Large regression across all media metrics |
||||
Issue descriptionThe chome AV perf sherrif reported approximately 30 alerts today, across all tracked metrics (https://chromeperf.appspot.com/alerts?sheriff=Chromium%20Perf%20AV%20Sheriff&sortby=end_revision&sortdirection=down). I have opened bugs for each of the metrics, grouping when it seemed reasonable. Here is the first finished bisect (unattached to a bug unfortunately, but this was for the vm_working_set_delta_size/video.html?src_crowd1080.mp4 metric on win10x64) https://chromeperf.appspot.com/buildbucket_job_status/8986466217566509040 Linking to https://build.chromium.org/p/tryserver.chromium.perf/builders/winx64_10_perf_bisect/builds/899 Which found https://codereview.chromium.org/2693423005 as the potential culprit. None of the other bisects have completed ATM. I have tagged all of the bugs with the "media_regression_2_27" (query here: https://bugs.chromium.org/p/chromium/issues/list?can=2&q=label%3Amedia_regression_2_27) samans@, do you think your CL could have affected video performance?
,
Feb 28 2017
It's not immediately obvious to me how I could have caused this, but I won't rule it out because we have had problems with CompositorFrameSinkSupport in the past (and present).
,
Feb 28 2017
I'm confused by the dependencies. These are android specific which doesn't even use DelegatedFrameHost (Android has its own DelegatedFrameHostAndroid).
,
Feb 28 2017
Yeah, I actually didn't notice the Android stuff at first. If this is something that equally affects Android, then probably it's not my fault. It's probably best to wait for the Android bisect jobs to finish.
,
Feb 28 2017
I am following up offline to see why the Android bisect jobs have not run yet.
,
Mar 3 2017
None of the other bisect jobs has finished yet except for one which actually pointed to another CL. I'm going to assume this is not my fault.
,
Mar 3 2017
@samans, I have been following up with the infra team and the android perf bots have only been back online as of yesterday (and one of them had a faulty device which caused it to fail fast, and it "swallowed" and failed most of the queue). I agree that the Android regression are not related to your CL, and that this 'umbrella bug' is not your fault. I am following up with you in the bug that tracks the regression that the perf bots have correlated to your CL.
,
Mar 26 2017
Ping tguilbert@. The Android perf bots should be back online by now.
,
May 13 2017
All of the associated bugs have been closed. Some were fixed, some trended back towards normal results and some were too flaky. Closing this umbrella bug. |
||||
►
Sign in to add a comment |
||||
Comment 1 by samans@chromium.org
, Feb 28 2017