Problem/Motivation
In #3350972: [random test failure] Drupal\Tests\layout_builder\FunctionalJavascript\LayoutBuilderUiTest::testReloadWithNoSections() @nod_ introduced a "fix" for problems with the correct opening of an off-canvas dialog which potentially could fix all JS random test failures for tests using the drupal.dialog.off_canvas somehow.
The basic test is:
- Unskip
\Drupal\Tests\layout_builder\FunctionalJavascript\InlineBlockTest::testDeletion. - Run _only_ now unskipped
\Drupal\Tests\layout_builder\FunctionalJavascript\InlineBlockTest::testDeletiona lot of times (usually we go for 1500x) as-is. - Run _only_ now unskipped
\Drupal\Tests\layout_builder\FunctionalJavascript\InlineBlockTest::testDeletiona lot of times (usually we go for 1500x) _without_ the changes in #3350972: [random test failure] Drupal\Tests\layout_builder\FunctionalJavascript\LayoutBuilderUiTest::testReloadWithNoSections(). - If 2., passes and 3. doesn't, we can safely turn the test back on again.
Per @xjm in #3353085-4: [meta] Determine impact of [#3350972] fix in off-canvas.js on currently disabled FunctionalJavascript tests we now need to go for a 5000-8000 times run for both patches.
Steps to reproduce
Proposed resolution
Remaining tasks
User interface changes
API changes
Data model changes
Release notes snippet
| Comment | File | Size | Author |
|---|---|---|---|
| #6 | 3353100-6.patch | 811 bytes | spokje |
Comments
Comment #2
spokjeComment #3
spokjeTest seems too slow for 1500x, let's try 1000x.
Comment #4
spokjeRight...
Let's do what I've should have done from the start, test _only_
\Drupal\Tests\layout_builder\FunctionalJavascript\InlineBlockTest::testDeletionComment #5
spokjeComment #6
spokjeVery minimal difference, only one fail in the should_fail, but it is with
which is the exact failure that was fixed in #3350972: [random test failure] Drupal\Tests\layout_builder\FunctionalJavascript\LayoutBuilderUiTest::testReloadWithNoSections(), so I believe this one to be fixed as well.
Comment #7
catchAmazing.
Comment #8
spokjeComment #9
spokjeComment #10
spokjeComment #11
spokjeBoth should_fail and non_fail patch ran 5 times. That's 7500 individual tests per patch.
Back to RTBC.
Comment #12
catchI've just committed #3353096: [random test failure] Try to un-skip and fix InlineBlockPrivateFilesTest::testPrivateFiles() in context of [#3353085], will aim to commit this one (or another similar RTBC one) in roughly 24 hours.
Comment #13
xjmOur fail patch is not failing here. Queueing more jobs to see if it's just unlucky or a lower fail rate. If it continues to pass we should double-check that we're fixing the right thing, or look to external factors like the AWS instances.
Comment #14
xjmBlergh, fixing attribution.
Comment #15
xjmOne of the should-fail-jobs in #5 now shows the expected fail, so we are good there. See what I mean about needing more test runs. :)
Comment #16
spokjeTo be fair, it did already fail: https://www.drupal.org/pift-ci-job/2639266, but more tests are better tests.
Comment #17
xjmBased on the current fail rate, even 24,000 jobs (16 test runs queued) would still have a 1% chance of a false pass, so queueing even more. I think this is definitely one where the race condition is worse depending on external factors like the AWS instance or such (which is also why it's important to queue the pass- and fail-patch test jobs at the same time on the same environments).
Comment #19
catchPatch no longer applied for some reason, but I just made the change manually. Committed/pushed to 10.1.x, thanks!
Moving to 10.0.x for backport.
Comment #20
spokjePatch applies to 10.0.x, 9.5.x and 9.4.x
Comment #21
smustgrave commentedSeems several testing rounds happened in #6
Comment #22
catchUsing 'to be ported' as 'need to cherry-pick this once all the related patches are in 10.1' so it's easier to tell which have already been committed vs. not.
Comment #25
catchCherry-picked to 10.0.x and 9.5.x, thanks!