With the addition of buildbot masters now supporting "Slave initiated shutdown" (e.g. graceful) in a way that does not require us to curl the buildbot WebStatus page, we should make use of this. For the "Why should we do this": * Allows us to stop killing the buildslave when we see an error.flg * Allows us to use slavealloc for enabled/disabled/what-master state for mobile devices. * Allows us to run verify.py *before* the job, eliminating many developer/sheriff-viewable issues. * Allows us to improve end-to-end time by reducing the run-by-buildbot verify.py checks. This is meant to be a pretty standalone improvement whose only improvement is side-effect of running verify.py outside of buildbot after every job, before the next buildbot run starts.
Created attachment 753853 [details] [diff] [review] [device buildbot.tac] manual diff To be manually applied to our hundreds of devices. Yea sadface. (r? to catlee since he wrote this upstream patch and backported it to our end to begin with)
Assignee: nobody → bugspam.Callek
Status: NEW → ASSIGNED
Attachment #753853 - Flags: review?(catlee)
Created attachment 753858 [details] [diff] [review] [buildbotcustom] graceful the slave This actual buildbot patch is untested, but the concept of touching a shutdown.stamp after a job has started shutting down the job if buildbot was started with the change to buildbot.tac has been. I will test this in staging before deploying this one.
Attachment #753858 - Flags: review?(catlee)
(In reply to Justin Wood (:Callek) from comment #2) > Created attachment 753858 [details] [diff] [review] > [buildbotcustom] graceful the slave > > I will test this in staging before deploying this one. Tested and looks good, now just awaiting catlee's review.
Live in production.
This is now live, and seems to be holding steady with no ill effects.
Status: ASSIGNED → RESOLVED
Last Resolved: 5 years ago
Resolution: --- → FIXED
Product: mozilla.org → Release Engineering
You need to log in before you can comment on or make changes to this bug.