[Webkit-unassigned] [Bug 247843] [GLIB] Deadlock in Layout, API and WebDriver tests in debug bots

bugzilla-daemon at webkit.org bugzilla-daemon at webkit.org
Mon Dec 12 07:46:55 PST 2022


https://bugs.webkit.org/show_bug.cgi?id=247843

--- Comment #8 from Lauro Moura <lmoura at igalia.com> ---
(In reply to Carlos Alberto Lopez Perez from comment #6)
> (In reply to Lauro Moura from comment #5)
> > After some manual testing on the debug external bot:
> > 
> > - This issue is not debug-specific, also happens to release builds on the
> > external host and started happening on the WPE release tester too, which is
> > a standalone machine.
> > - Bisecting consistently shows this issue starting in move to the FDO-SDK
> > 22.08 (255085 at main)
> > - The SDK upgrade bumped Python from 3.9.9 to 3.10.6, which might be related.
> > 
> > Could not reproduce yet on my desktop or with a subset of the tests other
> > than the full suite.
> 
> I landed this the other day...
> 
> [GTK][WPE][run-webkit-tests] layout tests are deadlocking when a test
> crashes due to unhandled BrokenPipeError
> https://bugs.webkit.org/show_bug.cgi?id=248533
> 
> 
> What I found is that when running the layout tests with one worker (export
> NUMBER_OF_PROCESSORS=1) it helps to debug this issues, because with one
> worker you will see the python exception that need to be fixed, otherwise
> with several workers the exception is hidden and you get a deadlock.

This commit indeed fixed the 1 worker case, but I'm still getting the #c1 backtrace whenever I run with n>1 workers under heavy load (managed to reproduce locally with 12 child processes in my 8-core laptop).

Changing the version inside flatpak[1], I could reliably reproduce the issue beginning with python 3.9.13 and 3.10.5.

[1] https://gist.github.com/lauromoura/5159c8428048d7d592b2562a142f3cf2

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20221212/3220a3a2/attachment.htm>


More information about the webkit-unassigned mailing list