bpo-46711: increase timeout for test_logging::test_post_fork_child_no_deadlock

[sobolevn]

Closes #31205
Thank you, @notarealdeveloper for noticing it!

However, I am not sure how to proceed here:

• We can merge bpo-46679: Don't ignore timeout argument in test.support.wait_process. #31205 first, then I rebase this PR (this way the original author will get a credit for fixing the problem)
• Or we can just merge this one (it is easier)

https://bugs.python.org/issue46711

[vstinner]

Can you please change the default timeout in wait_process() instead?

[sobolevn]

I think it might have some consequences 🤔

Right now SHORT_TIMEOUT is just 30 seconds, while LONG_TIMEOUT is 5 minutes.
It is a 10x increase.

Secondly, current docs state (https://docs.python.org/3/library/test.html#test.support.SHORT_TIMEOUT):

If a test using SHORT_TIMEOUT starts to fail randomly on slow buildbots, use LONG_TIMEOUT instead.

Initially I've followed this recommendation.

Do we have other tests that fail due to this timeout? Is it a global problem? You totally have more data than me on this problem.

Lastly, regrtest has --timeout, which we can change for slower runners.

If you think that we still should change the default, I have several questions:

1. I guess we need to add .. versionchanged:: 3.11 to wait_process docs and change the default value there
2. In this case we would have to recommend using SHORT_TIMEOUT if users expect some test to be fast, am I right? Or should we just remove this from the docs completely?
3. Should we change --timeout= flags that are used in the project somehow? https://cs.github.com/python/cpython?q=--timeout%3D

[vstinner]

Oh. I forgot that I doubled the ASAN buildbot last week: python/buildmaster-config@5a37411

    # Sometimes test_multiprocessing_fork times out after 15 minutes
    test_timeout = TEST_TIMEOUT * 2

Lib/test/libregrtest/setup.py adapts SHORT_TIMEOUT to --timeout option:

        # For a slow buildbot worker, increase SHORT_TIMEOUT and LONG_TIMEOUT
        support.SHORT_TIMEOUT = max(support.SHORT_TIMEOUT, ns.timeout / 40)

So SHORT_TIMEOUT should now be 60 seconds rather than 30 seconds. Why did https://buildbot.python.org/all/#/builders/621/builds/466 fail at 52.5 seconds seconds in this case?

Maybe my computation is wrong, or this build is older than the buildbot configuration change?

[vstinner]

cc @pablogsal

[pablogsal]

I'm ok not changing the default. I'm convinced by @sobolevn comments 👍

[pablogsal]

So SHORT_TIMEOUT should now be 60 seconds rather than 30 seconds. Why did https://buildbot.python.org/all/#/builders/621/builds/466 fail at 52.5 seconds seconds in this case?

I have no idea. I checked and the builder should have picked the change you talk about but maybe I have missed something :(

[sobolevn]

So SHORT_TIMEOUT should now be 60 seconds rather than 30 seconds. Why did https://buildbot.python.org/all/#/builders/621/builds/466 fail at 52.5 seconds seconds in this case?

I've added the information about the timeout to the error message. It can probably help the next time we have the same problem.

[notarealdeveloper]

Closes #31205 Thank you, @notarealdeveloper for noticing it!

However, I am not sure how to proceed here:

• We can merge [bpo-46679: Don't ignore timeout argument in test.support.wait_process. bpo-46679: Don't ignore timeout argument in test.support.wait_process. #31205](bpo-46679: Don't ignore timeout argument in test.support.wait_process. #31205) first, then I rebase this PR (this way the original author will get a credit for fixing the problem)
• Or we can just merge this one (it is easier)

https://bugs.python.org/issue46711

@sobolevn No worries! I don't mind not getting credit. :) This was just a minor thing I noticed while working on #30310.

[vstinner]

The Lib/test/support/init.py change is nice. I'm no longer sure that the Lib/test/test_logging.py change is correct. Maybe revert test_logging.py change?