Make pytest unit tests more robust

[fcollonval]

Currently tests are flaky

They pass when run individually but when running them all it always fails.

[krassowski]

It looks like we may need to add some cleanup sequence which ensures that kernel workers get stopped. It looks like the idea is for them to shut down when kernel does but it is not yet implemented. Currently they just keep running until they receive CancelledError:

jupyter-server-nbmodel/jupyter_server_nbmodel/handlers.py

Lines 259 to 284 in ef8181c

	async def _kernel_worker(
	kernel_id: str,
	client: jupyter_client.asynchronous.client.AsyncKernelClient,
	ydoc: jupyter_server_ydoc.app.YDocExtension | None,
	queue: asyncio.Queue,
	results: dict,
	pending_input: PendingInput,
	) -> None:
	"""Process execution request in order for a kernel."""
	get_logger().debug(f"Starting worker to process execution requests of kernel {kernel_id}…")
	to_raise = None
	while True:
	try:
	uid, snippet, metadata = await queue.get()
	get_logger().debug(f"Processing execution request {uid} for kernel {kernel_id}…")
	get_logger().debug("%s %s %s", uid, snippet, metadata)
	client.session.session = uid
	# FIXME
	# client.session.username = username
	results[uid] = await _execute_snippet(
	client, ydoc, snippet, metadata, partial(_stdin_hook, kernel_id, uid, pending_input)
	)
	queue.task_done()
	get_logger().debug(f"Execution request {uid} processed for kernel {kernel_id}.")
	except (asyncio.CancelledError, KeyboardInterrupt, RuntimeError) as e:

but it seems like in tests teardown the read from the queue will happen beforehand in:

jupyter-server-nbmodel/jupyter_server_nbmodel/handlers.py

Line 272 in ef8181c

uid, snippet, metadata = await queue.get()

Resulting in:

=================== 1 failed, 4 passed, 1 warning in 14.57s ====================
Deleting active ExecutionStack. Be sure to call `await ExecutionStack.dispose()`.
/home/runner/work/jupyter-server-nbmodel/jupyter-server-nbmodel/jupyter_server_nbmodel/handlers.py:340: RuntimeWarning: coroutine 'ExecutionStack.dispose' was never awaited
  self.dispose()
RuntimeWarning: Enable tracemalloc to get the object allocation traceback
Could not destroy zmq context for <jupyter_client.asynchronous.client.AsyncKernelClient object at 0x7f063efd5df0>
Task was destroyed but it is pending!
task: <Task pending name='Task-124' coro=<_kernel_worker() running at /home/runner/work/jupyter-server-nbmodel/jupyter-server-nbmodel/jupyter_server_nbmodel/handlers.py:272> wait_for=<Future pending cb=[Task.task_wakeup()]>>
Exception ignored in: <coroutine object _kernel_worker at 0x7f063ceb5fe0>
Traceback (most recent call last):
  File "/home/runner/work/jupyter-server-nbmodel/jupyter-server-nbmodel/jupyter_server_nbmodel/handlers.py", line 301, in _kernel_worker
    raise to_raise
  File "/home/runner/work/jupyter-server-nbmodel/jupyter-server-nbmodel/jupyter_server_nbmodel/handlers.py", line 272, in _kernel_worker
    uid, snippet, metadata = await queue.get()
                             ^^^^^^^^^^^^^^^^^
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/asyncio/queues.py", line 160, in get
    getter.cancel()  # Just in case getter is not done yet.
    ^^^^^^^^^^^^^^^
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/asyncio/base_events.py", line 794, in call_soon
    self._check_closed()
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/asyncio/base_events.py", line 540, in _check_closed

I guess we could implement something like putting a special value on the queue (or just cancelling the worker but I could not get it to work) when the kernel is shut down?