The Python kernel's control thread may hang during shutdowns

[seeM]

The Python kernel's control thread may hang during shutdowns. See this Playwright report for example.

The last few logs before the hang:

[Python] [PositronIPKernelApp] Control received: {'header': {'msg_id': 'defdcdb49e', 'msg_type': 'shutdown_request', 'session': 'python-0d121134', 'username': 'runner', 'date': datetime.datetime(2025, 4, 4, 21, 16, 46, 731000, tzinfo=tzutc()), 'version': '5.3'}, 'msg_id': 'defdcdb49e', 'msg_type': 'shutdown_request', 'parent_header': {}, 'metadata': {}, 'content': {'restart': False}, 'buffers': []}
[Python] [positron.positron_ipkernel] Shutting down the kernel
[Python] [positron.help] Stopping pydoc server thread
[Python] [positron.help] Pydoc server thread stopped
[Python] [PositronIPKernelApp] Kernel has no children.
[Python] [PositronIPKernelApp] {'header': {'msg_id': '4ff340cc-0dac95f225aa852099896c76_155264_59', 'msg_type': 'shutdown_reply', 'username': 'runner', 'session': '4ff340cc-0dac95f225aa852099896c76', 'date': datetime.datetime(2025, 4, 4, 21, 16, 47, 7045, tzinfo=datetime.timezone.utc), 'version': '5.3'}, 'msg_id': '4ff340cc-0dac95f225aa852099896c76_155264_59', 'msg_type': 'shutdown_reply', 'parent_header': {'msg_id': 'defdcdb49e', 'msg_type': 'shutdown_request', 'session': 'python-0d121134', 'username': 'runner', 'date': datetime.datetime(2025, 4, 4, 21, 16, 46, 731000, tzinfo=tzutc()), 'version': '5.3'}, 'content': {'status': 'ok', 'restart': False}, 'metadata': {}, 'tracker': <zmq.sugar.tracker.MessageTracker object at 0x7fe8024fbd90>}
[Python] [PositronIPKernelApp] Stopping control ioloop
[Python] [PositronIPKernelApp] Stopping shell ioloop
[Python] [asyncio] Close <_UnixSelectorEventLoop running=False closed=False debug=True>
[Python] [PositronIPKernelApp] Cleaning up sockets
[Python] [PositronIPKernelApp] Closing heartbeat channel
[Python] [PositronIPKernelApp] Closing iopub channel
[Python] [PositronIPKernelApp] Closing control thread

[seeM]

This may be addressed in ipykernel once v7 comes out of pre-release. There was a huge refactor to the way that concurrency is handled in ipython/ipykernel#1079.

[seeM]

I haven't been successful tracking this down, but I suspect it is a pre-existing issue in ipykernel. Some options below:

Option A: The supervisor currently waits for a kernel process to exit before it considers a kernel exited. We could update it to wait only for the shutdown_reply message. If the process hangs around afterwards, Positron could kill it in the background and log it somewhere.

Option B: More simply, we could timeout if the control thread doesn't exit quickly (or more generally, if the shutdown sequence takes longer than expected) and forcefully exit, perhaps with an error status code.

[jonvanausdeln]

We haven't seen this issue in automation since the fix. We'll keep an eye on it.