fix: add retry/trycatch logic

src/agent/runner.py

@@ -5,6 +5,7 @@ from dataclasses import asdict
 from typing import Dict, Optional
 import uuid
 
+from agent.utils import with_retries
 from agent.image_agent import generate_image_prompt
 from agent.tools import generate_scene_image
 
@@ -36,7 +37,7 @@ async def process_step(
         assert choice_text, "choice_text is required"
         graph_state.choice_text = choice_text
 
-    final_state = await llm_game_graph.ainvoke(asdict(graph_state))
+    final_state = with_retries(await llm_game_graph.ainvoke(asdict(graph_state)))
 
     user_state: UserState = await get_user_state(user_hash)
     response: Dict = {}

src/audio/audio_generator.py

@@ -94,8 +94,11 @@ async def cleanup_music_session(user_hash: str):
     if user_hash in sessions:
         logger.info(f"Cleaning up music session for user hash {user_hash}")
         session = sessions[user_hash]["session"]
-        await asyncio.wait_for(session.stop(), settings.request_timeout)
-        await asyncio.wait_for(session.close(), settings.request_timeout)
+        try:
+            await asyncio.wait_for(session.stop(), settings.request_timeout)
+            await asyncio.wait_for(session.close(), settings.request_timeout)
+        except Exception as e:
+            logger.error(f"Error stopping music session for user hash {user_hash}: {e}")
         del sessions[user_hash]