Try to debug non-closed iopub socket

[Carreau]

We sometime get

 FAILED tests/test_connect.py::test_port_bind_failure_gives_up_retries - pytest.PytestUnraisableExceptionWarning: Exception ignored in: <function Socket.__del__ at 0x7efe22e4d760>
Traceback (most recent call last):
File "/opt/hostedtoolcache/Python/3.12.9/x64/lib/python3.12/site-packages/zmq/sugar/socket.py", line 185, in __del__
warn(
ResourceWarning: Unclosed socket <zmq.asyncio.Socket(zmq.PUB) at 0x7efe1eb32820><F48>

This adds a tracemalloc fixture for the crashing test module in order to try to narrow that down. I'm thorn between restart the tests here until we have the failure; or merge and see once we get the error in another PR and then remove the fixture once we've fund the culprit.

---

The only zmq.PUB I can find is iopub, and we don't seem to be closing it

In addition : should app be a contextmanager ?
And should the atexit close registring be replaced with a weakref finalizer ?

[Carreau]

Ah ! No, in the real application the iopub_socket is closed when you close the thread, but in conftest it is not !

[Carreau]

well, in conf.py it is also closed...

[Carreau]

@minrk do you believe that using tracemalloc in the del in pyzmq to print where the socket was created could help and would you accept a PR ?

[minrk]

@Carreau 100%! I thought you had to -X tracemalloc / PYTHONTRACEMALLOC=1 to get this output from ResourceWarnings. Can we run the tests with that?

[Carreau]

@Carreau 100%! I thought you had to -X tracemalloc / PYTHONTRACEMALLOC=1 to get this output from ResourceWarnings. Can we run the tests with that?

I think you need both ? My understanding is that -X tracemalloc / PYTHONTRACEMALLOC=1 just enable tracemalloc, but that you still need to call tracemalloc.get_object_traceback(obj) in the code that emit the warnings, and tracemalloc.get_object_traceback(obj) ?

Maybe I'm wrong.

[Carreau]

Seem that you are right it shoudl show up automatically. I will try differently.

[Carreau]

Thanks for the review. I'm going to merge and hope it triggers a more helpful error message soon.