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

Issue 670413 link

Starred by 1 user

Issue metadata

Status: WontFix
Owner: ----
Closed: Jul 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

14.8%-18% regression in blink_perf.layout at 433544:434478

Project Member Reported by lanwei@chromium.org, Dec 1 2016

Issue description

See the link to graphs below.
 
Cc: nedngu...@google.com
Owner: nedngu...@google.com

=== PERF REGRESSION ===


=== Auto-CCing suspected CL author nednguyen@google.com ===

Hi nednguyen@google.com, the bisect results pointed to your CL, please take a look at the
results.


===== BISECT JOB RESULTS =====
Status: completed


===== SUSPECTED CL(s) =====
Subject : Revert of [Telemetry] Disable system_health.memory_desktop on win10 (patchset #1 id:1 of https://codereview.chromium.org/2520423005/ )
Author  : nednguyen
Commit description:
  
Reason for revert:
System health benchmark should be disabled at story level, not the whole benchmark level (if possible).

Original issue's description:
> [Telemetry] Disable system_health.memory_desktop on win10
>
> It crashes on Windows Zen Book, and Windows High DPI: Both win10 configurations.
>
> BUG= 667941 
> CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.perf:linux_perf_cq;master.tryserver.chromium.perf:mac_retina_perf_cq;master.tryserver.chromium.perf:winx64_10_perf_cq
>
> Committed: https://crrev.com/a3e6a48f576f8130788f731c8d8e6c7525118edf
> Cr-Commit-Position: refs/heads/master@{#434132}

TBR=eakuefner@chromium.org,fmeawad@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG= 667941 

Review-Url: https://codereview.chromium.org/2522273002
Cr-Commit-Position: refs/heads/master@{#434200}
Commit  : c1fa4049b9bf140f14ae31b2bd6bdac5b81df4ad
Date    : Wed Nov 23 18:15:51 2016


===== TESTED REVISIONS =====
Revision         Mean     Std Dev  N   Good?
chromium@433559  188.591  20.8672  25  good
chromium@433933  188.261  18.4484  25  good
chromium@434120  188.537  14.4354  25  good
chromium@434167  194.556  34.2936  25  good
chromium@434191  199.981  18.536   25  good
chromium@434197  197.479  35.2223  25  good
chromium@434199  202.627  22.9335  25  good
chromium@434200  210.176  17.8501  25  bad    <--
chromium@434203  209.4    33.0845  25  bad
chromium@434214  205.973  26.2939  25  bad
chromium@434307  197.949  19.7967  25  bad

Bisect job ran on: linux_perf_bisect
Bug ID: 670413

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests blink_perf.shadow_dom
Test Metric: ChangingSelect/ChangingSelect
Relative Change: 4.96%

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/linux_perf_bisect/builds/6870
Job details: https://chromeperf.appspot.com/buildbucket_job_status/8994448634165693440


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5794597152751616

| 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 Tests>AutoBisect.  Thank you!
Cc: dtu@chromium.org simonhatch@chromium.org
+simonhatch, dtu: see #3, bisect is not correct. I am not sure blink_perf.shadow_dom is bisectable, though. I'll bisect blink_perf.layout.

=== PERF REGRESSION ===


=== Auto-CCing suspected CL author chromeos-commit-bot@chromium.org ===

Hi chromeos-commit-bot@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


===== BISECT JOB RESULTS =====
Status: completed


===== SUSPECTED CL(s) =====
Subject : Automated Commit: Committing new LKGM version 9011.0.0 for chromeos.
Author  : chromeos-commit-bot
Commit description:
  
Cr-Commit-Position: refs/heads/master@{#433814}
Commit  : 1e66e96c50f124c0615cc29c0152dcd45ec9c2a5
Date    : Tue Nov 22 08:10:37 2016


===== TESTED REVISIONS =====
Revision         Mean     Std Dev  N   Good?
chromium@433543  1823.6   38.5743  25  good
chromium@433740  1810.16  155.647  25  good
chromium@433789  1785.26  109.428  25  good
chromium@433802  1814.62  107.753  25  good
chromium@433808  1816.42  120.244  25  good
chromium@433811  1796.47  102.837  25  good
chromium@433813  1803.3   134.274  25  good
chromium@433814  1751.53  116.322  25  bad    <--
chromium@433838  1757.39  133.052  25  bad
chromium@433936  1780.58  99.6244  25  bad
chromium@434329  1761.62  109.298  25  bad

Bisect job ran on: linux_perf_bisect
Bug ID: 670413

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests blink_perf.layout
Test Metric: flexbox-row-wrap/flexbox-row-wrap
Relative Change: 3.40%

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/linux_perf_bisect/builds/6872
Job details: https://chromeperf.appspot.com/buildbucket_job_status/8994439870743314640


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=4566148727701504

| 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 Tests>AutoBisect.  Thank you!
Cc: bashi@chromium.org
Owner: bashi@chromium.org

=== PERF REGRESSION ===


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

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


===== BISECT JOB RESULTS =====
Status: completed


===== SUSPECTED CL(s) =====
Subject : memory coordinator: Set variation parameters if available
Author  : bashi
Commit description:
  
In V0 implementation we have several parameters to determine
the global state. Override these parameters when variations
are available.

BUG= 617492 

Review-Url: https://codereview.chromium.org/2518653002
Cr-Commit-Position: refs/heads/master@{#433987}
Commit  : b935609873e0305e0666b89c4cff2c2a89901bf9
Date    : Tue Nov 22 21:39:05 2016


===== TESTED REVISIONS =====
Revision         Mean     Std Dev  N   Good?
chromium@433543  3585.84  284.937  25  good
chromium@433936  3639.97  320.653  25  good
chromium@433986  3718.3   117.454  25  good
chromium@433987  3605.03  326.494  25  bad    <--
chromium@433988  3664.15  230.062  25  bad
chromium@433990  3628.75  144.384  25  bad
chromium@433993  3606.66  147.739  25  bad
chromium@433999  3636.73  187.576  25  bad
chromium@434011  3624.01  449.316  25  bad
chromium@434035  3550.68  183.991  25  bad
chromium@434133  3496.6   521.125  25  bad
chromium@434329  3383.74  420.878  25  bad

Bisect job ran on: linux_perf_bisect
Bug ID: 670413

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests blink_perf.layout
Test Metric: flexbox-column-nowrap/flexbox-column-nowrap
Relative Change: 5.64%

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/linux_perf_bisect/builds/6871
Job details: https://chromeperf.appspot.com/buildbucket_job_status/8994439891535893920


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5814275350724608

| 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 Tests>AutoBisect.  Thank you!

Comment 9 by bashi@chromium.org, Dec 2 2016

Owner: ----
I don't think my CL caused the regression. The CL changed the code which run only when --enabled-feature=MemoryCoordinator.
Owner: dtu@chromium.org
re: #c4

From what I can see, this is what happens in bisect 3:

The reference value are established at roughly 188 (GOOD) and 197 (BAD)
chromium@433559  188.591  20.8672  25  good
... other commits
chromium@434307  197.949  19.7967  25  bad

Couple revisions are tested with values very close to GOOD:
chromium@433559  188.591  20.8672  25  good <-- lkgr
chromium@433933  188.261  18.4484  25  good <--- STEP
chromium@434120  188.537  14.4354  25  good <--- STEP
... other commits
chromium@434307  197.949  19.7967  25  bad  <-- fkbr


Then we get a set of values on chromium@434214 that gives a mean far outside either the good or bad set. Compare Samples gives a REJECT to both, and in revision_state.py:_check_revision_good() when this happens, we declare it either good or bad depending on the distance of the mean of the set to the means of the last known good and first known bad (lkgr/fkbr).
chromium@433559  188.591  20.8672  25  good
chromium@433933  188.261  18.4484  25  good
chromium@434120  188.537  14.4354  25  good <-- lkgr
... other commits
chromium@434214  205.973  26.2939  25  bad <--- STEP
chromium@434307  197.949  19.7967  25  bad  <-- fkbr


Similar thing happens on chromium@434167, Compare Samples gives REJECT on both, and GOOD was closer this time vs the BAD revision established in previous iteration.
chromium@433559  188.591  20.8672  25  good
chromium@433933  188.261  18.4484  25  good
chromium@434120  188.537  14.4354  25  good <-- lkgr
chromium@434167  194.556  34.2936  25  good <--- STEP
... other commits
chromium@434214  205.973  26.2939  25  bad  <-- fkbr
chromium@434307  197.949  19.7967  25  bad


Same thing happens on chromium@434191:
chromium@433559  188.591  20.8672  25  good
chromium@433933  188.261  18.4484  25  good
chromium@434120  188.537  14.4354  25  good
chromium@434167  194.556  34.2936  25  good <-- lkgr
chromium@434191  199.981  18.536   25  good <--- STEP
... other commits
chromium@434214  205.973  26.2939  25  bad  <-- fkbr
chromium@434307  197.949  19.7967  25  bad


Maybe Dave has some ideas on how to handle situations like this?
Cc: robert...@chromium.org
Handling of multiple regressions perhaps?
Project Member

Comment 14 by 42576172...@developer.gserviceaccount.com, Dec 14 2016

Cc: thakis@chromium.org
Owner: thakis@chromium.org

=== PERF REGRESSION ===


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

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


===== BISECT JOB RESULTS =====
Status: completed


===== SUSPECTED CL(s) =====
Subject : Disable Precompile-platform with clang-cl.
Author  : thakis
Commit description:
  
It currently asserts while trying to build that PCH file.

BUG= 667891 , 495697 
NOTRY=true

Review-Url: https://codereview.chromium.org/2520833005
Cr-Commit-Position: refs/heads/master@{#434028}
Commit  : 5b7b04e7fa59d53b8b7917d4c0cf4081abdfb0e3
Date    : Tue Nov 22 23:57:45 2016


===== TESTED REVISIONS =====
Revision         Mean     Std Dev  N   Good?
chromium@433543  1788.94  125.398  30  good
chromium@433936  1795.0   170.952  30  good
chromium@433986  1817.44  116.149  30  good
chromium@434011  1823.07  107.919  30  good
chromium@434023  1842.29  69.0779  30  good
chromium@434026  1844.08  132.499  30  good
chromium@434027  1845.87  245.589  30  good
chromium@434028  1781.49  167.705  30  bad    <--
chromium@434029  1767.58  241.003  30  bad
chromium@434035  1750.39  258.749  30  bad
chromium@434133  1755.26  207.858  30  bad
chromium@434329  1756.67  176.699  30  bad

Bisect job ran on: linux_perf_bisect
Bug ID: 670413

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests blink_perf.layout
Test Metric: flexbox-row-nowrap/flexbox-row-nowrap
Relative Change: 1.80%

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/linux_perf_bisect/builds/6938
Job details: https://chromeperf.appspot.com/buildbucket_job_status/8993377034242768432


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5772778819551232

| 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 Tests>AutoBisect.  Thank you!
Owner: ----
That's a win-only change affecting only clang, which in win isn't used anywhere yet. Your bisecter seems to be broken.
Status: WontFix (was: Untriaged)
This alert was found before M-60 branched. Closing as WontFix as this is believed to either be invalid or non-reproducible.

Sign in to add a comment