gh-123828: fix data raceing for _waiting_release method in _interpchannelsmodule.c

[Zheaoli]

Fix #123828

• Issue: TSAN race in _waiting_release #123828

[ZeroIntensity]

@picnixz out of the scope of this PR, but would it be feasible to migrate things using PyThread locks over to PyMutex in the future?

[picnixz]

Maybe. Haven't had the time to think about it and I don't know if it's possible for every case. Could you open another issue for that? also how would performances / maintenance cost be impacted?

[ZeroIntensity]

Maybe. Haven't had the time to think about it and I don't know if it's possible for every case. Could you open another issue for that? also how would performances / maintenance cost be impacted?

It's probably worth discussing on discourse before opening an issue. AFAIK, PyMutex is supposed to be much more efficient than pthread mutexes.

[Zheaoli]

@picnixz May I ask if this pr is worth to merge?

[picnixz]

I am not well-versed enough in this part of the code to be sure that it's correct, so I'll let Eric decide (or @ZeroIntensity)

[ZeroIntensity]

The core change looks correct, I noticed this with Eric at the sprint. Please add a test though.

[Zheaoli]

The core change looks correct, I noticed this with Eric at the sprint. Please add a test though.

Sorry about this. I have noticed that seems We don't have tests for the data race before(correct me if I'm wrong). You mean that we enable some test in CI?

[ZeroIntensity]

Just add something that stresses this concurrently, we'll get around to enabling TSan on test_interpreters later.

[ZeroIntensity]

I think we should remove this news entry. _interpchannels isn't exposed to any user-facing API.

[Zheaoli]

Just add something that stresses this concurrently, we'll get around to enabling TSan on test_interpreters later.

This issue is triggered by TSAN test. So I think if we don't need add more test if we are planning to enable TSAN

[picnixz]

I think @ZeroIntensity wanted add a test that specifically triggers it, not one that indirectly triggers it.

[Zheaoli]

@ZeroIntensity @picnixz Actually I think we don't need new tests anymore. In #129824, we found that the current test will triger the TSAN report. And this PR can fix the test and I reenable the test now

[ZeroIntensity]

Ok, thanks, this looks good.

@ericsnowcurrently This is the race I was telling you about at the sprints. As far as I can tell, this is the only thing blocking test_interpreters from running on the free-threaded build.

[Zheaoli]

The test has been failed https://buildbot.python.org/#/builders/75/builds/3415

I'm trying to figure it out

[ZeroIntensity]

That might be the leak from #139073.

[ericsnowcurrently]

Any reason not to backport this to 3.14?

[ZeroIntensity]

I didn't think it was that big of a deal because _interpchannels is only used in tests at the moment, but I don't mind backporting it. We'll have to wait until 3.14.1 because 3.14 is currently locked down for the RC, though.

[miss-islington-app]

Thanks @Zheaoli for the PR, and @ZeroIntensity for merging it 🌮🎉.. I'm working now to backport this PR to: 3.14.
🐍🍒⛏🤖

[bedevere-app]

GH-139517 is a backport of this pull request to the 3.14 branch.