FR: pre-cq-launcher should respect DetectRelevantChanges updates from pre-cq slaves |
||||||||
Issue descriptionThis pre-cq run failed: https://luci-milo.appspot.com/buildbot/chromiumos.tryserver/pre_cq/48055 It failed because cheets_keyboardtest is flaky (Issue 653048), not because of the CL under test. Looking at the DetectRelevantChanges stage, the pre-cq correctly identified that the CL wasn't relevant. However it still marked it as failing. It would nice if in that case the pre-cq let the CL go through. That's what the CQ does already I think?
,
Aug 21 2017
Good point. Currently pre-cq-launcher only looks at the success/failure of pre-cq slave when deciding if a CL should be promoted to CQ. pre-cq-launcher can be updated to do as cq-master does, and only reject changes that are relevant to the failed pre-cq slave.
,
Aug 21 2017
,
Aug 21 2017
It's possible that there's a bug on TOT on that build, so it makes sense to not reject the CL if the CL is not relevant to the failed pre-cqs. My concern is if we don't reject the CL, pre-cq-launcher will keep retrying the pre-cq; and if the bug on TOT doesn't get resolved in time, the pool will be overwhelmed by this build.
,
Aug 22 2017
I think nxia is qualified to make a decision on the best course of action.
,
Aug 22 2017
,
Mar 30 2018
,
Mar 30 2018
,
May 31 2018
This will be a good improvement to reduce false rejection in pre-cq. Passing this to CI team. |
||||||||
►
Sign in to add a comment |
||||||||
Comment 1 by kirtika@google.com
, Aug 11 2017