New issue
Advanced search Search tips

Issue 702417 link

Starred by 1 user

Issue metadata

Status: Duplicate
Merged: issue 702268
Owner:
Closed: Mar 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

1294% regression in time-to-play media.tough_video_cases around 456705:456986

Project Member Reported by chcunningham@google.com, Mar 16 2017

Issue description

Spike in time to play for two tests on win8.

Similar spike in  Issue 702332  (also time-to-play on win), but the ranges don't overlap (though they are close).
 
All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=702417

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg9ITc_AgM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg9Je0rAkM


Bot(s) for this bug's original alert(s):

chromium-rel-win8-dual
Project Member

Comment 3 by 42576172...@developer.gserviceaccount.com, Mar 17 2017

Cc: sugoi@chromium.org
Owner: sugoi@chromium.org

=== Auto-CCing suspected CL author sugoi@chromium.org ===

Hi sugoi@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : sugoi
  Commit : cf232fd17d6a4137620296afebef13eac0d39f18
  Date   : Tue Mar 14 21:24:41 2017
  Subject: SwiftShader DEPS update

Bisect Details
  Configuration: win_8_perf_bisect
  Benchmark    : media.tough_video_cases
  Metric       : time_to_play/tulip2.m4a
  Change       : 1.89% | 94.9889285714 -> 86.4941666667

Revision             Result                   N
chromium@456834      94.9889 +- 94.8262       14      good
chromium@456837      87.4379 +- 1.83898       14      good
chromium@456838      86.6208 +- 0.633223      6       bad       <--
chromium@456839      86.77 +- 3.2565          14      bad
chromium@456844      86.5467 +- 1.16204       6       bad
chromium@456853      86.7133 +- 1.23298       6       bad
chromium@456872      87.6279 +- 9.37582       14      bad
chromium@456910      86.455 +- 1.5516         9       bad
chromium@456986      86.4942 +- 2.25769       6       bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests media.tough_video_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984924113811559904

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6442930077696000


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!

Comment 4 by sugoi@chromium.org, Mar 20 2017

Cc: capn@chromium.org
This bisect looks like it found the wrong culprit. I started another bisect from 
456834 to 456837.
Cc: crouleau@chromium.org
Owner: chcunningham@chromium.org
Assigning this to chcunningham@ to find the cause of the regression and assign appropriately.
Why is the initial bisect result suspected wrong?
The original change is from ~86 ms to ~95 ms, but The bisect identified a difference from ~86 ms to ~87 ms as the culprit. Where are the other 7 ms?
We just need to wait and see what the bisect comes back with.
Project Member

Comment 11 by 42576172...@developer.gserviceaccount.com, Mar 21 2017


=== BISECT JOB RESULTS ===
NO Perf regression found

Bisect Details
  Configuration: mac_10_12_perf_bisect
  Benchmark    : media.tough_video_cases
  Metric       : vm_resident_set_size_delta_size/crowd.wav_total

Revision             Result                  N
chromium@456834      7122.86 +- 5317.83      21      good
chromium@456837      6782.1 +- 5326.01       21      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests media.tough_video_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984559461009706992

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5552382349410304


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Whoops, I bisected the wrong thing. I'm trying it again. 
Project Member

Comment 14 by 42576172...@developer.gserviceaccount.com, Mar 21 2017


=== BISECT JOB RESULTS ===
NO Perf regression found

Bisect Details
  Configuration: win_8_perf_bisect
  Benchmark    : media.tough_video_cases
  Metric       : time_to_play/video.html?src_tulip2.m4a_type_audio

Revision             Result                  N
chromium@456834      404.45 +- 2290.69       21      good
chromium@456837      475.418 +- 2815.14      21      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=video.html.src.tulip2.m4a.type.audio media.tough_video_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984494230468833968

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5244048660496384


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Summary: 1294% regression in time-to-play media.tough_video_cases around 456705:456986 (was: 1294% regression in time-to-play media.tough_video_cases at 456835:456986)
A few observations

1. The mac alerts are all now ignored (matching ref graph spike). The windows alerts remain.

2. The "all graphs for this alert" link is busted here (or I'm missing something) - currently shows no alerts. So to see the windows alerts, use this link:
https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg9ITc_AgM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg9Je0rAkM

3. The range in the title is misleading (I'm changing it now). The windows graphs both show the alert at rev 456986. The last good value on these graphs comes from rev 456705. 

4. Given the above, I think this is a duplicate of another time-to-play windows regression in this range:  Issue 702332 . The bisects in that issue found the following CL to be at fault:

Author : sugoi
Moving SwiftShader from component to bundled library
Review-Url: https://codereview.chromium.org/2715563002
Cr-Commit-Position: refs/heads/master@{#456764}

Will attempt another bisect with this range in mind, but I think we have a good case for resolving as duplicate either way. Caleb, yell if you disagree.
> The "all graphs for this alert" link is busted 

Now fixed. Turns out I had ignored the alerts by accident
 Issue 702416  has been merged into this issue.
Project Member

Comment 21 by 42576172...@developer.gserviceaccount.com, Mar 21 2017

Mergedinto: 702268
Status: Duplicate (was: Untriaged)

=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : sugoi
  Commit : 842efc6922b0cd66febd6f36f04318ebf396c184
  Date   : Tue Mar 14 18:10:39 2017
  Subject: Moving SwiftShader from component to bundled library

Bisect Details
  Configuration: win_8_perf_bisect
  Benchmark    : media.tough_video_cases
  Metric       : time_to_play/video.html?src_tulip2.m4a_type_audio
  Change       : 220.69% | 26.9433333333 -> 86.4033333333

Revision             Result                  N
chromium@456705      26.9433 +- 53.838       6      good
chromium@456741      20.1328 +- 65.922       9      good
chromium@456759      71.0656 +- 509.95       9      good
chromium@456762      14.4733 +- 20.6233      6      good
chromium@456763      14.0408 +- 16.2005      6      good
chromium@456764      87.2483 +- 7.10498      9      bad       <--
chromium@456768      88.1006 +- 13.9772      9      bad
chromium@456776      86.84 +- 2.72597        9      bad
chromium@456846      85.5433 +- 2.04148      6      bad
chromium@456986      86.4033 +- 8.43244      6      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=video.html.src.tulip2.m4a.type.audio media.tough_video_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984486515702711312

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5823099909439488


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Project Member

Comment 22 by 42576172...@developer.gserviceaccount.com, Mar 22 2017


=== BISECT JOB RESULTS ===
NO Perf regression found

Bisect Details
  Configuration: win_8_perf_bisect
  Benchmark    : media.tough_video_cases
  Metric       : time_to_play/tulip2.m4a

Revision             Result                  N
chromium@456834      87.5381 +- 3.2447       21      good
chromium@456837      88.9705 +- 21.7783      21      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests media.tough_video_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984483875981838624

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5864198115950592


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Project Member

Comment 23 by 42576172...@developer.gserviceaccount.com, Mar 22 2017


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : sugoi
  Commit : 842efc6922b0cd66febd6f36f04318ebf396c184
  Date   : Tue Mar 14 18:10:39 2017
  Subject: Moving SwiftShader from component to bundled library

Bisect Details
  Configuration: win_8_perf_bisect
  Benchmark    : media.tough_video_cases
  Metric       : time_to_play/tulip2.m4a
  Change       : 716.18% | 10.6033333333 -> 86.5425

Revision             Result                   N
chromium@456705      10.6033 +- 1.13317       6      good
chromium@456741      10.4025 +- 0.448874      6      good
chromium@456759      11.5325 +- 2.0725        6      good
chromium@456762      11.0883 +- 1.51759       6      good
chromium@456763      10.8492 +- 0.975664      6      good
chromium@456764      87.6842 +- 1.29733       6      bad       <--
chromium@456768      87.9933 +- 1.5058        6      bad
chromium@456776      87.8367 +- 1.17547       6      bad
chromium@456846      86.1633 +- 1.67282       6      bad
chromium@456986      86.5425 +- 1.73066       6      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests media.tough_video_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984486497372455376

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5856786948554752


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Cc: chcunningham@chromium.org
Owner: sugoi@chromium.org
I was wrong. sugoi@'s commit seems to be the culprit.
turned out to be a different sugoi@ commit, so you were *technically* correct ;)

Comment 26 by sugoi@chromium.org, Mar 22 2017

Still looking into it. It looks like the GPU process is getting created for no reason in these tests. I need to prevent that from happening without breaking anything, which is the tricky part :)
Project Member

Comment 27 by bugdroid1@chromium.org, Mar 25 2017

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/7ae2ba1892a8b4dee8d7c231ac0e0fbe66588912

commit 7ae2ba1892a8b4dee8d7c231ac0e0fbe66588912
Author: sugoi <sugoi@chromium.org>
Date: Sat Mar 25 13:59:47 2017

Solving some telemetry tests' slowness

After investigation, it turns out that most telemetry test perfomance
regressions were linked to an observer indefinitely waiting for gpu
info. This comes from SystemInfoHandler::GetInfo(), where, when
SwiftShader was being used,
GpuDataManager::IsEssentialGpuInfoAvailable() would always return false,
so the gpu info collection would think it needs to set a GpuObserver in
order to wait for the info to be loaded. By having
GpuDataManager::IsEssentialGpuInfoAvailable() return true when
SwiftShader is used, the gpu info is returned immediately and this
solves the hang.

BUG= 702417 

Review-Url: https://codereview.chromium.org/2770933008
Cr-Commit-Position: refs/heads/master@{#459651}

[modify] https://crrev.com/7ae2ba1892a8b4dee8d7c231ac0e0fbe66588912/content/browser/gpu/gpu_data_manager_impl_private.cc

Project Member

Comment 30 by 42576172...@developer.gserviceaccount.com, Apr 11 2017

Cc: jbudorick@chromium.org
Owner: jbudorick@chromium.org

=== Auto-CCing suspected CL author jbudorick@chromium.org ===

Hi jbudorick@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : jbudorick
  Commit : 761f60864c37e7af6fe9dba49256a8b014860049
  Date   : Tue Mar 14 21:20:28 2017
  Subject: [android] Clean up test_runner.py arguments.

Bisect Details
  Configuration: win_8_perf_bisect
  Benchmark    : media.tough_video_cases
  Metric       : time_to_play/video.html?src_tulip2.m4a_type_audio
  Change       : 65.12% | 1410.28277778 -> 491.912222222

Revision             Result                  N
chromium@456834      1410.28 +- 1868.72      9      good
chromium@456835      88.46 +- 1.00591        6      bad       <--
chromium@456836      89.85 +- 5.24778        6      bad
chromium@456837      491.912 +- 1219.0       9      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=video.html.src.tulip2.m4a.type.audio media.tough_video_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8982640125828895344

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5244048660496384


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!

Sign in to add a comment