Page MenuHomePhabricator

Jenkins tests shouldn't go red when it's not its fault
Closed, DeclinedPublic

Description

The MobileFrontend tests (in Firefox) have been flakey for some time. Take a look at the list on the left for examples of recurring issues.

Example:
https://integration.wikimedia.org/ci/view/Mobile/job/browsertests-MobileFrontend-en.m.wikipedia.beta.wmflabs.org-linux-firefox-sauce/333/

I propose that when a test fails and the content of the Stacktrace contains one of
*Timeout::Error (Timeout::Error)

  • Net::ReadTimeout
  • 500 error (Our servers are currently experiencing a technical...)

the test should be marked as unreliable and should not be considered failure.

Could the test go amber when this happens instead? It would allow for better visual scanning when things break.

If there are 2 timeout errors but one actual error the test should stay red as there is a genuine error there.


Version: wmf-deployment
Severity: normal

Details

Reference
bz72722

Event Timeline

bzimport raised the priority of this task from to Needs Triage.Nov 22 2014, 3:47 AM
bzimport set Reference to bz72722.
bzimport added a subscriber: Unknown Object (MLST).

Sorry Jon, a big part of the problem is https://bugzilla.wikimedia.org/show_bug.cgi?id=72366

which should get fixed soonish

Is this still happening? I do not remember seeing it since we moved to wikimedia jenkins.

zeljkofilipin set Security to None.
greg claimed this task.
greg added subscribers: zeljkofilipin, greg.

Is this still happening? I do not remember seeing it since we moved to wikimedia jenkins.

Closing this for now but let's keep open the idea of better handling of flaky tests.

See also: