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

Issue 688448 link

Starred by 1 user

Issue metadata

Status: WontFix
Owner:
Last visit > 30 days ago
Closed: Feb 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

1.4%-16.1% regression in system_health.memory_mobile at 447578:447645

Project Member Reported by jasontiller@chromium.org, Feb 3 2017

Issue description

See the link to graphs below.
 

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

Bisect Details
  Configuration: android_nexus5X_perf_bisect
  Benchmark    : system_health.memory_mobile
  Metric       : memory:chrome:all_processes:reported_by_os:system_memory:java_heap:proportional_resident_size_avg/load_search/load_search_amazon

Revision             Result                  N
chromium@447577      9376719 +- 3275937      21      good
chromium@447645      9675191 +- 3842124      21      bad

Please refer to the following doc on diagnosing memory regressions:
  https://chromium.googlesource.com/chromium/src/+/master/docs/memory-infra/memory_benchmarks.md

To Run This Test
  src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=load.search.amazon system_health.memory_mobile

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8988655071917839584

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5172304981524480


| 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!

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

Bisect Details
  Configuration: android_nexus5X_perf_bisect
  Benchmark    : system_health.memory_mobile
  Metric       : memory:chrome:all_processes:reported_by_os:system_memory:native_heap:proportional_resident_size_avg/load_news/load_news_hackernews

Revision             Result                  N
chromium@447577      35187496 +- 699256      21      good
chromium@447645      35200662 +- 939307      21      bad

Please refer to the following doc on diagnosing memory regressions:
  https://chromium.googlesource.com/chromium/src/+/master/docs/memory-infra/memory_benchmarks.md

To Run This Test
  src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=load.news.hackernews system_health.memory_mobile

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8988639290587133040

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6111291871068160


| 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: asvitk...@chromium.org
Owner: asvitk...@chromium.org

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

Hi asvitkine@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 : asvitkine
  Commit : a86e8e115054ad167458d0379980db5473cb20de
  Date   : Wed Feb 01 19:33:59 2017
  Subject: Don't set cpu architecture field on iOS in UMA logs.

Bisect Details
  Configuration: android_nexus5X_perf_bisect
  Benchmark    : system_health.memory_mobile
  Metric       : memory:chrome:all_processes:reported_by_os:system_memory:java_heap:proportional_resident_size_avg/load_social/load_social_facebook
  Change       : 1.31% | 8138678.85714 -> 8228205.71429

Revision             Result                  N
chromium@447577      8138679 +- 1698665      14      good
chromium@447578      9652565 +- 448626       6       bad       <--
chromium@447579      9496576 +- 83542.3      6       bad
chromium@447580      9513643 +- 507215       6       bad
chromium@447582      9734144 +- 288074       6       bad
chromium@447586      9846443 +- 42828.8      6       bad
chromium@447594      9645397 +- 151564       6       bad
chromium@447612      9542315 +- 119327       6       bad
chromium@447645      8228206 +- 2254966      14      bad

Please refer to the following doc on diagnosing memory regressions:
  https://chromium.googlesource.com/chromium/src/+/master/docs/memory-infra/memory_benchmarks.md

To Run This Test
  src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=load.social.facebook system_health.memory_mobile

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8988639296319866736

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5031634266488832


| 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!
Bisect is wrong. My change was iOS-specific and could not have affected Android.
Owner: jasontiller@chromium.org
Specifically, here's my change:

   metrics::SystemProfileProto::Hardware* hardware =
       system_profile->mutable_hardware();
+#if !defined(OS_IOS)
+  // On iOS, OperatingSystemArchitecture() returns values like iPad4,4 which is
+  // not the actual CPU architecture. Don't set it until the API is fixed. See
+  // crbug.com/370104 for details.
   hardware->set_cpu_architecture(base::SysInfo::OperatingSystemArchitecture());
+#endif
   hardware->set_system_ram_mb(base::SysInfo::AmountOfPhysicalMemoryMB());


So the above should not have resulted in any difference in compiled code on Android.

Re-assigning back to reporter.
Thanks asvitkine! Sorry for the incorrect ping from Bisect. I've kicked off a couple more to see if we can get a correct bisect. 
Cc: baxley@chromium.org
Owner: baxley@chromium.org

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

Hi baxley@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 : baxley
  Commit : 3432563dba5d8feb7819262688a0741a66c6bf35
  Date   : Wed Feb 01 21:20:41 2017
  Subject: Update web shell app to not load dev.chromium.org on launch.

Bisect Details
  Configuration: android_nexus5X_perf_bisect
  Benchmark    : system_health.memory_mobile
  Metric       : memory:chrome:all_processes:reported_by_os:system_memory:java_heap:proportional_resident_size_avg/load_tools/load_tools_stackoverflow
  Change       : 16.57% | 9255936.0 -> 7722666.66667

Revision             Result                  N
chromium@447577      9255936 +- 2518157      6      good
chromium@447594      7925760 +- 194053       6      good
chromium@447603      8212139 +- 271384       6      good
chromium@447607      8515584 +- 220819       6      good
chromium@447609      8769536 +- 307729       6      good
chromium@447610      9131008 +- 190805       6      good
chromium@447611      7748949 +- 66404.9      6      bad       <--
chromium@447645      7722667 +- 72025.4      6      bad

Please refer to the following doc on diagnosing memory regressions:
  https://chromium.googlesource.com/chromium/src/+/master/docs/memory-infra/memory_benchmarks.md

To Run This Test
  src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=load.tools.stackoverflow system_health.memory_mobile

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8988383767213701856

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5792117953134592


| 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: jasontiller@chromium.org
My change was also iOS specific (part of web shell app and tests, not even iOS Chrome code), so I don't think this could affect Android.

Assigning back to jasontiller@ - let me know if you have any questions
Cc: sullivan@chromium.org perezju@chromium.org
Yikes! Sorry about that, y'all. CCing in speed infra (+sullivan) and the benchmark owner (+perezju) who might have some more insight into what's going on here.
Cc: simonhatch@chromium.org
The bisects both look like they got bimodal results.
Kicked off a bisect on  "memory:chrome:gpu_process:reported_by_os:system_memory:java_heap:proportional_resident_size_avg/load_news/load_news_bbc" where the regression looks cleaner from:
https://chromeperf.appspot.com/report?sid=79a4b6e4980e7300d110f1f024dc7902b52b4dfa9d2b7d6563f68276f14aac93&rev=447645

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

Suspected Commit
  Author : asvitkine
  Commit : a86e8e115054ad167458d0379980db5473cb20de
  Date   : Wed Feb 01 19:33:59 2017
  Subject: Don't set cpu architecture field on iOS in UMA logs.

Bisect Details
  Configuration: android_nexus5X_perf_bisect
  Benchmark    : system_health.memory_mobile
  Metric       : memory:chrome:gpu_process:reported_by_os:system_memory:java_heap:proportional_resident_size_avg/load_news/load_news_bbc
  Change       : 3.14% | 1369429.33333 -> 1412437.33333

Revision             Result                  N
chromium@447577      1369429 +- 27272.5      6      good
chromium@447578      1601024 +- 32855.9      6      bad       <--
chromium@447579      1570304 +- 17058.1      6      bad
chromium@447580      1557504 +- 52693.7      6      bad
chromium@447582      1518421 +- 31342.3      6      bad
chromium@447586      1492992 +- 66347.0      6      bad
chromium@447594      1473536 +- 16190.9      6      bad
chromium@447611      1447083 +- 16914.1      6      bad
chromium@447645      1412437 +- 56546.1      6      bad

Please refer to the following doc on diagnosing memory regressions:
  https://chromium.googlesource.com/chromium/src/+/master/docs/memory-infra/memory_benchmarks.md

To Run This Test
  src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=load.news.bbc system_health.memory_mobile

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8988327389205641216

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5793065530294272


| 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: primiano@chromium.org hjd@chromium.org
+hjd, +primiano: hmm.. those results in #18 suspiciously look like the metric monotonically increasing in time. Hector, could you have a look?

Comment 20 by hjd@chromium.org, Feb 7 2017

Owner: hjd@chromium.org
Status: Assigned (was: Untriaged)

Comment 21 by hjd@chromium.org, Feb 8 2017

So far:

* We're not restarting the phone between runs in the bisect (unlike on the dashboard).
* The metric is indeed monotonically increasing in time in the bisect.
* Hard to say if this repos locally as the metric is quite noisy.
* reported_by_chrome's java_heap metric is rock solid (down to the byte) (both locally and in the bisect).
* However there is a weird movement of memory from allocated_objects to <unspecified> in java_heap as time goes on both locally and on the bisect.


Comment 22 by hjd@chromium.org, Feb 8 2017

Status: WontFix (was: Assigned)
Seems like a false regression, all graphs have recovered:

https://chromeperf.appspot.com/report?sid=79a4b6e4980e7300d110f1f024dc7902b52b4dfa9d2b7d6563f68276f14aac93&rev=447645
https://chromeperf.appspot.com/group_report?bug_id=688448

However we discovered two new bugs:
 crbug.com/690040  java_heap:allocated_objects_size_avg is super noisy
crbug.com/690042 bisect 'leaks' java_heap:allocated_objects_size_avg over many runs
Labels: Performance-Memory

Sign in to add a comment