5.96 - 27.08% wikipedia / wikipedia LastVisualChange / wikipedia fcp / wikipedia loadtime (linux64-shippable, linux64-shippable-qr, macosx1014-64-shippable-qr) regression on push facbcc4306d1a824cb607cb3bbc54a7b13d1038c (Fri January 8 2021)
Categories
(Core :: Layout: Generated Content, Lists, and Counters, defect)
Tracking
()
| Tracking | Status | |
|---|---|---|
| firefox-esr78 | --- | unaffected |
| firefox84 | --- | unaffected |
| firefox85 | --- | unaffected |
| firefox86 | --- | fixed |
| firefox87 | --- | fixed |
People
(Reporter: Bebe, Assigned: emilio)
References
(Regression)
Details
(Keywords: perf, perf-alert, regression, Whiteboard: [perf:alert:1] )
Attachments
(3 files)
Perfherder has detected a browsertime performance regression from push 04b98b03122fc7cbb8017c01d654cb3d9efecd83. As author of one of the patches included in that push, we need your help to address this regression.
Regressions:
| Ratio | Suite | Test | Platform | Options | Absolute values (old vs new) |
|---|---|---|---|---|---|
| 27% | wikipedia | loadtime | linux64-shippable | nocondprof warm | 604.75 -> 768.50 |
| 26% | wikipedia | loadtime | linux64-shippable-qr | nocondprof warm webrender | 625.29 -> 789.17 |
| 24% | wikipedia | loadtime | linux64-shippable-qr | nocondprof warm webrender | 641.27 -> 792.08 |
| 20% | wikipedia | fcp | linux64-shippable-qr | nocondprof warm webrender | 641.88 -> 771.42 |
| 19% | wikipedia | fcp | linux64-shippable | nocondprof warm | 625.33 -> 745.62 |
| 18% | wikipedia | linux64-shippable | nocondprof warm | 678.58 -> 801.58 | |
| 18% | wikipedia | linux64-shippable-qr | nocondprof warm webrender | 697.97 -> 824.49 | |
| 18% | wikipedia | LastVisualChange | linux64-shippable | nocondprof warm | 980.00 -> 1,156.67 |
| 16% | wikipedia | LastVisualChange | linux64-shippable-qr | nocondprof warm webrender | 993.33 -> 1,156.67 |
| 6% | wikipedia | loadtime | macosx1014-64-shippable-qr | nocondprof warm webrender | 945.29 -> 1,001.62 |
Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the offending patch(es) will be backed out in accordance with our regression policy.
For more information on performance sheriffing please see our FAQ.
|
||
Comment 2•3 years ago
|
||
Set release status flags based on info from the regressing bug 1685078
|
Assignee | |
Comment 3•3 years ago
|
||
https://hg.mozilla.org/mozilla-central/rev/66d167aa16ea#l4.139
Had no equivalent in the new code. This seems important for Wikipedia's
visual metrics.
|
|
Assignee | |
Updated•3 years ago
|
Pushed by ealvarez@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/6083cdb61f42 Restore bullet image load priority boost which was lost in bug 1685078. r=jrmuizel
|
||
Comment 5•3 years ago
|
||
| bugherder | ||
|
|
Assignee | |
Comment 6•3 years ago
|
||
There seemed to be some high-confidence progressions in my try push but most of the graphs there remained flat, so taking another look at this.
|
|
||
Updated•3 years ago
|
|
|
Assignee | |
Updated•3 years ago
|
|
|
Assignee | |
Comment 7•3 years ago
|
||
Do you know how can I see a before / after profile / captures? I wonder why this only affected Linux, too... Maybe the test is bi-modal and now the load event fires the same way that on other platforms or something.
|
Reporter | |
Comment 8•3 years ago
|
||
to create gecko profile you just select the job and click the "*** > Create gecko profile" button in treeherder
from the graph the test is not bimodal
|
|
Assignee | |
Comment 9•3 years ago
|
||
So I'm looking at this profile, but there don't seem to be any symbols?
|
|
Reporter | |
Comment 10•3 years ago
|
||
:Sparky do you know anything about the profiles generated by browsertime?
|
||
Comment 11•3 years ago
|
||
When the patch in this bug lands, you'll be able to generate good profiles - there were some issues with symbolication and profile organization that were fixed: https://bugzilla.mozilla.org/show_bug.cgi?id=1686327
|
|
Reporter | |
Comment 12•3 years ago
|
||
:emilio as of :spaky patch we don't have gecko profiles generation out of the box for this alert
I think you can cherrypick and run some try builds to get the appropiate data
|
|
Assignee | |
Comment 14•3 years ago
|
||
It's on my list of things to investigate, but this week and the next I'm basically on PTO (university exams). We can revert the regressing bug on beta meanwhile, so that the regression is scoped to nightly for now.
|
|
Assignee | |
Comment 15•3 years ago
|
||
This will give me more time to investigate.
Approval Request Comment
[Feature/Bug causing the regression]: bug 1686220
[User impact if declined]: performance regression
[Is this code covered by automated tests?]: yes
[Has the fix been verified in Nightly?]: n/a
[Needs manual test from QE? If yes, steps to reproduce]: n/a
[List of other uplifts needed for the feature/fix]: none
[Is the change risky?]: no
[Why is the change risky/not risky?]: cleanish backout (only test expectations needed merge)
[String changes made/needed]: none
|
||
Comment 16•3 years ago
|
||
Comment on attachment 9199912 [details]
Back out bug 1686220 from beta.
Back out bug 1686220 for causing a perf regression, approved for our next 86 beta, thanks!
|
||
Comment 17•3 years ago
|
||
I guess you mean back out bug 1685078. Bug 1686220 is this one.
|
|
||
Comment 18•3 years ago
|
||
| bugherder uplift | ||
|
|
||
Updated•3 years ago
|
|
|
Assignee | |
Comment 19•3 years ago
|
||
So, for my own, reference. Baseline (central minus the regressing bug):
- https://treeherder.mozilla.org/jobs?repo=try&revision=653fea65c9e2dc5fd46a161319470cd45941e76d (profiles).
Central:
|
|
Assignee | |
Comment 20•3 years ago
|
||
Comparison that shows the regression: https://treeherder.mozilla.org/perfherder/comparesubtest?originalProject=try&newProject=try&newRevision=59280cf544e217e6ca69e680913028750fa5f331&originalSignature=3072456&newSignature=3072456&framework=10&originalRevision=653fea65c9e2dc5fd46a161319470cd45941e76d
|
|
Assignee | |
Comment 21•3 years ago
|
||
Invalidating paint on size available is too soon (we can't decode the
image anyways). Bullet frames didn't use to do this and just reflowed.
Scheduling a paint causes us to do layout way too soon when there's more
important stuff to do for the load of the page.
This is a big improvement on load times and visual metrics in a variety
of websites:
|
|
Assignee | |
Comment 22•3 years ago
|
||
Took a little bit to confirm what was going on, but I think it was worth it!
|
||
Comment 23•3 years ago
|
||
Pushed by ealvarez@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/8591dc31ccda Invalidate paint on image load, not on size available. r=tnikkel
|
|
Reporter | |
Comment 24•3 years ago
|
||
Backout did its thing:
== Change summary for alert #28633 (as of Thu, 04 Feb 2021 18:24:59 GMT) ==
Improvements:
| Ratio | Suite | Test | Platform | Options | Absolute values (old vs new) |
|---|---|---|---|---|---|
| 23% | wikipedia | loadtime | linux64-shippable-qr | nocondprof warm webrender | 778.04 -> 601.54 |
| 18% | wikipedia | fcp | linux64-shippable-qr | nocondprof warm webrender | 763.29 -> 626.42 |
| 16% | wikipedia | linux64-shippable-qr | nocondprof warm webrender | 822.71 -> 694.86 | |
| 15% | wikipedia | LastVisualChange | linux64-shippable-qr | nocondprof warm webrender | 1,133.33 -> 960.00 |
| 7% | wikipedia | loadtime | macosx1014-64-shippable-qr | nocondprof warm webrender | 985.92 -> 921.08 |
For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=28633
|
|
||
Comment 25•3 years ago
|
||
| bugherder | ||
|
||
Updated•3 years ago
|
Description
•