Jupyter Collaboration prevents JupyterLab from being killed from the terminal

[martinRenou]

Issue

Installing jupyter_collaboration results in not being able to kill JupyterLab from the terminal.

Reproduce

1. Install jupyterlab and jupyter_collaboration
2. Create a Notebook, so that you create at least one room
3. Try to kill jupyterlab with Ctrl + C in the terminal

Attempts to fix the issue

After some investigation, it looks like this sys.exit call does not work as expected even though it properly raises a SystemExit.

One reason for sys.exit to not work can be a hanging thread or subprocess that has not been terminated.

My investigations so far have been:

• Removed the _confirm_exit thread from jupyter_server to see if that thread did not prevent from finishing
• Refactored the tasks cleanup in ypy-websockets and jupyter_collaboration
  • Refactor room and server tasks cleanup y-crdt/ypy-websocket#76
  • Refactor tasks cleanup #160
• Remove any AnyIO usage in ypy-websockets. I tried this because there seems to be an "AnyIO worker thread" thread hanging around upon termination.

None of these attempts fixed the issue. Though I think the refactor of the tasks cleanup could be good to include anyway.

[martinRenou]

I'm out of ideas for now. I feel like the "AnyIO worker thread" could be the culprit, but I don't know who creates it.

[davidbrochart]

One idea: BaseException catches all built-in exceptions, while Exception catches all built-in, non-system-exiting exceptions. I think KeyboardInterrupt is a system-exiting exception.
Maybe we catch BaseException instead of Exception somewhere, preventing it to be propagated up the stack?

[jordanhboxer]

are there any updates on this? after installing jupyter collab recently i'm running into this.

[davidbrochart]

Maybe #514 can improve the situation. Otherwise, have you tried jupyverse?

[jordanhboxer]

thanks for pointing me there! unfortunately it's too aggressive of a clean up and jupyverse is a no go considering how big of a shift that would be... however i came up with a solution that seems to work!

i monkey patched YDocExtension.stop_extension to iterate ywebsocket_server.rooms then iterate list(room.clients) which i can then iterate the _background_tasks and cancel those. I'm gonna hammer this a bit more but I can look into putting up a diff at some point

[jordanhboxer]

it turns out that approach wasn't as aggressive as I wanted it to be so I ended up monkey patching the ServerApp._cleanup method instead which seems to work like a charm..

def _patch_shutdown_cleanup() -> None:
    import asyncio

    from jupyter_server.serverapp import ServerApp

    _original = ServerApp._cleanup

    async def _patched(self) -> None:
        await _original(self)
        loop = asyncio.get_event_loop()
        current = asyncio.current_task()
        pending = [
            t for t in asyncio.all_tasks(loop) if not t.done() and t is not current
        ]
        if pending:
            for task in pending:
                task.cancel()
            await asyncio.gather(*pending, return_exceptions=True)

    ServerApp._cleanup = _patched

you can suppress noisy logging with a logging filter if you'd like too