New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 884587 link

Starred by 2 users

Issue metadata

Status: Verified
Owner:
Closed: Sep 21
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Chrome
Pri: 2
Type: Bug-Regression



Sign in to add a comment

video_ChromeHWDecodeUsed.h264 / vp9 regressed between CrOS 11067-11071.

Project Member Reported by deanliao@chromium.org, Sep 17

Issue description

https://cros-goldeneye.corp.google.com/chromeos/healthmonitoring/testDetails?milestone=71&daysBack=30&testName=video_ChromeHWDecodeUsed.h264.mse

https://cros-goldeneye.corp.google.com/chromeos/healthmonitoring/testDetails?milestone=71&daysBack=30&testName=video_ChromeHWDecodeUsed.vp9.mse

Mostly x86_64 families are affected in video_ChromeHWDecodeUsed.h264 and some x86_64 boards are affected in video_ChromeHWDecodeUsed.vp9.

Regression range from CrOS 11067.0.0 to 11071.0.0. All became green starting from 11072.0.0.

The error is that the test failed to observe histogram difference on histograms/Media.GpuVideoDecoderInitializeStatus . (0 status count didn't increase for about ten retries.)

Finding culprit CL.

noted that from 11071.0.0 to 11072.0.0 (bad to good), Chrome version are the same. So culprit is unlikely in Chrome repository.
 
CrOS changes between 11071.0.0 and 11072.0.0 (bad to good):
https://crosland.corp.google.com/log/11071.0.0..11072.0.0

ARC++ changes between 11071.0.0 and 11072.0.0 (bad to good):
https://android-build.googleplex.com/builds/5013055/branches/git_master-arc-dev/targets/cheets_arm-user/cls?end=5012931
Suspect cras issue. Verifying.
I'm curious. Did you bisect manually or use bisect tools?
Cc: dgreid@chromium.org hychao@chromium.org
Labels: OS-Chrome
Status: Fixed (was: Assigned)
https://chromium-review.googlesource.com/c/chromiumos/third_party/adhd/+/1180782 was the culprit and https://chromium-review.googlesource.com/c/chromiumos/third_party/adhd/+/1226926 fixed the regression.

While I set up bisect tool env, I ran test against 11071 and 11072 and found cras has crash issue on 11071. So I manually looking into cras related issues to do manual bisect.

It took long to confirm the above finding as it took more than 30 minutes to cros flash to a lab machine and I got only fragment time bisecting during vacation. I would see if bisect tool can work well on this case.
Cc: khmel@chromium.org agawronska@chromium.org sammiequon@chromium.org deanliao@chromium.org hiroh@chromium.org
 Issue 891415  has been merged into this issue.
Cc: kcwu@chromium.org
Wu-cheng, I tried using bisect-kit to find culprit CL and found some issue. Asking Kuang-che to investigate: crbug.com/893441
Cc: -agawronska@chromium.org -khmel@chromium.org -sammiequon@chromium.org
Remove some cc who were mistakenly added because I set the wrong duplicate issue.
According to the healthmonitor dashboard, the regression is between 11066.0.0 and 11067.0.0 and it's not flaky. However, I don't know why my local build 11066.0.0 failed with the same error message as well.

Sign in to add a comment