tests against the in-development version of kubernetes I identified when this started happening by looking for an increase in build-log size is the first build where this started happening, which happens to be the first build after merged, which is the PR that introduced this new way of invoking <details</details<details</details<details</details<details</details<details</details<details</details\n/sig testing /triage accepted /assign To put this on your radar\n/cc\nCorrect. This happens because of the ordering in Tests gets skipped before is called. This avoids the expensive namespace creation/deletion for tests that don't need to run. The for then gets executed although the framework didn't get initialized. My preferred solution is to document that the framework instance might not be initialized in an AfterEach callback and add a nil check to .\nSee Only was affected. The other framework extensions use DeferCleanup, which only runs when needed."}
{"_id":"q-en-kubernetes-05984e835f1e73cb4c0b2223127020e6f9a06b8a22eb8f8d9b6ab8ff80249f40","text":"The replicas is 3 but just create 1 pod 'web-0' and one pvc, then kube-controller-manager panic happend. Should create 3 pod and 3 pvc add '--feature-gates=StatefulSetAutoDeletePVC=true' arg for kube-controller-manager, then create statefulset with 2 replicas and volumeClaimTemplates. No response <details</details<details</details<details</details<details</details<details</details<details</details\n/sig apps\nI can reproduce the issue and want to fix it. It seems like variables cause . /assign\n/triage accepted\nGreat find! Does this only happen on an upgraded cluster? I'll spin up a cluster and see if the e2e test catches this problem.\nThe e2e tests are still passing on a 1.24 cluster, so I suspect this is an upgrade problem. I wonder if there are other holes here.\nI installed the Kubernetes 1.25 cluster directly using and still reproduced this issue, so it was probably not caused by upgrading the cluster.\nInteresting. I thought that would set the policy to be non-nil. Let's keep this bug open after you submit your PR as I think something else is going on here---I don't think we should be seeing the null policy in the first place.\nI suspect that somewhere accidentally set it to nil, and I'll keep looking at this code.\nI tried the following (and also with parallel pod management), and it didn't seem to cause the panic. This was a 1.24 cluster, maybe this is only happening on head versions?\nI just tried this on a kind cluster running on k8s head, and the problem didn't seem to reproduce. But this was setting the feature flag default in I wonder----does the api server need to have the feature gate as well? That would explain why defaulting wasn't happening.\nAs far as I'm concerned,I just set the in kube-controller-manager manifests to reproduce this issue. I repeat the yaml used as follows: I tested that in statefulset yaml, setting or not would cause the component to panic. The panic logs are the same as above, so ignore them here. statefulset with : The environment\nI believe you---this doesn't seem to be well-documented, but it's my understanding that you need to set the feature gate on all of the system components. Obviously failing to do so shouldn't cause a panic, so thank you again for finding & fixing this bug! Let me ask around to see what the expected behavior is. I know there may be skew between kubelets and master versions, but I didn't expect to have to deal with skew between master components.\nI read the code, and now I understand what you mean. This feature gate must be enabled in both the and components. We need to describe it in the to prevent someone from ignoring the component, which also needs to enable the feature gate.\nI don't think we need all the system components, like the component, right? I read the original KEP documentation. This feature gate only involves components and :\nI had a bit of discussion about this with others. Because upgrades don't occur atomically, any feature gate has to work regardless of how the other system components are configured (things shouldn't panic!). I suppose if the cluster administrator wants to track what needs to be enabled on each feature they can try to sequence which component has which feature gate. If it were me I'd try to make all the components consistent though.\nThis is certainly a method, but it is a bit more cumbersome for the administrator. It is best to have a place for the document to be queried by the administrator and avoid confusion. I don't think everyone is very familiar with the k8s mechanism. I suggest adding a partial description to this section of the : Before: After: I can submit this PR on .\nThis is going to be true of any api change. When you tried to use it with only adding the controller manager feature gate, I think you would be able to set the retain policy, but there wouldn't be any defaulting or validation (?).\nI've seen other feature gates that say what components are needed to enable it to use the new feature. For example: If it's not necessary, then I close .\nonly adding the controller manager feature gate, I think you would be able to set the retain policy, but there wouldn't be any defaulting or validation (?). I remember just enabling it in kube-controller-manager. it using the retain policy, but the statefulset does not have the policy from output of (It should be the apiserver authentication forbids it)\nYeah I think that makes sense. Thanks for finding this issue! After your PR with the panic fix is submitted, I think we can close this?\nYes. I agree. /close\nClosing this issue. <detailsIn response to : Yes. I agree. Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-8d104fb08a54b1aafbf227b5ff87163913c73afc8318e39b926ee160b7ff0192","text":"release-master-informing: capz-windows-containerd-masterKubernetes e2e suite: [It] [sig-node] Container Lifecycle Hook when create a pod with lifecycle hook should execute prestop https hook properly [MinimumKubeletVersion:1.23] [NodeConformance] Kubernetes e2e suite: [It] [sig-node] Container Lifecycle Hook when create a pod with lifecycle hook should execute poststart https hook properly [MinimumKubeletVersion:1.23] [NodeConformance] Both the tests have been failing during most of the runs since 20th October Kubernetes e2e suite: [It] [sig-node] Container Lifecycle Hook when create a pod with lifecycle hook should execute prestop https hook properly [MinimumKubeletVersion:1.23] [NodeConformance] | 1m6s -- | -- {Oct 31 18:42:02.425: Timed out after 30.001s. Expected <errors.errorString | 0xc0015283b0: { s: \"failed to match regexp \"GET /echo?msg=prestop\" in output \"I1031 18:41:06. 8988 ] Started HTTP server on port 9090nI1031 18:41:06. 8988 ] Started UDP server on port 9091n\"\", } to be nil failed (0xc000f1d000) +0x469 () +0x335} Kubernetes e2e suite: [It] [sig-node] Container Lifecycle Hook when create a pod with lifecycle hook should execute poststart https hook properly [MinimumKubeletVersion:1.23] [NodeConformance] | 5m11s -- | -- {Oct 31 19:05:50.205: timed out while waiting for pod container-lifecycle-hook-6892/pod-with-poststart-https-hook to be running and ready failed (PodClient).CreateSync(0xc000eb5920, 0x22?) +0x94 (0xc0045e8c00) +0x73 () +0x325} No response_ /sig windows cc\n/triage accepted /cc\nThese failures happen also with Kubernetes master + containerd master with Flannel CNI. Testgrid link:\nI figured out why these new failures appear. I've submitted the fix via PR\nThis issue is also affecting 1.25 tests: If possible, we should cherry-pick to the release-1.25 branch once it's merged on master."}
{"_id":"q-en-kubernetes-1237b61a9a420f0c0b04915325043b5410a2d3bac53dd2246fffcf46a7cfe4ca","text":"Why is this needed? The functionality that the package provides is already present in golang's builtin errors (using in ) and the github repo is now read-only. /kind cleanup /priority backlog /good-first-issue Notes to people that want to collaborate Avoid creating a single PR for the entire repo. Create multiple PRs, each targeting a specific SIG. Do not submit PRs for\n/area code-organization\n/kind cleanup\nthis is only partially true, e.g. %w does not provide stack traces\n/assign\nI see. Should we close this issue then?\nI think it depends on how it's being used. If the traces are not the goal then we can do a simple substitution. It's probably best to drop read only dependencies ASAP but it's worth noting that if we depended on the full functionality in any way we'll need a more complex replacement (KIND did/does, not sure about k/k)\nk/k git log suggests that some cleanup has already been done in the past (maybe in other commits too): This issue triggered 3 PRs and 2 of them are merged. there does not seem to be calls to so most cases it looks it's just wrapping errors.\nfor cmd/kubeadm in kubernetes/kubernetes we decided to not drop usage of pkg/errors even if it's a deprecated project. prior ticket and comments are here\nfor cmd/kubeadm in kubernetes/kubernetes we decided to not drop usage of pkg/errors even if it's a deprecated project. prior ticket and comments are here\nI see. Added a note.\nis it still open ?\nshould be replaced with the standard library for the following reasons: archived library with no maintenance or security updates the go community used pkg/errors to introduce eventually into the standard library hence the approach that was selected is more idiomatic stack traces are hardly used in error wrapping and are not a reason not to refactor\nOn place left in Except the k/k repo, as is archived, some other repos may need an update.\n/reopen\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/triage accepted\nThe label cannot be applied. Only GitHub organization members can add the label. <detailsIn response to : /triage accepted Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\nis it necessary to remove \"\" from every location? I noticed it in many places. Also, do I need to create separate PRs for each file where I remove it?\n/assign\nFrom kuberentes/kubernetes repo view, only two locations of kubectl code should be updated. (Do not submit PRs for ) see For other repos, we should follow up until the dependencies' repo is updated. I will track them in\n/assign\n/assign\nto remove dependency you can use the standard Go error package or creating your own handling package, this could contain custom functions for generating and handling errors.\nlet me work on this issue now please /assign\nWe are in code freeze. I suggest waiting for a couple of weeks.\nHello, is this still up?\nLet's get this going\n/assign\nJust opened a PR for this issue\n/assign\nUpdate the remaining todo items.\n/sig architecture umbrella / code org\n/assign\nHey Are you still working on this one? If not- would you mind if I take a stab at it? Thanks!\nIs this issue really still open? When I do I only get hits for cmd/kubeadm. Am I missing something?\n/close\nClosing this issue. <detailsIn response to : /close Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-ec1f2b5e3111beba9c3d5972d4887c5fb5222f18baca7cc1195b15978a145ffe","text":"It was found during a review that stateful set status might not be correctly updated when new revision is being rolled out and API error occurs when deleting a pod. The status should be updated correctly Not easy to reproduce, but should be possible with this unit test No response_ <details</details<details</details<details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig apps"}
{"_id":"q-en-kubernetes-f2910a5a7757744dd88700c551fb3f5fe81db92c4ae532e768f1444ef3ba7b10","text":"It's at line 111 When the server stopped before timeout, the first goroutine will be block on because the receiver in the second goroutine is not exist. Changing to a buffered one may fix it. no goroutine leaks run related test with leak checker No response <detailsdev </details<detailsunknown </details<details</details<details</details<details</details<details</details\n/sig api-machinery\n/assign /triage accepted\n/cc\nIt should close the channel rather than write to it, that's the easiest fix"}
{"_id":"q-en-kubernetes-7718a29788b98b66982a35f9fbeb8ad977aaca8c841112c95b4581bb5c967dce","text":"Events from node taint manager doesn't report involved object apiVersion: ! apiVersion is reported as other events. For example: ! a pod on node A (not master) off node A is reported, but without apiVersion in involvedObject. The apiVersion is not reported in The issue is still not fixed in master: ! <details</details<detailsIrrelevant </details<details</details<details</details<details</details<details</details\n/sig node /sig apps\n/assign In the latest version, the issue is confirmed\none suggestion you should wait for the corresponding SIG to triage the issue and accept it before sending PRs, the chances of your PR getting reviewed and approver are much higher\nYou are right sorry about that. It won’t happen again.\nno no worries, it was a suggestion, is not that you did something wrong, but it is possible that your PR doesn't get reviews and you get frustrated\n/remove-sig node please re-add if this is node-related issue\nI think it is a node related issue. The root cause in taint manager.\n/sig node\n/assign /triage accepted"}
{"_id":"q-en-kubernetes-00d3f41b4011457970a44dc6eb8b0abccf584d90cb550d7555d38ecc0463a883","text":"For 1.27 we should graduate the ReadWriteOncePod feature to beta. This is needed to rollout the feature to GA. In beta this feature will support preemption of pods using PVCs. See:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig storage /sig scheduling"}
{"_id":"q-en-kubernetes-f808b7b53b871e2ad503a490ab8a8853ac035cebad77b88506654bcda75080a5","text":"A status of Pod mounted NFS PV remain as terminating when a node shutdown and restart. A Pod is completely terminated. Deploy a Pod mounted NFS PV Shutdown the node where the Pod is running Wait for 5 min and the Pod status become Restart the node The Pod status is still and the Pod is never terminated completely <detailsPrepare a NFS Server.(I used ) Deploy a Pod mounted NFS PV like this. Shutdown the node where the Pod is running.(In this case ) And the status of node02 becomes . After 5 min later(by default), the Pod status becomes . (The time a Pod becomes status depends on .) Then, restart the node and the status of node02 becomes . But the Pod status is still and the Pod is never terminated completely. It seems that a faild to delete Pod because of trying to unmount a not mounted directory. Force delete Pod. Delete Pod's directory on the node. </detailsNo response <details</details<detailsnone(on premise) </details<details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig storage\n/sig node\nIt dose not happen using , so it seems to depend on NFS volume plugin.\n/assign\nI try to build and run kubelet with comment out this logic. Then it seems to get well. related: This change checks the result of instead of the mount point check. However, check logic is not implemented in and (these functions are used in NFS volume plugin), despite it is done in .\n/remove-sig node /sig storage\nReopening, fix had to be reverted due to unit tests requiring root permissions ()\nI am terribly sorry... Please wait a moment. I'll try to fix it.\nno worries, just get to review the rework, making sure the units tests don't require root permissions\nThank you! Fixed in .\nAre there any plans to backport this to older Kubernetes versions? This seems like a critical fix."}
{"_id":"q-en-kubernetes-f8e0420493392a2143d7cf5c6fd188de94eceb8d975eec8e70d5b38a422ea455","text":"It \"connects\" to dockershim There is no Run crictl v1.26.0, without setting up (or using -r) Worked in crictl v1.25.0, that still had CRI v1alpha2 This patch fixed the issue, to make the fallbacks work: <detailsv1.26.0 </details<detailsLima </details<detailsPRETTY_NAME=\"Ubuntu 22.04.1 LTS\" </details<details</details<detailsv1.6.12 </details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\nI agree that we can check for unavailable connection because it will only be done on kubelet startup."}
{"_id":"q-en-kubernetes-bea37ab7d4f377c040141f3da5d077816029ad50feb7b6c4d294ebb6078f8089","text":"If a CR converter is wired in and it modifies the input list, the delegating CR converter does not protect against this. A converter cannot effect the input list Not really possible externally - it's a coding issue that the current converters don't hit. No response <details</details<detailsN/A </details<details</details<details</details<details</details<details</details\n/sig api-machinery /assign\n/triage accepted"}
{"_id":"q-en-kubernetes-5ae73f290b7cbed5a20c411ecd6584ae3e0bb9d2c39b6882eb361e82242274a2","text":"It seems that it was recently and it was always failing ! /sig cli /kind failing-test /assign\n/triage accepted"}
{"_id":"q-en-kubernetes-67ba6216c2f0a82dc146b5437e105db9d39096631ebad5fab166c989773fd1f7","text":"master-blocking build-master Overall build is failing ! 2023-02-09 23:09:58 +0000 UTC make: [Makefile:487: cross] Error 1 !!! [0210 00:35:55] Call tree: !!! [0210 00:35:55] 1: kube::build::runbuildcommandex(...) !!! [0210 00:35:55] 2: kube::build::runbuildcommand(...) make: [Makefile:374: release] Error 1 level=fatal msg=\"failed to run: running make release: command /usr/bin/make release did not succeed: !!! [0210 00:35:54] Call tree:n!!! [0210 00:35:54] 1: hack/make- kube::golang::buildbinaries(...)nmake[1]: [Makefile:92: all] Error 1nmake: [Makefile:487: cross] Error 1n!!! [0210 00:35:55] Call tree:n!!! [0210 00:35:55] 1: kube::build::runbuildcommandex(...)n!!! [0210 00:35:55] 2: kube::build::runbuildcommand(...)nmake: * [Makefile:374: release] Error 1n\" No response_ /sig testing\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/priority critical-urgent\nlooks like the same issue again: /cc\nThis would fall under sig-release or sig-testing? couldn't really gather much from logs about this though failure is so I guess sig-release then.\nI'm going to reopen for that, which should work around the issue.\nthat's correct :) /remove-sig testing /sig release\nThe PR does not seem to fix the problem without disabling test cases. Opening a slack discussion and closing the loop:\nFailing again. /reopen\nReopened this issue. <detailsIn response to : Failing again. /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nthis is a different error: can we spawn a new issue to track rather than reopening the same one and mixing investigations for different issues?\nOh just the stack trace for make was the same, my bad, opened /close\nClosing this issue. <detailsIn response to : Oh just the stack trace for make was the same, my bad, opened Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-bf562bafe869012db15c0a17b6d5b8a3c02c53240756e36c1b24f8106551c642","text":"Currently, when the Label Selector in the Pod Affinity is empty, no objects will be match. That means if it's in a required PodAffinity, the Pod will never be schedulable. And if it's in a required PodAntiAffinity, that PodAntiAffinity will be no-op. So, I'm thinking PodAffinity with empty Label Selector is very likely mistake. It'd be good if we could notice it for users. To prevent such a misconfiguration. Probably, we cannot filter out such a Pod creation in the validation for a backward compatibility. /assign /sig scheduling\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI'm wondering if it has already discussed somewhere or not. Please just let me know if we have some decisions on it.\nI am not sure, where do you propose to put the warning? If the pod is unschedulable, that will show up in the events and pod status.\nDo you mean the place to implement this warning? I suspect ↓ is a suitable place:\nA warning sounds useful, even if the pod is unschedulable, as it gives you a heads up at the time when you create the Pod. Whereas the Pod scheduling failure might be a bit harder to interpret.\nok, I thought you were referring to something that the scheduler will produce after attempting to schedule the pod. Yeah, if there is already a setup to print warnings on pod creation, it is certainly useful for this case.\nCan you please also add it in topology spread?\nSure"}
{"_id":"q-en-kubernetes-a2a207ab5c5251abdddcfb02c46fba9f20ebf7280d6accb8963d5b3a92615c69","text":"The EndpointSlice API says and it works as expected in normal circumstances I delete the pod and the endpoints conditions are the expected However, if during this time I set we run into an incoherence with the API doc, but it seems that we already documented this edge case in the code I think that at this point is just to amend this information to the API, no? /kind bug /kind documentation\n/sig network /assign\nI would like to work on this. I just want a bit more clarification on what needs to be done!\nI proposed a solution, that is clarify it in the doc ... once we reach an agreement in this issue we implement it ...\n+1 to docs. I'd make it a little more vague, like: \"\"\" when the normal readiness behavior is being explicitly overridden, for example when the associated Service has set the publishNotReadyAddresses flag. \"\"\"\nso, the thing is to update the docs here with Tim's suggestion and here and do to update the generated code and submit a PR\n/assign\nI am not able to run command in windows, I am checking it. This error message is coming: *\n/triage accepted\nIs there a PR for this? Happy to review\nI will raise the PR if my above problem gets solved. I am trying to make an update but the error mentioned above is coming.\nWhat platform are you on? If the Makefile were really broken EVERYTHING would be failing.\nhe's on windows ... you will need a linux/macos. see\nLet me know if you want someone to take it over.\nI think in that case, you can assign this to someone else."}
{"_id":"q-en-kubernetes-04e3b909e513627c5b2af9f40bf2e8ca20eddfe7646c93203a450ce8a9a58db3","text":"From the Deprecated API Migration Guide: , since v1.19, the has been using the new reportingController and reportingInstance to replace the decrecated source.component and But when I checked the Kube events in a cluster with version 1.26.0, those new fields are empty and the supposed to be deprecated fields have not been renamed. Seeing the values in the new fields like reportingController and reportingInstance an Kubernetes cluster with version 1.26.0 some Kube events the Kube events details for the field of reportingController, reportingInstance, source.component, No response <detailsServer Version: version.Info{Major:\"1\", Minor:\"26\", GitVersion:\"v1.26.0\", GitCommit:\"\", GitTreeState:\"clean\", BuildDate:\"2022-12-11T00:15:47Z\", GoVersion:\"go1.19.4\", Compiler:\"gc\", Platform:\"linux/amd64\"} </details<detailsAzure </details<details</details<details</details<details</details<details</details\n/sig instrumentation\nI have also encountered this issue in my local 1.26.0 Kubernetes cluster, which does not align with the deprecation guide described in I believe I can investigate this further\n/assign\nThank you for making the PR to fix this! Also a comment in the PR for the doc update.\n/triage accepted"}
{"_id":"q-en-kubernetes-1f4123c828672256c348bc539b38c34a946e71292e23552bc51c4890f70b8975","text":"tc.shutdownActive is captured by 'func' literals in 'go' statements(Line177), it may cause a data race. I use t.Run(..., ){...}() or t.Run() {...}(tc) to avoid it . I don't know whether it is accurate, so I 'm looking forward to your reply. Data Race Nil No response <detailsNil </details<detailsNil </details<details</details<details</details<details</details<details</details\n/sig node\nTarget file:\n/area test /priority backlog\nhave you observed any test failures or this is found by code analysis?\nYes, it is found by code analysis. In terms of the source code, the scenario can happen.\n/assign\n/remove-kind bug /kind failing-test since it is not a product bug, but the test issue\nAlthough each test case calls a goroutine, tests run serially and block on the goroutine return, so it doesn't look like it could cause an issue. However, to be safe we can create a local copy of inside the loop. I'll send a PR briefly.\nCreated to address this.\n/triage accepted /priority important-longterm"}
{"_id":"q-en-kubernetes-f7b92d2ae4c7de8a3091c04ca1d11b65c719334aa69b8d7a201e03a33f42c804","text":"Example: this PR seems to be flaking: comparing age of a node from the output of two commands: Kubernetes e2e suite: [It] [sig-cli] Kubectl client kubectl subresource flag GET on status subresource of built-in type (node) returns identical info as GET on the built-in type expandless 5 hours back No response No response No response_ /sig cli\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/priority important-soon\nit seems that is comparing subresources and there is something there that changes over time As Sergey correctly points out is the Age The test can not assume the executions will be done in the same second\n/assign"}
{"_id":"q-en-kubernetes-1d19a749ded70ac40f5a4e3891ac8db5a4f9a3cf9321999067374a962e0dd578","text":"The following log line is often (~30times) logged during our tests in : . This suggests a possible bug as it was assumed to do not happen. No log line saying . Look at artifacts of all (or almost all) builds of: No response <details</details<detailsReproducible in tests </details<details</details<details</details<details</details<details</details\n/sig node FYI\n/triage accepted since it's a new error, it is a regression. So adding a corresponding priority /priority important-soon\nare you working on this? Should I assign bug to you?\nYes, I'm taking a look at this. /assign\n/cc\nThis is a metrics only error (we don’t report a fraction of metrics) and has no functional impact.\nThat makes sense, but is it expected that this is a valid state in first place? . My assumption was that if static is false, (i.e. API pod) then we should have a config?\nI took a closer look at the logs, it seems that this is happening when we enter while a static pod is being synced which has UID reuse and another pod which has the same full name did not terminate yet. In this case, we block the static pod from starting i.e. () returns false: Because the pod can't start, we don't set : which means that when we log metrics there is no set and thus or are both defaulted to false:\nBased on the above, it looks like this case is expected. As a result, maybe the simplest option is to simply drop the log statement since by design we only want to into account a subset of the states for the metrics?"}
{"_id":"q-en-kubernetes-37578c7ef1fc35bcc9b20a625a2453047827b433a40713775cf7720c37f7ee33","text":"If a pod is created and then deleted an entry in processedPods map remains forever (sometimes, it's racy). The problem comes from the fact that calls volume_manager asynchronously. If the last call happens after populator cleaned up resources, an entry is to processedPods and will never be seen. An entry in processedPods map must be deleted if pod doesn't exist and pod runtime should removed according to . To verify that the issue really exists I applaied a simple printing resources in map. Here is a reproducer: brand-new fresh k8s cluster with entries in : a pod, check its uid: that this uid is accurately reported: the pod, wait for a few mins, check for again: The entry will be kept forever because kubelet doesn't know anything about this pod anymore: so no more events will ever come to populater for this uid, and the loop in won't find it because it was already deleted from . n/a <details</details<detailsany </details<details</details<details</details<details</details<details</details\n/sig storage\n/triage accepted\n/priority impoerant-soon\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /priority impoerant-soon Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/priority important-soon\n/assign\nPR () fixed this issue. It is ready to merge since Mar 21.\n/assign"}
{"_id":"q-en-kubernetes-482aab5c60c854e49844e9f7b501458de493923e9ab6838dc8496864ffefd996","text":"Not , but removed: Looking at this comment: this should be removed once the migration to lease in leader-election was done. Can this be removed already? I mean, the following 2 lines: Security folks internally were asking if we could get rid of this upstream if it is not needed anymore (if this \"migration to lease in leader-election\" was fully done)\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig sig-scheduling\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /sig sig-scheduling Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig scheduling\n/cc\n/assign\nSupport for endpoints and configmap based leader election has been removed with this PR: Validation to error out leader elections configuration for resource lock type other than \"lease\" has been implemented in With this, it looks safe to remove the two lines mentioned in this issue."}
{"_id":"q-en-kubernetes-8ef15e9cb2af20fdd7b292d3dde46c3ffb64b6b0323788d679b26abe3d6fd30a","text":"Followup feedback from In a few e2e node tests (e.g. tests of deletion grace period) we have some busybox pods that wait for SIGTERM and perform some action like a small sleep to simulate SIGTERM taking some time. These tests follow the pattern of -- see The will block the SIGTERM handler from being run until after the time is elapsed. We should consider to remove the and instead use to avoid blocking the SIGTERM handler, see pattern followed in n/a n/a No response n/a n/a <details</details<details</details<details</details<details</details\n/sig node /cc\n/assign\n/triage accepted"}
{"_id":"q-en-kubernetes-1ffde837085a7cf8fda14061815d3ec79e4cb9af8d9e5a4c3a326fb5831a1bf9","text":"No response No response /sig windows /kind failing-test\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\n/sig windows\ni think this problem can be fixed when it imports os/net in Am i right?\nI also found some other unit test bugs in CI, such as running linux-related tests on windows. I will try to fix them. /assign\n/cc\nThanks for opening the issue! I didn't notice the failure through the testgrid dashboard. I've sent a PR here: . While fixing this issue, I noticed there's one unit test that was failing, so that PR also addresses that as well.\nYep, the imports you mentioned are needed. Keep in mind that imports are in alphabetical order. But in addition to that, there was another failing unit test which was masked by this import errors."}
{"_id":"q-en-kubernetes-9969bd986508ea8ad3b87aed5184f6e5ca5490b6e36b50732a630680ba555908","text":"<!-- Please use this template while reporting a bug and provide as much info as possible. Not doing so may result in your bug not being addressed in a timely manner. Thanks! If the matter is security related, please disclose it privately via --What happened: Following the example here: Everything here works when you use , but when using 1.27.1, I see the following panic: What you expected to happen: When downgrading, this works as expected: How to reproduce it (as minimally and precisely as possible): <!-- Please make sure you are able to reproduce the bug using a supported version and version skew of Kubernetes. See for which versions and are currently supported. --You will need a Kubernetes cluster with Agones installed (I'm testing on 1.25.x, as you can see above), then follow this quickstart: Anything else we need to know?: The is implemented as an API Extension, not a CRD -- which is probably where the issue lies. Looking at the line: There seems to be an assuming that there will be a under and , in - there should be some nil checks in there. Environment: Kubernetes client and server versions (use ): See above Cloud provider or hardware configuration: GKE OS (e.g: ): Linux\nseems like an another edge case; Due to the server version is old and openapiv3 is returning empty?\nThanks for opening this issue /triage accepted\nI'm digging into the Agones code, but I'm 99.4% sure we don't even implement openapiv3 at all for our api endpoint\nYep, confirmed -- I see a bunch of 404's in my API Extension logs for requests for\n/transfer kubernetes /kind regression\n/sig cli /assign\nI can also work on tomorrow if Sean can not get time to look at it\nIf nothing else, this does also force me to publish an openapi spec for our API extension\nnot sure if needed to handle nils inside the schema better, or if didn't fall back to v2 correctly in the absence of v3 data\nIt appears to be the former, where an OpenAPI V3 document is returned without , while the code assumes this field is non-nil.\nseems like that should still fall back to v2\nCan we get a kubectl -v8 to see what actually happens with downloading the OpenAPI?\nAs requested!\nThanks for the quick response. It looks like the OpenAPI V3 document is almost completely empty The code is expecting an Open API document with . I will have a fix out shortly.\nI was digging through that - I found that really interesting. We only expose an basically empty spec for the api extension. It's always \"just worked\", so never needed to update it. I am curious where K8s is even getting the something that it is outputting for , but that's a side note."}
{"_id":"q-en-kubernetes-f7bde9e341cfa5fb4aef7bf1e2ef4213057d2455b858e397a168aa2266e9e827","text":"is recording the time that a Pod is gated. We use the timestamp when the scheduler inserts the pod into the queue: The period of time when a Pod fails PreEnque (like being gated) shouldn't be accounted in the podschedulingdurationseconds. Create a Pod with scheduling gates. Wait some time before removing the gate. Observe the metric No response <details</details<detailsAny </details<details</details<details</details<details</details<details</details\n/sig scheduling\n/help /triage accepted\nThis request has been marked as needing help from a contributor. Please ensure that the issue body includes answers to the following questions: Why are we solving this issue? To address this issue, are there any code changes? If there are code changes, what needs to be done in the code and what places can the assignee treat as reference points? Does this issue have zero to low barrier of entry? How can the assignee reach out to you for help? For more details on the requirements of such an issue, please see and ensure that they are met. If this request no longer meets these requirements, the label can be removed by commenting with the command. <detailsIn response to : /help /triage accepted Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nwould you want to take a stab on this issue?\nYup, I can give it a shot. /assign helayoty\n, as we discussed, we can address this issue with different solutions: the pod timestamp to the current time , so the gated time is substituted. timestamp . Use a pointer to the InitialAttemptTimestamp and interpret as the newest timestamp so the pod will go to the back of the queue. I'll go ahead and implement solution no. 3, wdyt?\nsgtm. Any concerns ?\nI think it should be option 1 + 3, and pInfo.Timestamp should be pInfo.InitialAttemptTimestamp, right? as records the time based on :\nI think we should revisit this issue, in terms of its assumption: So is it reasonable to always exclude time spent in PreEnqueue from ? Note that PreEnqueue != SchedulingGates, is one of the PreEnqueue's usecases. In other words, some user of PreEnqueue may want to include the time spent in PreEnqueue.\nis a signal that cluster providers can use to build SLOs. The time spent before passing PreEnqueue is not under control of kube-scheduler, so considering it would not allow to build an SLO. Are you aware of other users of PreEnqueue?\nTrue, but comparing to state that metric X can or cannot be used to build SLO (which is hard IMO), I'm in favor of giving suggestion on SLI/SLO's semantic definition, for example: the official SLI definition of has a footnote clearly explaining its definition: So if a pod is gated, it's not \"immediately schedulable\", and hence doesn't fall into the definition of podstartup_latency.\nA viable solution might be providing an extra label like to explicitly exclude the time spent on preenqueue.\nThat is true, but there is no way today to exclude those pods from the metric, or separate the time. Maybe we need an extra metric, but I think the new metric could be the one that includes the time guarded in PreEnqueue. It should probably be part of the KEP. The old metric is already stable, so we can't add labels. And it might already be in use for SLOs.\ncc\nI think we should be working on that. And treat it as part of GA requirement for PodSchedulingReadiness.\nAn edge(maybe not edge) case is a pod's scheduling attempts is 0 but the is a long time, so prefer not to account for the waiting time when gated. For users not using PreEnqueue, they will not be impacted, but for users using PreEnqueue, it's a break change for it increases the duration time. And we can add a new metric like to measure how long a pod stays in a scheduler.\nI thought of \"attempts\", it's just I'm not sure it's a good idea to rely on \"attempts\" to define a SLI/SLO, b/c \"attempts\" is obtained from the in-memory object and in a HA config env, it's only updated by the primary scheduler instance. In other words, the value is inaccurate when the lease owner switches.\nattempt=0 is good enough for most scenarios\ndo we have a conclusion? Should we move on with the solution or remove this issue and the linked PR from the current release?\nI would like to move forward with this proposal. ?\nTo move it forward, I think we should add new metric(s) to explicitly tell it's including or excluding the time spent on PreEnqueue. Given is in stable level: We will need a new metric like \"podschedulingexcludingpreenqueuedurationseconds\".\n28 Bug Triage Shadow here. Just checking-in, if this is still on track for the 1.28 release?\nWe don't have a conclusion yet, but as a bug, it should qualify for changes after the freezes."}
{"_id":"q-en-kubernetes-e778adaebb7204aaccdb6b0ad852288af0cc70a46f89c26d93abbff84abe1ce5","text":"Data as of June 3rd, 9:00 PM Eastern. Latest data can be found here: Additional Context:\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig release\ncc Do we want to keep this job and fix it, or do something different considering the debs/rpms efforts?\nI think we should keep this job as long as we build packages for Google infra (those are using different specs then packages for OBS).\nfixed it:"}
{"_id":"q-en-kubernetes-fc4591d5a968b672a0570388697c9d60ee563c13cd8a5a0b6a61949383cd02ed","text":"This pertains more for clusters running in GCP but not necessarily in a GKE environment. Clusters that using a CNI supporting overlay such as Calico, do not need explicit routes to be created on the infrastructure level, meaning that the route controller should not be necessary. The following snippet sets the condition for new GCE nodes when a new node is picked up by the node controller. This condition is supposed to be removed later by the route controller in However, as stated above there are cases where the route controller is not needed and should not be running. When Calico is deployed in such a way that the calico-node pods are ignoring any and all taints including the cloudprovider's taint (which we consider reasonable for a system pod such as calico), the above check creates the following race condition: a Calico pod may start and set the node network conditions ( to false) before the node controller reconciles the node. After that the node controller picks up the node and sets the condition to an otherwise healthy node. Leaving aside the race condition itself, I have to ask whether the check is meaningful in the first place. AFAIK, the kubelet in Kubelet only sets the condition for provider, but clusters should already be using kubelet with cloudprovider= For the node controller to not set the condition unnecessarily. An easy way to do it with a custom CCM image is to introduce some delay in the node sync loop, enough for pods such as calico-node to start on that node. No response <details</details<detailsGCP </details<details</details<details</details<details</details<detailsCalico </details\nThis should be fixed by /cc\n/sig cloud-provider /sig network /assign\n/triage accepted"}
{"_id":"q-en-kubernetes-770086606bae719d268caabc123abc30e449a3085cd77438e71558b5b1142417","text":"Add the following to be safe sysctls: net.ipv4.tcpfintimeout net.ipv4.tcpkeepaliveintvl net.ipv4.tcpkeepaliveprobes These are related settings to net.ipv4.tcpkeepalivetime that was enabled in These are related ipv4 TCP settings for tuning network connections between a client and a pod.\n/sig network\nMay you help provide some tests to ensure that setting this sysctl can't affect any other pods or the node.\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : May you help provide some tests to ensure that setting this sysctl can't affect any other pods or the node. /sig nodes Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\nI adjusted it with the plugin.\nWould this change apply at the node level or the pod level? I assumed it would be at the pod level, and should not affect any other pods--however Kubernetes does not natively have any limits on network connections that I am aware of--but only the pod with the changed net settings would be affected.\nTo verify that a sysctl has an effect on other pods, we need to verify that it's a namespace sysctl.\nHere's a simple test Use net.ipv4.tcpfintimeout sysctl to create a container check if it set on the host =It is the default value on host run another container attach to the network namespace of the one where the sysctl is set =results in sysctl being set Verify in another container under a different network namespace that it is not set =It is the default value According to the test results, net.ipv4.tcpfintimeout is a namespaced sysctl\nI likewise verified and , which are also namespaced sysctl, and I believe we can add them to the safe sysctl /assign\nProbably need to wait for to finish before start working on it\n/triage accepted"}
{"_id":"q-en-kubernetes-a0f44ff49904c630a236a41ae82dec0f4b59be25538001c026bb3cf6d3eff7eb","text":"N/A 07/18 02:38 PDT No response /sig release\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nSlack thread:\n/sig cli"}
{"_id":"q-en-kubernetes-4ac47953c966d73cdb4f8db4df90c7a83ec7544295f353614ecb5b18cb1183e4","text":"/sig scheduling /kind regression caused a performance degradation in the 5k nodes scalability tests ! The PR changed how to split the load among parallel threads. Previously, we would parallelize over the number of nodes. The PR does it over the number of plugins, which is much smaller. Performance to improve or remain the same https://perf- No response 28 N/A <details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign The quickest fix is to revert, but I'm open to other suggestions.\n/priority critical-urgent\nCan we just use another Parallelizer to handle the score process for each plugin? I ran the UT locally and it seems fine, but I'm not familar with\nDepending on how many cores you have available, you might just overload the routines. I think it's better to revert the parallizer over plugins while keeping the change from map to slice. Although I don't expect a major impact for the latter. You might want to run a profiler for the benchmark with a higher number of nodes.\nI'll create a revert version first :)\nLet's revert this first since we reached the code freeze. We can do the refactor later.\nSorry for causing this trouble and thanks everyone for a prompt action"}
{"_id":"q-en-kubernetes-606c1d95a05ee07ac0fb0270728c195eba6457caa0c36194149dc09bc972c356","text":"This flakiness is reproducible on x86 and ppc64le architectures. https://prow.ppc64le- in /sig\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig testing\n/sig node /assign"}
{"_id":"q-en-kubernetes-3da1785f6bd753c45736bbd293a0a746d294fd81adc36664d7eaa0f7a74eafb0","text":"/sig scheduling Include the Pod UID in the event message for preemption by kube-scheduler Minimizes the need for an end user to consult the ops team to learn about what caused their pods to be preempted, without compromising PII. From\n/triage accepted\nCan I pick this up?\nYes, just type . Note that we are in code freeze, so this won't be merged until about a month from now.\n/assign\nPII? I know that as personally identifying information (ie, data about humans).\nBTW, ideally we'd also document the change.\nIn a multi-tenant environment, pod names could be sensitive information."}
{"_id":"q-en-kubernetes-06fbdef5e504dc3aae9c51820398a7847ef7322d8eaf0fbadc20091d1460d8a8","text":"Calculate sidecar containers resource in pod autoscaler. From 1.28 user could run Sidecar containers in , those sidecar containers remain running during the entire life of the Pod, it should be calculated in . If missing the sidecar resources, we will get a wrong pod resource utilization. We should also correct the related Cadvisor metrics.\n/assign /sig autoscaling\n/triage accepted /priority important-soon"}
{"_id":"q-en-kubernetes-042b753178b6ae4cd5699af58dadfe9c574839d637069cd42426a55d84a81866","text":"We have a pod fake function in , but this is not a base util anymore, we have a lot of arguments, we can replace this with here /sig scheduling /kind cleanup /help This is an easy one to get start. Cleaner code for maintenance.\nThis request has been marked as needing help from a contributor. Please ensure that the issue body includes answers to the following questions: Why are we solving this issue? To address this issue, are there any code changes? If there are code changes, what needs to be done in the code and what places can the assignee treat as reference points? Does this issue have zero to low barrier of entry? How can the assignee reach out to you for help? For more details on the requirements of such an issue, please see and ensure that they are met. If this request no longer meets these requirements, the label can be removed by commenting with the command. <detailsIn response to : ### What would you like to be ? /kind cleanup /help Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/remove-kind feature\nI want to try this one, the problem we solve here is to replace the *v1.Pod with type PodWrapper {}? /assign\nReplace with in\nyou mean replace ` in I check the code and didn't found out that we have makeBasePod in\nI have tried it. But there will be error about cycle import, because package import package, where locate.\nyes, I've tried it as well. guess I need to wait for kerthcet's response <img width=\"805\" alt=\"image\" src=\"\"\nSorry, forgot to reply on this, I introduced another PR to help solve this import cycle, but waits for other maintainers' response. So generally, if that pr accepted, we can work on this one then. Thanks for the volunteer.\nthe import cycle is resolved, you can start your work now, thanks!\nthanks for the ping!"}
{"_id":"q-en-kubernetes-0754ac7bcf463db71ff62e3e2bf376844b1f4020d08fde9752e5a4b08b9e97da","text":"remove and was introduced in to fix This is to overcome the situation that volume is to DSW at first and then deleted before it is to ASW, which causes mounted volume neither in DSW nor ASW and won't execute unmount/detach: The reason the volume cannot be to ASW in time is , which is . However, there is another PR also solved the issue. What it did is to check whether ASW already has the volume before deleting it from DSW, this makes sure that volume can be detached/unmounted after it is deleted from DSW. I suggest that we can keep the second solution and remove related code to make the reconstruction process and reconcile loop cleaner?\n/sig storage\ncc\nShould I proceed with this issue? Your guidance is appreciated.\nI think we'd better wait for the feedback from\ncc\nSorry for a late reply. This is how behind we are with reviews :-(. Reading Now that is on by default, it reconstructs volumes first and always puts them to ASW, it should never happen that a reconstructed volume is in DSW and not in ASW, right? So we can remove easily. I would not be so sure without , because the code is 5 years old and who knows all the implications. You could be right though, it may not be necessary there too.\n/triage accepted\n/assign"}
{"_id":"q-en-kubernetes-e5239da31475cc408e37fe1270e0f340149f1b179f79b6ece1d3aa35708ff9fc","text":"We need to explore how DRA interacts with One likely outcome is that the DRA helper code watches for nodes with the special label and helps DRA drivers clean up. /sig node\n/assign\n/triage important-soon as this is a DRA Beta criteria and DRA is going to be graduated to Beta in 1.29\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /triage important-soon as this is a DRA Beta criteria and DRA is going to be graduated to Beta in 1.29 Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/triage accepted\nHere's a test that was written for volumes:\nThank you. I’ve found it last week. Unfortunately kind image doesn’t have ssh, so stopping kubelet is not possible the way this test does. I’ve found another solution.\nI was able to simulate non-graceful node shutdown by running . It turned out that after adding tent, all node pods are deleted by the GC controller and the resource claim controller successfully deallocated the resource. So, it looks like a non-graceful shutdown is handled correctly and no code changes are required. However, the fact that can potentially be called without calling should be explained in the DRA KEP, so that DRA plugin authors will be aware of this. I'll add an explanation of handling non-graceful node shoutdown to the and add a correspondent e2e test case to the DRA e2e test suite.\nKEP update PR: E2E PR:"}
{"_id":"q-en-kubernetes-9833d436ed0d7cd87bff14a39dccbe4377f5e87cdff15de0c9b2b0d561aa663a","text":"transformation expandless | 14m1s Failed; === RUN TestEncryptionConfigHotReload TestEncryptionConfigHotReload When I see this flake, 10-09 is the first one.(But the job history in the testgrid only started from 10-03) some PR that may be related: /sig auth /cc\n/cc\n/triage accepted\n/cc failed two times in recent 3 runs:\n/reopen as of this is still .\nReopened this issue. <detailsIn response to : /reopen Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details"}
{"_id":"q-en-kubernetes-b0df5b59049720637ee6899078c8355df04a35b6393ba5d6263465e50214aa4a","text":"4 other Test Failures in ci-crio-cgroupv1-node-e2e-features. Test has failed 61 times (no passing result found). Earliest failure found at commit/ (Thu Oct 19 21:41 PDT). /cc { failed [FAILED] Error creating Pod: pods \"pod-dualstack-host-ips9ddb9810-134c-4d35-a0c6-\" is forbidden: violates PodSecurity \"restricted:latest\": host namespaces (hostNetwork=true) In [It] at: @ 10/23/23 01:38:06.024 } No response_ /sig network node\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\nis also failing on these tests\nthe test has a bug, it checks the contrary\n, let's start over, 1 commit for the e2e test, other for the promotion, and let's validate everything is alright before merging"}
{"_id":"q-en-kubernetes-550f4382eaf6f2574b193eb10497a48e4e8c7bfa87371b6f8dd7ca8be3c8d2cd","text":"we just log that the user should install etcd, which is impractical in an ephemeral container we actually already call installing etcd in the verify script, but we don't export it into the current PATH for the process so it doesn't work ... ref: /sig api-machinery /assign /kind bug\n/triage accepted\nfor one possible fix\n/sig etcd"}
{"_id":"q-en-kubernetes-d28453dfc6bcf829133f35627fac482c86b9caf8a20943cea67f030688a8c436","text":"Runs k8s with master golang: test Failure was observed on Aug 25th, it is passing after that. However it can be reproduced with stress tool. No response No response It is failing on both ppc and x86 architectures. Steps to reproduce: up golang tip version. clone kubernetes master /sig testing\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\nit is a sig node test, sig testing charter is not about doing or maintain tests, we need to remove that old mindset of having a QA department, that is not how software development works or should work :) Besides, thanks for spotting it and submittn a patch /sig node /remove-sig testing\nFlaky in\nmerging this PR will fix the issue"}
{"_id":"q-en-kubernetes-18a5f78a5b4039c2d6c090416c6a2637771f1d0c6604c73c3afd50f185f1a3f4","text":"The swap fedora job is failing consistently for OOM. For a while. It looks like OOM is not actually killing these pods when swap is enabled. I did not see any examples of this failing without swap enabled so I think this is a real issue. There is a failure in OOMKilled as part of the serial fedora job but that is not related to this. This is part of and should be required for promotion to Beta2. /sig node\n/triage accepted /priority important-longterm"}
{"_id":"q-en-kubernetes-f8cba96d7af492bbd89bb089b4aed949aa9507e872bf2bb7a99f04a32adbe03f","text":"pkg/registry/core/persistentvolume/storage TestUpdateStatus unknown No response /sig storage\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/assign\n/priority important-soon as it failed in master blocking"}
{"_id":"q-en-kubernetes-eb5589dd8109b546f3f24867d42590a81e8b93836006e7a26d7a84d50aa0ef54","text":"When unit tests in are invoked with 1, it fails Unit tests should work for any . This is important for troubleshooting flaking tests. Discovered when I tried to add parallel execution as described in master N/A N/A N/A N/A N/A\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig network /area kube-proxy\n/cc"}
{"_id":"q-en-kubernetes-b02b380dc71b93d08e0ccc7768a8b4ce44c6911102d809bff80ab3a21d026127","text":"Created a RWX volume and mounted it on a pod having readOnly flag set to true. While the volume provisioning happened successfully, we did observe some warning message on describing the pod: Warning FileSystemResizeFailed 27m kubelet MountVolume.NodeExpandVolume failed for volume \"pvc--a6ba-49f4-a560-\" requested read-only file system It seems like this message is being triggered from here: But we never did an expand volume call, so we should not be seeing warning message. It can be confusing to see this warning message. The warning message should not be displayed when expand volume was not even called. Create a RWX volume and mount it on a pod with readOnly flag set to true. Wait for a few minutes and describe the pod. No response_ 28 vSphere <details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig storage\n/assign"}
{"_id":"q-en-kubernetes-4cedb06a566ea432b70bad38aa75afb5c41b6d4b476ff1019ee7757e2034ed6b","text":"When running subcommand like this: a warning message will display: which is incorrect in mode. It is only correct in mode. Warning message to only display when flag is passed in. Run No response <details</details<details</details<details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig cli"}
{"_id":"q-en-kubernetes-6593a5f6eee2ae8487ea83fb7883fea797662788ba1018ee349fd7ca0ec07e76","text":"! Please also see /kind failing-test <!-- If this is a flake, please add: /kind flake --<!-- Please assign a SIG using: /sig SIG-NAME --/sig testing\ncc\nslack thread:\n/triage accepted\ngot one good run!"}
{"_id":"q-en-kubernetes-dbf813432d184c58d7382219162d9513652adc2416b194ce1ef0f3c6ef9ee302","text":"/kind failing-test /sig node\n/assign\n/triage accepted\nSeems to me that problem is related with test-infra config ? I noticed Checking kubelet logs there are complains about kubelet not finding configuration file. Even if this job run in AWS it is defined under ( ) and not under aws subfolder. Performing vimdiff periodic- ./aws/periodic-eks- i notice some variables are missing, but i don't have permissions to re-run the job to be able to debug more on this. Test code for InPlacePodVerticalScaling doesn't seem to be the root cause here, if we could increase the kubelet log level and perform a rerun might be helpful to debug more. Checking containerd logs , it seems that a container failed to stop, this might explain why pod resize failed due to lack of resources)\nThese two tests are permanent failing on this job\nTrue, but looking deeper it seems for the first test it fails ( unable to resize there are no available resources ) because a previous pod has not been deleted according to containerd logs . For second error it seems to be a config error , it complains that it could not login. I believe something is wrong with the setup ( repeated kubelet config errors in kubelet logs )\n/sig storage\nMore information for the container delete error, it belongs to mainpod-with-poststart-http-hookrestartable-init-container-lifecycle-hook. I will try to check if InPlacePodVerticalScaling test can be improved to cover this corner case. Please note, Garbage collect error happens before the InPlacePodVerticalScaling test failure, perhaps this is the root cause of the failure in InPlacePodVerticalScaling error. Perhaps we should update Issue title and include the garbage collect error for this issue? ! Source: / . Please search for"}
{"_id":"q-en-kubernetes-8fba26730e225229cebfcc9e9c81c6027c2c910e6c9f6867d93937be1135e10f","text":",%20google-gce all I can't say , but at least a few weeks ,%20google-gce No response No response /sig network\n/assign\n/triage accepted"}
{"_id":"q-en-kubernetes-29413a491ce4700b5090f2ccec5e3645bbc47d10dffa11e343f3d9f092c18349","text":"The following api endpoints are currently in - As each group has no stable endpoints they are reported in as promoted to stable without tests. The current logic for endpoints will address this issue.\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig architecture /area test"}
{"_id":"q-en-kubernetes-1fddb3b599751010fd024d59e4002c926ec297661adb30ed03755bb80c0f9141","text":"After restart Kubelet, node will become notReady in first kubelet update period. node condition is : in kubelet's logs: when go in to next period (default 10s), node status resume to Ready. Restart Kubelet should not make node status change to notReady. Restart Kubelet, and watch node status for second The root case maybe here: When kubelet start, it will start two go routine, one detect runtime status, one check node status and update to APIServer; detect runtime status: in it will update : in it will check runtimeStatus include : is one of the notReady setter func: So if nodestatus execute before , kubelet will update a node status with notReady condition: In kubelet starting, both go routine and will be called before , so the first node status is notReady. In our cluster is v1.22.17, but I belive it can reproduce in latest version. <details</details<details</details<details</details<details</details<details</details<details</details\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig node\nThe label(s) cannot be applied, because the repository doesn't have them. <detailsIn response to : /sig node Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI can reproduce this issue in v1.30 When the container runtime is healthy, the kubelet should not report with the reason We called in the function to update the status of the container runtime, but we did not do this in , We only need to execute function once before and , just like we did in the function, than this problem can be avoided. But in , seems to have some different opinions...\nYeah, in our env, we modify kubelet the same way as you said, to let it execute once before , it worked. I can submit a PR later.\nI may be missing something but what happens if something has changed that caused the kubelet to not be ready ... imagine, the Network plugin is not working ... during the initial period until the check changes the state pods will be scheduled on the node and will fail .... I think that we are operating the assumption that this is a restart and during that time nothing changed, but is this a safe assumption or can we guarantee somehow that nothing has changed that could impact the node readiness state?\nIn this issue, what I found is that nothing has changed, every time I restart kubelet, node will become notReady only in first sync period, in next period it will be ready; It's not an assumption. or other problem of course will cause node notReady, but they are not in this issue's scope, in this scope, notReady reason is message, have not relationship with other problem;\nok, I misread this issue sorry, so it is not to blindly set the node to ready, is just to perform the runtime check before the other checks ... I think you both are right ... actually it seems that if wins the race then there is no problem, righ ?\nYes, but and run in different goroutines, so we cannot guarantee that will complete faster. So it seems like a good choice to execute before running , or do you have better suggestions?\nWe need to put the goroutines in the right order to make the startup more predictable\nYes, if execute very fast, may there is no problem, because in it will execute . I have test in v1.28 (which code in fastStatusUpdateOnce almost same as 1.30 ) , it will not always reproduce, but still occurr some times. In 1.22, will sleep for about 100ms before everything, it can be reproduced everytimes.\nArranging the goroutines in order can solve our problems in most cases, but in the method, we need to call the containerRuntime api to get the status. I am worried that its response will not be timely enough in special scenarios, there may be a risk that will complete faster.\nmoving will execute first\nI think I must not have expressed clearly. What I'm worried about is that under special circumstances, calls the API of the container runtime and cannot return immediately (not sure if this situation really exists). Even if the method is executed first, it will still complete later than :)\nAdd some detail: in it call runtime api to check runtime status, then set variable . in , it will check variable, if is nil, this issue's problem will occurr. Even if goroutine call firstly, call container runtime check will still need some time, it can't guarantee that when firstly call, is setted.\nare you suggesting to lock syncNodeStatus on updateRuntimeUp? I'm afraid of some corner cases we can hit ... if the container runtime does not return do we block forever?\nContainer runtime API call have timeout.\nCan you try this on a newer version? 1.22 is out of support and we have made changes to node readiness since. /triage needs-information\nI have tried in 1.28, it can reproduce, can reproduce in 1.30.\n/remove-triage needs-information"}
{"_id":"q-en-kubernetes-f9f3a4a9e4c80ea389fb8a9736c195405e1e162356b7859d014b4284b5404210","text":"What happened: If you use to test whether a user can approve certificate signing requests, you get a response indicating that is not a valid verb An example of the verb being used can be seen What you expected to happen: should be recognized as a valid verb on objects. How to reproduce it (as minimally and precisely as possible): Run Anything else we need to know?: I think that needs to be to the kubectl auth can-i code Environment: Kubernetes client and server versions (use ): Cloud provider or hardware configuration: OS (e.g: ):\nThis issue is currently awaiting triage. SIG CLI takes a lead on issue triage for this repo, but any Kubernetes member can accept issues by applying the label. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nI don't think is a valid verb for kubectl auth command. Valid verbs are in\nso in that line of code there are a number of verbs which are only usuable on specific objects for example which works on users. AFAIK the point of that line is to inform kubectl what the valid verbs are on any object in Kubernetes, so that it knows what's possible when users use the command. the verb is valid on certificate objects as shown in the docs example I listed. Without being present in that line of code, will incorrectly state that there is no verb in Kubernetes.\nI tested it, seems working when we add in that list. But the question is should we add it or not?\nThere have been cases where Kubectl has had additions to handle non-standard resources/verbs in the past (e.g. I raised a couple of years back). AFAIK the explicit list of verbs here is designed to address this point, it's just not got approve in it. You could ask SIG-Auth about it, I'm sure they'd have more info.\n/transfer kubenertes /sig auth Decision can be made by sig-auth\nSomething went wrong or the destination repo kubernetes/kubenertes does not exist. <detailsIn response to : /transfer kubenertes /sig auth Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/transfer kubernetes/kubernetes\nSomething went wrong or the destination repo kubernetes/kubernetes/kubernetes does not exist. <detailsIn response to : /transfer kubernetes/kubernetes Instructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/transfer kubernetes\n/triage accepted\nThis will work even without adding verb there, adding verb just removes the warning. Submitted a PR to add that. /assign"}
{"_id":"q-en-kubernetes-d6a2cfcf65b740093bf3b8dc935076e8c61a689f6ed71e2e391888aac4bf6a68","text":"Everything 2024-05-24 No response /sig testing\n/sig node\nkubelet logs dmesg:\nSeems a regression in\n/cc\n/triage accepted /priority important-longterm"}
{"_id":"q-en-kubernetes-16e9d6242dd147113f7225a4b24c8c9581a1b180a6b05f0edc47ba4e0c49bde0","text":"master-blocking: ci-kubernetes-unit It starts flaking today 30-05-2024 The test fail is This failure may be caused by the error message's order sensitivity. The error message's order of Resource Classes versus is different. N/A /sig node /sig scheduling\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/remove-sig testing /sig scheduling (sig testing doesn't own all of the unit tests, just test frameworks, tooling, CI infra, etc) This appears to be a flake in scheduler unit tests which would be sig scheduling ( )\n.. and it appears to be DRA specific subset of scheduling so /sig node as well\nFYI - looks like this is happening every 3-6 hours give or take. Yesterday it was every six hours and today it seems to be every 3-4 hours. Looking into it a bit more, it looks like it may just be an issue with a connection via the test while it's trying to hit the server. I don't think the test itself is failing, I think the connection to the server is failing. ! CC\nHey All - hope you're well. Michael here from the Release Signal team. Just wanted to touch base and make sure we're good to go for the cut on the 4th. CC /sig-node"}
{"_id":"q-en-kubernetes-1526b3852d29761d739a37aed35c379fa23d213ea73b054963472bd46c418c30","text":"Kubelet soft admission handlers were created as a way for the Kubelet to block pods that cannot be run for reasons that can be resolved. In practice, this mechanism was only ever used by AppArmor (). Problems with soft-admission rejection: is only used by AppArmor, and only for situations that cannot realistically be resolved (host or build does not support apparmor) held in the blocked state must be manually cleaned up. While there they hold reserved capacity from the scheduler and continually retry syncpod. other features work this way - usually Kubelet hard rejects pods that it knows it cannot run, or the runtime fails to start the pod/container. This was originally to prevent controllers from continuously recreating pods that cannot run, but that is a separate problem that should be approached globally, rather than with this 1-off AppArmor solution. I propose we remove Kubelet soft admission entirely, and move the AppArmor admission handler to the hard-admission list. /sig node /milestone v1.31 /kind bug\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nWe have a different, alpha mechanism: . We could change AppArmor, seccomp and other mechanisms to work either with PodSchedulingContext or its eventual replacement. IMO we only need to sketch out an approach and see that it looks viable; we don't have to block graduation for AppArmor support on a fleshed-out design.\nFrom my admittedly limited PoV I'm not against removing soft admission, and I got no opinions about AppArmor specifically. However this seems to connected to the recent conversation we're having in WG serving and SIG-node about retriable vs unretriable errors so I think we would need some coordination cc\n+1\nMy understanding is that after this PR: soft admission just doesn't make sense any longer.\nHello ! Just a reminder, the code freeze is starting (about from now), and while there is still plenty of time, we want to ensure that each PR has a chance to be merged on time. As the issue is tagged for , is it still planned for this release?\nDiscussed in sig-node, decided to move forward with this (move AppArmor admission to the hard admission set, remove soft-admission code)\n/assign\n- do you want me to incorporate this change into"}
{"_id":"q-en-kubernetes-01221639ca87beb4d58486db98fdd31e7eab9aeb213cb02b235da5cb2047e034","text":"pull-kubernetes-unit TestStoreListResourceVersion in seems like July 7th 2024: ! Don't know, here is the log: No response /sig\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig api-machinery\nSee also\n/sig etcd\n/assign /cc\nBisected this to cc\nI'm not sure if this is the proper way to fix it, but setting a limit in the ListOptions deflakes the test. The flakiness comes from the fact that the cache is sometimes not initialized in time which causes it to return a \"too many requests\". Perhaps in tests we could wait till the cache is ready in order to proceed?\nRight, setting a limit when a resource version is set would delegate the list to etcd when cache is not initialized based on the KEP description. It is due to this type of request is not that costly to etcd. Looks like it is exactly how the wait cache ready is in See below So I think we should implement this same thing in when initializing the .\nAgree - waiting for cacher to be initialized when creating it here is a better fix for it:\nI thought about this as well, but waiting on is not possible outside of the cacher module, since it is a private field. I'm thinking to either add an option to cacherstorage.Config and make wait or add a new public method to . Do you guys have a better idea? Thanks.\nExposing Wait(context) method sounds fine\nI have pushed the changes we discussed. Thank you all for your guidance."}
{"_id":"q-en-kubernetes-e1c6f7407b346df6fb6212f23a2f930e04198e198bb03b9cff5041967802082d","text":"The PR above moved everything related to into because that was where most of the code already was, but didn't renamed the files to avoid a larger diff and merge conflicts. Once is merged, we can do the renaming. Code clarity. /sig node\n/triage accepted /priority important-soon\n/assign\na PR can be created for this now.\nwill do"}
{"_id":"q-en-kubernetes-7e3bf9f7cc9913c2c832ea7854e2bf968397de93b69e6c2711f2a32d2a4b8e55","text":"/kind failing-test <!-- If this is a flake, please add: /kind flake --<!-- Please assign a SIG using: /sig SIG-NAME --\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\nlooks like started a few days ago ! /sig cluster-lifecycle\ncc\nfix was sent earlier\n/sig cluster-lifecycle\nthanks :) i saw it yesterday and forgot to raise the issue. i should have checked the PR queue :)"}
{"_id":"q-en-kubernetes-86f72e96d548c4e81185c025a10e2d04ae6592e98dce4ac1020bbde868fad271","text":"ci-cgroupv1-containerd-node-arm64-e2e-serial-ec2-eks ci-cgroupv1-containerd-node-e2e-serial-ec2-eks It has been failing consistently for a long time. No response /sig node\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/cc\nThere's another bug that was filed a long time ago. But there's no update on it and the same test in gce jobs seem to be green now.\ncc"}
{"_id":"q-en-kubernetes-773b1fcc123536d87743dbcce61564d4d99d0ef8f1bc9e99dc4abf4640767e58","text":"ci-kubernetes-unit TestPickBestStrategy shows the jobs are success. but in test-grid and ci report, it shows a failure. After This failed the {Failed;Failed;Failed;Failed;Failed;Failed;Failed; === RUN TestPickBestStrategy === RUN TestPickBestStrategy/singlecandidate,singlepreferredstrategy === RUN TestPickBestStrategy/multiplecandidates,differentpreferredstrategiesshouldfail === RUN TestPickBestStrategy/multiplecandidates,multipleresolvedpreferredstrategy No response_ /sig testing\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/cc\nThanks for opening this Paco, just noticed.\ncc\n! It is weird as the test passed indeed and I cannot reproduce it.\nFixed in"}
{"_id":"q-en-kubernetes-503f4eb9e1101cd3b3a6307d9ade03528bfbb5ca0774f24cc21b9b3ac01d2621","text":"This issue was supposed to have fixed the issue where endpoint stay out of sync I also wrote a small script which would get me the out of sync endpoints compared to the endpointslices I expect the endpoints to eventually sync and reflect the most upto date information. I have just deployed the newer patch to our cluster and that has resulted in endpoints never ending up being updated if the status goes out of sync. No response Client Version: v1.29.7 Kustomize Version: v5.0.4-0.- Server Version: v1.29.7 <details</detailsalmalinux-9 <details</detailscri-o No response\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/sig network\nfrom the logs, endpoint controller was retrying test1/test-qa in the expected way. It has retried 5 times and the next retry should be after 320ms, which should long enough for the informer cache to get the updated endpoints. If there is no more logs related to the failure, I think it may have synced it successfully. The python script used to detect the failure seems capturing diff between endpoints and endpointslice at a moment, but that could happen if there is any Pod change in the cluster, and technically the script is comparing endpoints of one moment with endpointslice of another moment. So did you see the endpoints not being synced after endpoint controller stopped retrying it?\nYeah i still see it and see similar messages continuously. Let me post some additional log messages after the one i posted above.\nObserved the same issue in 1.29 and 1.30. I keep seeing the same error then new pods were behind a service. And the endpoint never catch up after all pods are settled.\ncould you provide complete log of kube-controller-manager and the yamls of the related endpoints and endpointslice?\n/assign Are you able to triage this since you are already looking at this?\nYes i can get that over by EOD today\nSure, I'm waiting for more logs and the content of the endpoints to understand how it happened. Currently there are only retry logs which is expected in some cases.\nWe found a customer that triggered an edge case in your change of , which will cause Endpoints never updated. They have a mutatingWebhook on Endpoints object that automatically injects a label, let's say . syncServices, the endpoints controller , and they won't match since Endpoints have the extra \"dummy-k: value\". endpoints controller will by removing the label. mutatingWebhook will automatically adds back the label as part of the update request. From the APIServer's POV, the Endpoints object don't have any changes, thus the update request will succeed but returns the same resourceVersion as before. change in will mark that resourceVersion as stale. future pods events won't be handled correctly due to the step didn't result in a new resourceVersion. I think we should fix this by check the resulting resourceVersion from the update request, and only mark the old resourceVersion as stale when a new resourceVersion is generated. Not sure whether this is the same case others in this thread encountered :D\nthanks for explaining the use case and the analysis. I did analyze the code after the issue was opened and suspected the only possibility is the returned resource version somehow equals to the one in the request but didn't know how it could happen after it has checked the desired endpoints and the current endpoints are different: Now I understand this could happen when the endpoints appear different on client side but are mutated to be same on server side. Do you plan to create a PR since you have proposed a fix? could you confirm if this is also your case, i.e. there is a mutatingWebhook for endpoints objects that could make the endpoints different from what endpoints controller expects?\nYeah, i can do a fix for this\n- can we assign to you?\nSure, I'm already assigned. Given that we've had three user reports on this issue, I think it's important to address it as soon as possible. Could you please let me know when you plan to create the PR? If you're currently occupied, I'm happy to handle it and submit the PR.\nSorry, were occupied by some internal incidents last week. Just created a WIP PR for it. Will test it in kops for this specific webhook scenario in the following two days and and remove the WIP flag\n/assign"}
{"_id":"q-en-kubernetes-a2e77bc9d5795a4a17af5bcc5860b35587b39d7e25d603a866fcb410798b82da","text":"seen in this inside a like loop we execute a function that return a bool and an error. If we want to fail fast we return the error directly and assert on the output of the wait function. If we want to retry we log the error and return false nil /kind flake /kind bug /sig testing /sig network\nThis issue is currently awaiting triage. If a SIG or subproject determines this is a relevant issue, they will accept it by applying the label and provide further guidance. The label can be by org members by writing in a comment. <detailsInstructions for interacting with me using PR comments are available . If you have questions or suggestions related to my behavior, please file an issue against the repository. </details\n/cc since I think you this test but most probable the code is inherited and is a copy paste thing, but besides that, the change is to remove those there\n/assign"}
{"_id":"q-en-kubernetes-49e05d2f2f317603fb57e106ed7f1dde07ec985be93eb80f96de35cb540b3d02","text":"The code says this (): While the comment says this (): So..., what happened? The feature is still Alpha or it is a Beta? Be consistent about the features we add or change. Read the code. No response 30.0 <details</details<details</details<details</details<details</details<details</details\nAccording to the changelog: This feature has been demoted from Beta to Alpha. That is fine. We need to revise the code comment then.\n/sig api-machinery (not sure who owns it per se)\nnot worry, i think this can be resolved with\nI found same like this inconsistency between the comment and the actual feature gate definition for APIServingWithRoutine in due to this\nI want to raise PR on this if need another PR separately\nLooks like this comment should be updated to alpha based on . Thanks for catching. if you'd like to raise a PR to fix the comment that'd be great!\n/triage accepted\nsure I will raise"}
{"_id":"q-en-kubernetes-5db59427f7480c01036171cb909e9a21e32b3fce4bc46b1955a70f229c09fbe9","text":"k8s component developers are using FQDNs as finalizers (e.g. or which triggers an error message that isn't particularly easy for an end user to understand: To a naive end user, a fully qualified domain name is a pretty safe thing that is unlikely to conflict with anything else (as long as it's only used by someone from within the domain that owns it, which is a reasonable assumption). If the finalizer has s inside it but no , the message should suggest that finalizers need a . Something like: Do whatever it is that triggered the warning in (this is probably simpler than installing cert-manager and asking for a certificate) introduced: <details</details<detailsGKE </details<details</details<details</details<details</details<details</details\n/sig api-machinery\nIf thats ok I would like to work on that.\n/assign\n/triage accepted"}
{"_id":"q-en-kubernetes-4e1e8e3df5d85e92b0a012a0f9c66d157162fa3694ce62d536459d8af744fb03","text":"[ ] two weeks soak end date : xxxx-xx-xx According to following APIsnoop query, there are six CSINode endpoints that are untested. Using a number of existing e2e test practices a new has been created to provide future Conformance coverage for the 7 endpoints. The e2e logs for this test are listed below. This query shows the following CSINode endpoints are hit within a short period of running this e2e test. If a test with these calls gets merged, test coverage will go up by 6 points This test is also created with the goal of conformance promotion. /sig testing /sig architecture /area conformance\n/triage accepted"}