[Python SDK] graceful shutdown does not track or cancel in-flight tasks

[santoshkumarradha]

Summary

agentfield.agent_server.AgentServer._graceful_shutdown does not track in-flight asyncio tasks, so when shutdown fires, long-running reasoner tasks continue executing in the background past the shutdown deadline. The agent process exits while work is still in progress — partial state, potential dropped responses, and a confusing audit trail in the control plane.

Where

• File: sdk/python/agentfield/agent_server.py
• Function: AgentServer._graceful_shutdown
• Discovered by: sdk/python/tests/test_agent_graceful_shutdown.py::test_graceful_shutdown_cancels_in_flight_tasks_within_deadline (skipped with pytest.skip("source bug: graceful shutdown does not track or cancel in-flight tasks"))

Reproduction

agent = make_shutdown_agent()
server = AgentServer(agent)

async def long_running():
    await asyncio.sleep(60)

tasks = [asyncio.create_task(long_running()) for _ in range(5)]
await server._graceful_shutdown(timeout_seconds=0)

# Expected: all tasks done (completed or cancelled) before _graceful_shutdown returns
# Actual:   tasks still running after _graceful_shutdown — server exits leaving them orphaned

Expected behavior

_graceful_shutdown should:

1. Maintain a registry of in-flight reasoner tasks (created when a request arrives, removed when it completes)
2. On shutdown, await all pending tasks up to the configured deadline
3. After the deadline expires, cancel any remaining tasks (see [Python SDK] graceful shutdown does not enforce timeout-based task cancellation #357)
4. Only call os._exit after the task drain is complete

Reference implementation: most ASGI servers (uvicorn, hypercorn) follow this pattern — track requests, drain on shutdown, cancel on deadline.

Acceptance criteria

• AgentServer maintains a set of in-flight reasoner tasks
• _graceful_shutdown awaits all in-flight tasks before exit
• Tasks finished before the deadline complete cleanly
• The skipped test in test_agent_graceful_shutdown.py::test_graceful_shutdown_cancels_in_flight_tasks_within_deadline is unskipped and passes

Related

• Companion: [Python SDK] graceful shutdown does not enforce timeout-based task cancellation #357 (graceful shutdown does not enforce timeout-based task cancellation)
• Coordinates with: [Python SDK] Agent.stop() is not implemented #355 (Agent.stop() not implemented)

Discovered via

PR #352 (test coverage improvements). One of 5 source bugs surfaced while writing failure-mode tests for the Python SDK.

[Luffy2208]

@santoshkumarradha i would like to work on this

[github-actions]

Thanks for picking this up, @Luffy2208! 🎉

Here's what to expect:

• We'll check in after 7 days if we haven't heard an update
• After 14 days, we'll send a second reminder
• After 21 days without activity, we'll unassign to let others contribute

Tips for success:

• Drop a comment when you start working on it
• Share any blockers - we're here to help!
• Link your PR with Fixes #356 in the description

Happy coding! 🚀

[santoshkumarradha]

@Luffy2208 🚀