Details

August 31, 2011 @ 9am PDT / 12pm EDT

dial-in x92 304

Backchannel #buildfaster on irc.mozilla.org

Agenda

Will Lachance has made some awesome build charts which show what's happening when for each commit. Need to discuss findings. Possible action items:
- Split mochitest-browser-chrome out from mochitest-other, to bring all test chunks into the 30-min target range. Currently mochitest-other takes 60-90 minutes on debug builds.
- Investigate what to do about long builds that are caused by full hg clones and full builds on clobbered machines.
Review the Action Items & Bugs

[DONE] will to follow up with catlee regarding dashboard calculations - new calculations are used in the GoFaster dashboard
[DONE] jgriffin to change test run comparison routine to compare against a range of recent m-c runs instead of one run to help mitigate the fact that test durations vary considerably per run, even when equivalent
[DONE] jgriffin to get data on infrequently failing tests, and how long they take to run
- for mochitest-plain, 95% of tests have never failed in the past 6 months
- list of tests that haven't failed in past six months
- list of tests that have failed in past six months
[DONE] ted to file bug about harnesses downloading symbols on demand - bug 679759
Any p2 bugs need to escalate?

we've put investigation of running debug tests with opt xpcshell on hold, since opt debug builds seem likely to materialize sometime soon (bug 669953)
still need a mechanism to track individual slow tests; jmaher suggested the correct metric to do this is test duration/test asserts; need to update the logparser to handle this

catlee to add builder name to the .csv that's being used to generate the buildcharts so we can filter out re-triggers and nightlies, etc
wlach to update buildcharts to indicate job submission time, so it's easier to see wait times
catlee and jgriffin to coordinate re: gofaster session at the all-hands