Bug 137892

Summary: Patch is spinning EWS for no obvious reason
Product: WebKit Reporter: Alexey Proskuryakov <ap>
Component: Tools / TestsAssignee: Jake Nielsen <jake.nielsen.webkit>
Status: RESOLVED FIXED    
Severity: Normal CC: ddkilzer, jake.nielsen.webkit, rniwa
Priority: P2    
Version: 528+ (Nightly build)   
Hardware: Unspecified   
OS: Unspecified   

Description Alexey Proskuryakov 2014-10-20 13:05:36 PDT
This keeps being retried by EWS, not sure why. It should be able to determine that the patch is a failure.

https://webkit-queues.appspot.com/patch/239920
Comment 1 Jake Nielsen 2014-10-20 13:41:47 PDT
Looks like this is the culprit:

(patchanalysistask.py:208)

if (not first_results.did_exceed_test_failure_limit() and
    not second_results.did_exceed_test_failure_limit() and
    self._results_failed_different_tests(first_results, second_results)):
    # We could report flaky tests here, but we would need to be careful
    # to use similar checks to ExpectedFailures._can_trust_results
    # to make sure we don't report constant failures as flakes when
    # we happen to hit the --exit-after-N-failures limit.
    # See https://bugs.webkit.org/show_bug.cgi?id=51272
    return False

Neither exceeds the failure limit, and they do fail different tests, so it defers. It should probably report the flakiness that it can, and fall through.

In general I do think that's just a band-aid though, this whole function needs some stream-lining. It's difficult to reason about because it's got so many special cases that get filtered out one-by-one. I think we can do better.
Comment 2 Alexey Proskuryakov 2014-10-20 14:39:10 PDT
But were these different tests? Looks like these runs failed the same tests:

https://webkit-queues.appspot.com/results/5528053639806976
https://webkit-queues.appspot.com/results/4704366959263744
Comment 3 Jake Nielsen 2014-10-31 11:47:09 PDT
Fixed in a different bug:
https://bugs.webkit.org/show_bug.cgi?id=138184