non-deterministic test_channel test failure

[apteryks]

Hi,

On Guix, building using Python 3.11.4 and the following dependencies:

python-hatch-vcs@0.5.0 python-hatchling@1.27.0 python-pytest-bootstrap@8.4.1

I encountered:

FAILED testing/test_channel.py::TestChannelBasicBehaviour::test_channel_passing_over_channel[main_thread_only-popen] - execnet.gateway_base.RemoteError: Traceback (most recent call last):
  File "/gnu/store/n9wcfy3gx8gk3dd2s4xgc5119ylykcha-python-execnet-2.1.1/lib/python3.11/site-packages/execnet/gateway_base.py", line 1291, in executetask
    exec(co, loc)
  File "<remote exec>", line 2, in <module>
AssertionError
= 1 failed, 465 passed, 614 skipped, 4 xfailed, 11 xpassed, 1 warning in 59.96s =

This was under high load, and it seems difficult to retrigger that same failure (I've just built the package/ran the test suite for 5 times without the problem being triggered).

[RonnyPfannschmidt]

It seems like a potential cleanup timing issue

Its not critical and can be fixed by repeating thr check

[apteryks]

I've disable the test in Guix for now, thanks for the assessment.

[sanvila]

Hello. On AWS instances of type r7a.medium and m7a.medium, which incidentally have a single CPU, I'm getting the failure reported here 100% of the time when trying to build the Debian package (which includes the tests). So, in this case, the failure is completely deterministic, and it may not be fixed by repeating the check.

If this is a bug in the test and not a bug in the code, I'd like to suggest that the test is disabled for everybody.

To reproduce the failure, try configuring GRUB with GRUB_CMDLINE_LINUX="nr_cpus=1".

Thanks.