1534538 - 2.99 - 10.36% raptor-tp6 (linux64, windows10-64) regression on push 3cddc7cd4da5fcd26b93becb55a028621ab68f64 (Fri Mar 8 2019)

Status: RESOLVED FIXED

(Core :: DOM: Security, defect, P2)

Description

[Ionuț Goldan [:igoldan]]

Raptor has detected a Firefox performance regression from push:

https://hg.mozilla.org/integration/autoland/pushloghtml?changeset=3cddc7cd4da5fcd26b93becb55a028621ab68f64

As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

10% raptor-tp6-microsoft-firefox fcp windows10-64 pgo 482.17 -> 532.12
10% raptor-tp6-wikia-firefox linux64 pgo 164.93 -> 181.44
5% raptor-tp6-wikia-firefox loadtime linux64 pgo 183.94 -> 192.29
5% raptor-tp6-imdb-firefox loadtime linux64 pgo 522.06 -> 545.62
4% raptor-tp6-imgur-firefox fcp linux64 pgo 271.75 -> 281.92
3% raptor-tp6-amazon-firefox fcp windows10-64 pgo 255.77 -> 264.54
3% raptor-tp6-imgur-firefox linux64 pgo 294.21 -> 303.02

Improvements:

6% raptor-assorted-dom-firefox linux64 pgo 26.69 -> 25.10

You can find links to graphs and comparison views for each of the above tests at: https://treeherder.mozilla.org/perf.html#/alerts?id=19829

On the page above you can see an alert for each affected platform as well as a link to a graph showing the history of scores for this test. There is also a link to a Treeherder page showing the Raptor jobs in a pushlog format.

To learn more about the regressing test(s) or reproducing them, please see: https://wiki.mozilla.org/Performance_sheriffing/Raptor

*** Please let us know your plans within 3 business days, or the offending patch(es) will be backed out! ***

Our wiki page outlines the common responses and expectations: https://wiki.mozilla.org/Performance_sheriffing/Talos/RegressionBugsHandling

Comment 1

[Ionuț Goldan [:igoldan]]

These are the Gecko profiles for tp6-amazon on Windows 10 (PGO builds):

before: https://perf-html.io/from-url/https%3A%2F%2Fqueue.taskcluster.net%2Fv1%2Ftask%2FbCMi807nSfmBGpqHiLTZzw%2Fruns%2F0%2Fartifacts%2Fpublic%2Ftest_info%2Fprofile_raptor-tp6-amazon-firefox.zip

after: https://perf-html.io/from-url/https%3A%2F%2Fqueue.taskcluster.net%2Fv1%2Ftask%2FZlKtv1XiRF6VgbwWyB0QUw%2Fruns%2F0%2Fartifacts%2Fpublic%2Ftest_info%2Fprofile_raptor-tp6-amazon-firefox.zip

For tp6-wikia on Linux 64bit (PGO builds):

before: https://perf-html.io/from-url/https%3A%2F%2Fqueue.taskcluster.net%2Fv1%2Ftask%2FRn2AnejURI6TEMjoxH_o8Q%2Fruns%2F0%2Fartifacts%2Fpublic%2Ftest_info%2Fprofile_raptor-tp6-wikia-firefox.zip

after: https://perf-html.io/from-url/https%3A%2F%2Fqueue.taskcluster.net%2Fv1%2Ftask%2FLYoLp1vcRGerEpGXMM2DOg%2Fruns%2F0%2Fartifacts%2Fpublic%2Ftest_info%2Fprofile_raptor-tp6-wikia-firefox.zip

Comment 2

[Andrea Marchesini [:baku]]

I'm debugging this regression. I'll tell you more about the time range for the fix for today, max tomorrow.

Comment 3

[Andrea Marchesini [:baku]]

Attachment: Bug 1534538 - Avoid sending CookieSettings from parent to child if not needed, r?ehsan

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/33d53c6ee136
Avoid sending CookieSettings from parent to child if not needed, r=Ehsan

Comment 5

[Dorel Luca [:dluca]]

https://hg.mozilla.org/mozilla-central/rev/33d53c6ee136

Comment 6

[Ionuț Goldan [:igoldan]]

I'm afraid push from comment 4 didn't fix the regressions. I only saw a partial fix on raptor-tp6-amazon-firefox fcp windows10-64-pgo-qr, but the other baselines remained unchanged.

Comment 7

[Ionuț Goldan [:igoldan]]

I also noticed a fix for raptor-tp6-microsoft-firefox fcp linux64 pgo:

== Change summary for alert #19948 (as of Mon, 18 Mar 2019 06:58:24 GMT) ==

Improvements:

4% raptor-tp6-microsoft-firefox fcp linux64 pgo 511.06 -> 488.92

For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=19948

Comment 8

[Ionuț Goldan [:igoldan]]

This is the Compare view for the new patch Andrea is working on. Keeping it here for easy reference.

Comment 9

[Andrea Marchesini [:baku]]

Attachment: Bug 1534538 - Principal serialization should not serialize base domain if not needed, r?ehsan

Comment 10

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/66501cf7b6a4
Principal serialization should not serialize base domain if not needed, r=Ehsan

Comment 11

[Noemi Erli[:noemi_erli]]

https://hg.mozilla.org/mozilla-central/rev/66501cf7b6a4

Comment 12

[Ionuț Goldan [:igoldan]]

Push from comment 10 still didn't fix the perf regressions. All of them remained the same.
There's lots of noise, especially after March 29, which makes the changes even more difficult to identify.

Comment 13

[Andrea Marchesini [:baku]]

Can you tell me the % of regression after comment 10?

Comment 14

[Ionuț Goldan [:igoldan]]

(In reply to Andrea Marchesini [:baku] from comment #13)

Can you tell me the % of regression after comment 10?

Yes, but you'll have to wait a bit so I can do some retriggers to obtain that percentage.

Comment 15

[Ionuț Goldan [:igoldan]]

I've done the retriggers. In 3 ours I should have enough data to reply back.

Comment 16

[Ionuț Goldan [:igoldan]]

I looked over the retriggers and and it's difficult to precisely state in percentages how much we've fixed.
The noise & big harness update around March 29 (which should mix into the equation) hide the small gains.

Comment 17

[Pascal Chevrel:pascalc]

Marking as wontfix for 67 as we are not sure that the performance regressions would be addressed by the patches in this bug and we are in the middle of the nightly cycle already.

Comment 18

[Andrea Marchesini [:baku]]

I did some improvements lately: bug 1554527 (not landed yet) and 1553867. Would be great if we can check if we still have regressions.