We should consider running them at the end of new-run-webkit-tests without other processes running so they're less flaky. Note that I already disabled them for debug builds because they're really flaky in debug. I don't have time for this now, but someone else might!
Created attachment 139795 [details] patch Perf tests are affected by external factors (load, i/o, etc), makes sense to lock other threads when running them. This patch solves the problem on my machine of perf tests flaking/timeouting when building a Debug version of EFL port. IMO we have such a few perf tests that it is OK to put them on the locked shard, should not affect the total elapsed time of running layout tests.
Comment on attachment 139795 [details] patch Seems fine to me. I'll give dirk a chance to review the code as well. I'd also be curious to know how this impacts total running time, although I suspect it's relatively small.
Comment on attachment 139795 [details] patch looks fine.
Comment on attachment 139795 [details] patch Clearing flags on attachment: 139795 Committed r116021: <http://trac.webkit.org/changeset/116021>
All reviewed patches have been landed. Closing bug.
Got no flakes on perf anymore on the EFL buildbot and no change on the run-webkit-tests total time.
It caused a regression, could you check and fix it, please? See https://bugs.webkit.org/show_bug.cgi?id=86800 for details.
(In reply to comment #7) > It caused a regression, could you check and fix it, please? > See https://bugs.webkit.org/show_bug.cgi?id=86800 for details. Thanks for informing. Working on it right now.