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

Issue 635918 link

Starred by 1 user

Issue metadata

Status: Archived
Owner:
Last visit > 30 days ago
Closed: Sep 2016
Cc:
EstimatedDays: ----
NextAction: ----
OS: Mac
Pri: 2
Type: Bug-Regression

Blocked on:
issue 639287
issue 639828



Sign in to add a comment

48.7%-71.8% regression in system_health.memory_desktop at 410080:410419

Project Member Reported by petrcermak@chromium.org, Aug 9 2016

Issue description

See the link to graphs below.
 
All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=635918

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgtsGSvgoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg1rnGrgkM


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

chromium-rel-mac-retina
Project Member

Comment 3 by 42576172...@developer.gserviceaccount.com, Aug 10 2016

Cc: rsesek@chromium.org
Owner: rsesek@chromium.org

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

Hi rsesek@chromium.org, the bisect results pointed to your CL below as possibly
causing a regression. Please have a look at this info and see whether
your CL be related.


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


===== SUSPECTED CL(s) =====
Subject : Revert of [Mac] Only enable the CFBundleBlocker in the browser process. (patchset #1 id:1 of https://codereview.chromium.org/2199313002/ )
Author  : rsesek
Commit description:
  
Reason for revert:
Speculative revert for perf regression.

BUG= https://crbug.com/634917 

Original issue's description:
> [Mac] Only enable the CFBundleBlocker in the browser process.
>
> In child processes, which are sandboxed in most cases, the sandbox will prevent
> access to the filesystem locations where the potentially blocked bundles are
> stored. Furthermore, on macOS 10.11 and higher, the Google Chrome build is
> codesigned in such a way where bundle loading is blocked by SIP.
>
> This reduces some of the "triggered DYLD shared region unnest for map" messages.
>
> BUG= 428858 
> R=mark@chromium.org
>
> Committed: https://crrev.com/65e732d3b4dc595512b731143fd49d372acc1a87
> Cr-Commit-Position: refs/heads/master@{#409244}

TBR=mark@chromium.org,thakis@chromium.org
# Not skipping CQ checks because original CL landed more than 1 days ago.
BUG= 428858 

Review-Url: https://codereview.chromium.org/2218163002
Cr-Commit-Position: refs/heads/master@{#410113}
Commit  : 3026ec1f3bd67bb2cac5e42fe6d1a1298efcd049
Date    : Fri Aug 05 18:22:47 2016


===== TESTED REVISIONS =====
Revision         Mean      Std Dev   N  Good?
chromium@410079  44525384  0.0       5  good
chromium@410101  44525384  0.0       5  good
chromium@410112  44525384  0.0       5  good
chromium@410113  76411003  147876    5  bad    <--
chromium@410114  76212245  147001    5  bad
chromium@410115  76343931  180226    5  bad
chromium@410117  82046485  13006664  5  bad
chromium@410122  76343931  180226    5  bad
chromium@410164  82569851  13995365  5  bad
chromium@410249  76343931  180226    5  bad
chromium@410419  82702459  14104744  5  bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 85.74%
Score: 99.9

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1558
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9004795970942841504


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5802975524552704

| 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 4 by rsesek@chromium.org, Aug 10 2016

Owner: ----
Status: Untriaged (was: Assigned)
Don't think that's right. That CL was a revert for a different perf regression.
Owner: petrcermak@chromium.org
Status: Assigned (was: Untriaged)
The result of the bisect seems very clear, but I'm running another bisect.
Project Member

Comment 8 by 42576172...@developer.gserviceaccount.com, Aug 11 2016


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


===== SUSPECTED CL(s) =====
Subject : Revert of [Mac] Only enable the CFBundleBlocker in the browser process. (patchset #1 id:1 of https://codereview.chromium.org/2199313002/ )
Author  : rsesek
Commit description:
  
Reason for revert:
Speculative revert for perf regression.

BUG= https://crbug.com/634917 

Original issue's description:
> [Mac] Only enable the CFBundleBlocker in the browser process.
>
> In child processes, which are sandboxed in most cases, the sandbox will prevent
> access to the filesystem locations where the potentially blocked bundles are
> stored. Furthermore, on macOS 10.11 and higher, the Google Chrome build is
> codesigned in such a way where bundle loading is blocked by SIP.
>
> This reduces some of the "triggered DYLD shared region unnest for map" messages.
>
> BUG= 428858 
> R=mark@chromium.org
>
> Committed: https://crrev.com/65e732d3b4dc595512b731143fd49d372acc1a87
> Cr-Commit-Position: refs/heads/master@{#409244}

TBR=mark@chromium.org,thakis@chromium.org
# Not skipping CQ checks because original CL landed more than 1 days ago.
BUG= 428858 

Review-Url: https://codereview.chromium.org/2218163002
Cr-Commit-Position: refs/heads/master@{#410113}
Commit  : 3026ec1f3bd67bb2cac5e42fe6d1a1298efcd049
Date    : Fri Aug 05 18:22:47 2016


===== TESTED REVISIONS =====
Revision         Mean      Std Dev  N  Good?
chromium@410079  32359536  1025801  5  good
chromium@410101  32359536  1025801  5  good
chromium@410112  32359536  1025801  5  good
chromium@410113  48793507  280.434  5  bad    <--
chromium@410114  48711638  112974   5  bad
chromium@410115  51636029  6355900  5  bad
chromium@410117  48670115  112828   5  bad
chromium@410122  48629053  92047.3  5  bad
chromium@410164  48670115  112594   5  bad
chromium@410249  48670115  112594   5  bad
chromium@410419  48752906  92254.1  5  bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:gpu:effective_size_avg/load_search_baidu
Relative Change: 50.66%
Score: 99.9

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1561
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9004700693610793536


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5795554492153856

| 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: rsesek@chromium.org
resek: Another bisect (on a different metric) confirmed that the revert (r410113) indeed caused the regression (+15.7 MiB in GPU effective size, +30.4 MiB in CC effective size). Please figure out whether this was expected and if anything needs to be done.

I can see on the charts (https://chromeperf.appspot.com/group_report?bug_id=635918), that the regression followed a symmetrical improvement between r409580 and r409624 (http://test-results.appspot.com/revision_range?start=409580&end=409624). Is any of the patches in that range related to your revert (r410113)?
No, my patch originally landed at 409244, which doesn't show as the starting point for the improvement. So either the starting point for the improvement is wrong, there's an issue with the bisect, or there's an issue with the test.

7490bda13d8646a6225b45bb45126be71e19358c and 8ee20259fc28430498df2b22cb4a506d215d26dd could be candidates, but it does not align with the graph. I'm inclined to WontFix.
This is a +15.7 (+30.4) MiB regression in the effective size of GPU (cc), so I'd really like to get to the bottom of this. I'll bisect the improvement to figure out what caused it and hopefully it will shed some light.

I don't think that the bisect is wrong here (given that it repeatedly blamed the same patch).
Project Member

Comment 13 by 42576172...@developer.gserviceaccount.com, Aug 12 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean      Std Dev  N   Good?
chromium@409579  44525384  0.0      12  good
chromium@409624  44525384  0.0      8   bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 0.00%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1565
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9004527254386230368


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5790290741297152

| 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!
Project Member

Comment 15 by 42576172...@developer.gserviceaccount.com, Aug 16 2016

Cc: dmazz...@chromium.org
Owner: dmazz...@chromium.org

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

Hi dmazzoni@chromium.org, the bisect results pointed to your CL below as possibly
causing a regression. Please have a look at this info and see whether
your CL be related.


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


===== SUSPECTED CL(s) =====
Subject : Add aboxhall as owner of modules/accessibility in Blink
Author  : dmazzoni
Commit description:
  
BUG=none

Review-Url: https://codereview.chromium.org/2204593006
Cr-Commit-Position: refs/heads/master@{#409246}
Commit  : 8b1118e99d2b589b523ab2f97b3c6cc500f31df1
Date    : Tue Aug 02 18:17:58 2016


===== TESTED REVISIONS =====
Revision         Mean      Std Dev   N  Good?
chromium@409210  76434632  116422    8  good
chromium@409236  82635387  14143037  5  good
chromium@409243  76410082  147345    5  good
chromium@409245  76478792  2827.61   5  good
chromium@409246  44525384  0.0       5  bad    <--
chromium@409249  44525384  0.0       5  bad
chromium@409262  48519240  11296331  8  bad
chromium@409314  44525384  0.0       5  bad
chromium@409417  44525384  0.0       5  bad
chromium@409624  44525384  0.0       5  bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 41.73%
Score: 99.9

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1566
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9004268388312819344


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5887771248427008

| 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: petrcermak@chromium.org
It's obviously not my change either, that's just a change to an OWNERS file, it doesn't affect Chrome at all.

No idea what's going on, but it's troublesome that when the test fails it has no standard deviation.

Cc: -dmazz...@chromium.org sullivan@chromium.org robert...@chromium.org dtu@chromium.org sdefresne@chromium.org
Labels: OS-Mac
I agree that this is quite worrying.

Here are the traces for the two revisions:

r409245 (cc=72.9 MiB): https://console.developers.google.com/m/cloudstorage/b/chrome-telemetry-output/o/trace-file-id_6-2016-08-15_17-19-42-28832.html
r409246 (cc=42.5 MiB): https://console.developers.google.com/m/cloudstorage/b/chrome-telemetry-output/o/trace-file-id_6-2016-08-15_16-01-13-25061.html

[Note that we were bisecting an *improvement*]

Clearly, the numbers must be wrong because r409246 just adds an owner.
The only rational explanation I can think of is that there must be a
difference between the two builds:

r409245: gs://chrome-perf/Mac Builder/full-build-mac_39bef912185b65331bb3f86c7f035eeb34b18792.zip
r409246: gs://chrome-perf/Mac Builder/full-build-mac_8b1118e99d2b589b523ab2f97b3c6cc500f31df1.zip

Unfortunately, the builder runs corresponding to these builds have
already been deleted. To figure out if there's any difference between
the two, I downloaded and unzipped them. I diffed the resulting folders
(diff -Nqr r409245 r409246 | sed 's/^Files r409245\/\(.*\) and r409246\/\(.*\) differ$/\1/g' | sort).
Firstly, the following groups of files seem irrelevant:

===== CRASHPAD & BREAKPAD [there were no crashes] =====
crashpad_handler
crashpad_handler-54.0.2812.0.breakpad
crashpad_handler-54.0.2813.0.breakpad
crashpad_handler-54.0.2814.0.breakpad
crashpad_handler-54.0.2815.0.breakpad
crashpad_handler-54.0.2816.0.breakpad
crashpad_handler-54.0.2817.0.breakpad
Google Chrome.app/Contents/Versions/54.0.2817.0/Google Chrome Framework.framework/Helpers/crashpad_handler
Google Chrome Framework-54.0.2812.0.breakpad
Google Chrome Framework-54.0.2813.0.breakpad
Google Chrome Framework-54.0.2814.0.breakpad
Google Chrome Framework-54.0.2815.0.breakpad
Google Chrome Framework-54.0.2816.0.breakpad
Google Chrome Framework-54.0.2817.0.breakpad
Google Chrome Framework.framework/Helpers/crashpad_handler

===== TEXT FILES WITH REVISION INFO =====
FULL_BUILD_REVISION
Google Chrome.app/Contents/Info.plist
Google Chrome.app/Contents/Versions/54.0.2817.0/Google Chrome Framework.framework/Resources/Info.plist
Google Chrome Framework.framework/Resources/Info.plist

===== MOJO [no difference after unzipping] =====
irt_x64/gen/ipc/ipc.mojom.srcjar
irt_x64/gen/mojo/public/interfaces/bindings/interface_control_messages.mojom.srcjar
irt_x64/gen/mojo/public/interfaces/bindings/pipe_control_messages.mojom.srcjar

===== PROTO =====
pyproto/chrome/common/safe_browsing/download_file_types_pb2.pyc
pyproto/google/__init__.pyc
pyproto/google/protobuf/descriptor_database.pyc
pyproto/google/protobuf/descriptor_pb2.pyc
pyproto/google/protobuf/descriptor_pool.pyc
pyproto/google/protobuf/descriptor.pyc
pyproto/google/protobuf/__init__.pyc
pyproto/google/protobuf/internal/api_implementation.pyc
pyproto/google/protobuf/internal/containers.pyc
pyproto/google/protobuf/internal/decoder.pyc
pyproto/google/protobuf/internal/encoder.pyc
pyproto/google/protobuf/internal/enum_type_wrapper.pyc
pyproto/google/protobuf/internal/__init__.pyc
pyproto/google/protobuf/internal/message_listener.pyc
pyproto/google/protobuf/internal/python_message.pyc
pyproto/google/protobuf/internal/type_checkers.pyc
pyproto/google/protobuf/internal/well_known_types.pyc
pyproto/google/protobuf/internal/wire_format.pyc
pyproto/google/protobuf/message.pyc
pyproto/google/protobuf/reflection.pyc
pyproto/google/protobuf/symbol_database.pyc
pyproto/google/protobuf/text_encoding.pyc
pyproto/google/protobuf/text_format.pyc
pyproto/google/third_party/six/six.pyc

The following differences seemed potentially relevant:

===== BINARY FILES =====
Google Chrome.app/Contents/Versions/54.0.2817.0/Google Chrome Framework.framework/Google Chrome Framework
Google Chrome.dSYM.tar.bz2 [The two contained files below differ]
  crashpad_handler.dSYM/Contents/Resources/DWARF/crashpad_handler
  Google Chrome Framework.dSYM/Contents/Resources/DWARF/Google Chrome Framework
Google Chrome Framework.framework/Google Chrome Framework
performance_browser_tests

Of these files, the following differ in size:

Google Chrome.dSYM.tar.bz2: 443484 -> 443448
  Google Chrome Framework.dSYM/Contents/Resources/DWARF/Google Chrome Framework: 1917072 -> 1917096
irt_x64/obj/ppapi/proxy/libppapi_proxy.a: 42372 -> 42368

===== NINJA FILES =====
build.ninja
toolchain.ninja

After sorting, the build.ninja files are identical. There are only 5
different lines between the toolchain.ninja files. Specifically,
TOOL_VERSIONs of filter_libtool.py, linker_driver.py and
compile_xcassets.py have changed:

                                         r409245       r409246
build/toolchain/mac/filter_libtool.py:   1468923233 -> 1468923090
build/toolchain/mac/linker_driver.py:    1468932295 -> 1468933041
build/toolchain/mac/compile_xcassets.py: 1468923233 -> 1468923090

these timestamps translate to 19th July 2016 with the following UTC
times (yes, the first and the third one seem to go backwards):

build/toolchain/mac/filter_libtool.py:   10:13:53 -> 10:11:30
build/toolchain/mac/linker_driver.py:    12:44:55 -> 12:57:21
build/toolchain/mac/compile_xcassets.py: 10:13:53 -> 10:11:30

Looking at the git history of the three files, the following two patches
match the region:

r406110 by rsesek@    landing on 18th July 2016 @ 21:59:17 UTC
r406253 by sdefresne@ landing on 19th July 2016 @ 12:43:59 UTC

rsesek,sdefresne: Is there any chance either of your changes could have
led to such a dramatic improvement in memory usage?

It seems unlikely that this could be due to a hardware/software change
on the builder. The modification timestamps of the
"Google Chrome Framework.framework/Google Chrome Framework" file are
just 3 minutes apart:

  r409245: 2nd August 2016 @ 11:53
  r409246: 2nd August 2016 @ 11:56

In fact, these two seem to be too close to come from the same device.
Perhaps the two builds were generated by different slaves.

sullivan,robertocn,dtu: Could there be a difference in build
configuration across build slaves of a single bot? Any other ideas how
the two builds could differ?
We did at one point have a problem where most of the Mac builders were running 10.11 but one was running 10.9, but that was fixed in bug 608233 long before the revision range of this regression.

For the perf builders, you can see some data on the sizes of the binaries they're building, and which revisions they built originally, by looking at the sizes chart:
https://chromeperf.appspot.com/report?sid=24d6dcc57f09b4bed2677c35f7c66ebf2a8125bb0a558d55496af4d4af30c55c&start_rev=409225&end_rev=409300

It looks like they are building every revision, as expected. Does anything look strange in the sizes to you?

I am not sure how to figure out:
* Whether any revisions the bisects used were built by mac_perf_bisect_builder
* Whether mac_perf_bisect_builder's slaves are all set up as expected.
https://build.chromium.org/p/tryserver.chromium.perf/builders/mac_perf_bisect_builder

Dave, do you know?
According to the sizes chart, there is no difference between r409245 and r409246 (both show "Value: 122,513,000").

Comment 20 by dtu@chromium.org, Aug 18 2016

In the past we have seen sizes differences across different builders despite them running the exact same hardware and OS version:
https://chromeperf.appspot.com/report?sid=0b68579269a9a5a4402ed6ac605e79dc9b793a1ae695f94b6ec0b94117e9fede&start_rev=407068&end_rev=407282

I can confirm that all the Mac Builders and mac_perf_bisect_builders are running OS X 10.11.5 (15F34):
https://docs.google.com/spreadsheets/d/1LTOSY9y1_sdDiL94XQTZrXmSzokn4p44hlgHQvLnEt4/preview
rsesek,sdefresne: ping (#17)
No, I don't think either of those CLs would have an effect on memory usage.
rsesek: Thanks!

This is really really weird. I'll try to bisect the improvement once again with the "Use archive" option unchecked.
Blockedon: 639287
Turns out that the "use archive" checkbox doesn't do anything :-(
Project Member

Comment 26 by 42576172...@developer.gserviceaccount.com, Aug 19 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean      Std Dev   N   Good?
chromium@409579  73760627  9207606   12  good
chromium@409624  84466568  14791277  8   bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 27.47%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1589
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003899331866169952


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5344056775802880

| 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!
OK, I really don't understand this. The bisects in #15 and #26 were on the same platform and the same value. Yet they reported completely different numbers for r409624:

       Revision         Mean      Std Dev   N  Good?
  #15: chromium@409624  44525384  0.0       5  bad
  #26: chromium@409624  84466568  14791277  8  bad

This is a complete enigma to me :-\
Furthermore, both bisects ran on the same slave a few days apart:

#15: https://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1566 (Monday, August 15th)
#26: https://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1589 (Friday, August 19th)

and had the same configuration (except for a different "good" revision):

#15: https://chromeperf.appspot.com/buildbucket_job_status/9004268388312819344
#26: https://chromeperf.appspot.com/buildbucket_job_status/9003899331866169952

I'm running another bisect with the same config as in #15.
Project Member

Comment 30 by 42576172...@developer.gserviceaccount.com, Aug 20 2016


===== BISECT JOB RESULTS =====
Status: failed


===== TESTED REVISIONS =====
Revision         Mean      Std Dev   N   Good?
chromium@409210  77676929  6094741   27  good
chromium@409262  75282267  6147158   27  bad
chromium@409314  76477740  2577.84   18  bad
chromium@409417  76480840  0.0       5   bad
chromium@409624  80409160  11121104  8   bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 8.60%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1592
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003878490358401616


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5285414902956032

| 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!
Project Member

Comment 32 by 42576172...@developer.gserviceaccount.com, Aug 20 2016


===== BISECT JOB RESULTS =====
Status: failed


===== TESTED REVISIONS =====
Revision         Mean      Std Dev   N   Good?
chromium@409210  77676929  6094741   27  good
chromium@409262  75282267  6147158   27  bad
chromium@409314  76477740  2577.84   18  bad
chromium@409417  76480840  0.0       5   bad
chromium@409624  80409160  11121104  8   bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 8.60%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1592
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003878490358401616


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5285414902956032

| 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!
Project Member

Comment 34 by 42576172...@developer.gserviceaccount.com, Aug 21 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean      Std Dev  N   Good?
chromium@409210  76393757  149107   12  good
chromium@409624  76436040  117011   8   bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 0.00%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1595
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003878490358401616


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5285414902956032

| 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!
Blockedon: 639828
Project Member

Comment 37 by 42576172...@developer.gserviceaccount.com, Aug 22 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean      Std Dev  N   Good?
chromium@409579  46692080  5191918  12  good
chromium@409624  46510576  5590339  8   bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:gpu:effective_size_avg/load_search_baidu
Relative Change: 0.97%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1599
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003624442702882688


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5778642873876480

| 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!
Project Member

Comment 39 by 42576172...@developer.gserviceaccount.com, Aug 23 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean      Std Dev  N   Good?
chromium@409210  48725019  101278   12  good
chromium@409624  48779088  19901.5  8   bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:gpu:effective_size_avg/load_search_baidu
Relative Change: 0.14%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1602
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003542269757162832


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6444331263590400

| 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!
Project Member

Comment 41 by 42576172...@developer.gserviceaccount.com, Aug 25 2016


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


===== SUSPECTED CL(s) =====
Subject : Revert of [Mac] Only enable the CFBundleBlocker in the browser process. (patchset #1 id:1 of https://codereview.chromium.org/2199313002/ )
Author  : rsesek
Commit description:
  
Reason for revert:
Speculative revert for perf regression.

BUG= https://crbug.com/634917 

Original issue's description:
> [Mac] Only enable the CFBundleBlocker in the browser process.
>
> In child processes, which are sandboxed in most cases, the sandbox will prevent
> access to the filesystem locations where the potentially blocked bundles are
> stored. Furthermore, on macOS 10.11 and higher, the Google Chrome build is
> codesigned in such a way where bundle loading is blocked by SIP.
>
> This reduces some of the "triggered DYLD shared region unnest for map" messages.
>
> BUG= 428858 
> R=mark@chromium.org
>
> Committed: https://crrev.com/65e732d3b4dc595512b731143fd49d372acc1a87
> Cr-Commit-Position: refs/heads/master@{#409244}

TBR=mark@chromium.org,thakis@chromium.org
# Not skipping CQ checks because original CL landed more than 1 days ago.
BUG= 428858 

Review-Url: https://codereview.chromium.org/2218163002
Cr-Commit-Position: refs/heads/master@{#410113}
Commit  : 3026ec1f3bd67bb2cac5e42fe6d1a1298efcd049
Date    : Fri Aug 05 18:22:47 2016


===== TESTED REVISIONS =====
Revision         Mean      Std Dev   N  Good?
chromium@410079  50849608  14141395  5  good
chromium@410101  44525384  0.0       5  good
chromium@410112  44525384  0.0       5  good
chromium@410113  76344341  180599    5  bad    <--
chromium@410114  76409877  147231    5  bad
chromium@410115  76476027  457.947   5  bad
chromium@410117  76345058  181266    5  bad
chromium@410122  76277986  180039    5  bad
chromium@410164  76279112  181592    5  bad
chromium@410249  82701742  14105711  5  bad
chromium@410419  76345058  181266    5  bad

Bisect job ran on: mac_retina_perf_bisect
Bug ID: 635918

Test Command: src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --also-run-disabled-tests system_health.memory_desktop
Test Metric: load_search-memory:chrome:all_processes:reported_by_chrome:cc:effective_size_avg/load_search_baidu
Relative Change: 50.14%
Score: 99.9

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/mac_retina_perf_bisect/builds/1607
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003435607012180608


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5772976805380096

| 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 42 by dtu@chromium.org, Aug 25 2016

The metric is bimodal and maybe even trimodal. It's never going to be able to bisect.

If you look at the individual values of the last run, even multiple iterations that ran consecutively produced bimodal values.

chromium@410079 [44525384, 76146504, 44525384, 44525384, 44525384]
chromium@410249 [76476232, 107933512, 76146504, 76476232, 76476232]

I've attached plots of the results from comment 37 and comment 41. They both show these kinds of bimodal outliers.
37.png
48.5 KB View Download
41.png
92.0 KB View Download

Comment 43 by dtu@chromium.org, Aug 25 2016

Is it possible that some kind of external factor could affect this metric? If, for example, some other process starting and stopping on the device could affect the metric, that could lead to this kind of bimodal behavior.
Labels: SystemHealth-Sheriff
Labels: Hotlist-SystemHealthBankruptcy
Status: Archived (was: Assigned)
Temporarily declaring bankruptcy on the *desktop* system health benchmark.
The number of alerts became unmanageable and the overall process needs to be improved to make it sustainable.
The alerts have been turned off and I'm archiving the outstanding regressions.
Note: this is just about desktop, the mobile system health stays up. 

Sign in to add a comment