potiuk commented on PR #28248:
URL: https://github.com/apache/airflow/pull/28248#issuecomment-1344392785

   Feel free to try :). 
   
   Though I think it could be a different behaviour (the test is not mine - 
just read the commit message it came with: 
https://github.com/apache/airflow/commit/f75dd7ae6e755dad328ba6f3fd462ade194dab25
  ) I belive the reason was that it could be triggered by a quick succession of 
those celery calls one-by-one and if we wrap it with flaky, it might simply 
never trigger. @yuqian90 - maybe you still remember this commit and the case 
and can shed some light on it
   
   In this case the whole idea of this test is not the best - I understand it 
was a way to make more certaintly that we do not trigger the case, but in 
essence this test is "flaky by design" when something happens so we might as 
well ignore it in the future (which we did by adding it to quarantined tests) - 
without actually knowing if the problem is still there or not.
   
   I believe though - that the problem is generally solved by the change, and 
maybe we do not even need to keep the test and can simply remove it (because 
regression is a) not likely b) we might not even notice the regression because 
of the flakey nature of the test). So it might be more distraction than help.
   
   @Taragolis :
   
   Yes, Using flaky might be a good idea in general. The problem with 
quarantined tests though is not only them being flaky - but also them having 
side-effects on other tests. Most of the tests in the "quarantined" group that 
we have had this nasty property, that they not only failed from time to time, 
but their sheer presence in the "main" group of tests caused the other tests to 
be also affected (that's why we invented the "quarantine" idea) to keep them 
isolated from the main pytest run.  (and yest it was at the beginnig of COVID 
so the name is inspired by real-life events :D). So "flaky" library is a bit 
different concept.
   
   Bu yes - I would love to get rid of the quarantined tests (while leaving the 
possibility of quarantining some tests if we find them breaking things again). 
I guess we could attempt (one-by-one ideally) to fix those tests and possibly 
move the quarantined tests to be "flaky" ones and see if that works. 


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to