Fix asyncio hanging issue when using crew.kickoff() with asyncio.to_thread()

[devin-ai-integration]

Fix asyncio hanging issue when using crew.kickoff() with asyncio.to_thread()

Summary

Fixes issue #3730 where crews would hang indefinitely when run with asyncio.to_thread(crew.kickoff) if they used async tools.

Root Cause: The tools code was calling asyncio.run() to execute async tool functions. When a crew is run via asyncio.to_thread(), it executes in a thread spawned by an existing event loop. Calling asyncio.run() in this context fails because it tries to create a new event loop in a thread that's conceptually part of an already-running loop, causing a hang.

Solution: Added a run_coroutine_sync() utility that:

• Uses asyncio.run() when no event loop is running (normal case)
• When a loop IS running, spawns a new thread with its own event loop to run the coroutine

Changes:

• Added src/crewai/utilities/asyncio_utils.py with run_coroutine_sync() helper
• Updated CrewStructuredTool.invoke() to use the new helper (2 call sites)
• Updated BaseTool.run() to use the new helper (1 call site)
• Added comprehensive test suite in tests/test_asyncio_tools.py

Review & Testing Checklist for Human

⚠️ Risk Level: MEDIUM-HIGH - This modifies a core execution path affecting all async tool usage

• Review the threading approach in run_coroutine_sync() - Is spawning a new thread with a new event loop the right solution? Consider:

  • Performance implications (creates a thread for every async tool call when in nested async context)
  • Potential thread safety issues
  • Resource leak risks (event loop cleanup)
  • Whether there's a better alternative (e.g., using asyncio.get_event_loop().run_until_complete() or other approaches)

• Test the actual issue scenario end-to-end - The new tests mock Agent.execute_task, so they don't verify full integration. Test manually:

  # Run this with a real LLM configured
  import asyncio
  from crewai import Agent, Crew, Task
  from crewai.tools import tool
  
  @tool
  async def test_tool(x: str) -> str:
      """Test async tool"""
      await asyncio.sleep(0.1)
      return f"Result: {x}"
  
  crew = Crew(
      agents=[Agent(role="Test", goal="Test", backstory="Test")],
      tasks=[Task(description="Use test_tool", expected_output="Result", tools=[test_tool])]
  )
  
  async def run():
      result = await asyncio.to_thread(crew.kickoff)
      print(result)
  
  asyncio.run(run())

  Verify it completes without hanging.

• Verify the lock file regeneration didn't break anything - The uv.lock was completely regenerated (not part of core fix). Run the full test suite to ensure no regressions.

• Check exception handling - Verify that exceptions raised in async tools are properly propagated with full tracebacks through the new threading approach.

Notes

• Issue reported by user with local Ollama setup experiencing hangs when running crew in background with asyncio.to_thread()
• The existing kickoff_async() method already worked correctly; this fixes the manual asyncio.to_thread(crew.kickoff) pattern
• All new tests pass, and existing thread safety tests pass

Link to Devin run: https://app.devin.ai/sessions/fa61a1f74deb410faa495932f4a86cad
Requested by: João ([email protected])

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring