1885376 - 6.2% ts_paint_webext (Windows) regression on Thu March 7 2024

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect)

Description

[Andrej (:aglavic)]

Perfherder has detected a talos performance regression from push db166f0115c4eb3a6a37cc2e1218293acc7f074b. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
7%	ts_paint	windows10-64-shippable-qr	e10s fission stylo webrender	430.67 -> 461.17
7%	ts_paint_webext	windows10-64-shippable-qr	e10s fission stylo webrender	433.00 -> 461.67
6%	ts_paint_webext	windows10-64-shippable-qr	e10s fission stylo webrender	433.79 -> 460.67

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 patch(es) may be backed out in accordance with our regression policy.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a sheriff to do that for you.

You can run these tests on try with ./mach try perf --alert 41787

For more information on performance sheriffing please see our FAQ.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1882358

Comment 2

[Brad Werth [:bradwerth]]

My theory is that this is happening because ts_paint is measuring the time it takes to complete the first paint submission calls, rather than measuring the time before the first paint completes. In other words, if the first paint is async, ts_paint as currently calculated will be lower. If this is correct analysis, it'll be easy to undo the ts_paint regression by no longer forcing the first flush to be sync (as was done in D202869). But really, we need to fix what ts_paint is measuring, which I will open another Bug to accomplish, presuming this fix works.

Comment 3

[Brad Werth [:bradwerth]]

Attachment: Bug 1885376: Allow first scene flush to be async.

This recovers most of the performance degradation of ts_paint, which
measures time to first MozAfterPaint event received. It's not critical
that the first flush be sync, so let's keep things flowing in the way
that ts_paint prefers.

Comment 4

[Timothy Nikkel (:tnikkel)]

Looks like this is what ts_paint runs

https://searchfox.org/mozilla-central/source/testing/talos/talos/startup_test/tspaint_test.html

It gets the firstPaint value from the startupinfo object via getStartupInfo.

That object should contain two fields

https://searchfox.org/mozilla-central/rev/04f7743d94691fa24212fb43099f9d84c3bfc890/toolkit/components/startup/StartupTimeline.h#18

firstPaint and firstPaint2. These are recorded

https://searchfox.org/mozilla-central/rev/04f7743d94691fa24212fb43099f9d84c3bfc890/view/nsViewManager.cpp#293

and

https://searchfox.org/mozilla-central/rev/04f7743d94691fa24212fb43099f9d84c3bfc890/view/nsView.cpp#1095

respectively. So firstPaint2 measures the time of the end of the first composite. I think that is what you are asking for here.

I created firstPaint2 years ago in bug 1556568 presumably due to a similar issue. After that I was hoping that we would have observed firstPaint2 long enough to validate that it was working as expected and then switch over our relevant perf metrics to it. And bug 1654398 got filed to switch ts_paint to the new metric, but I guess it never got picked it up.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/b0450657a485
Allow first scene flush to be async. r=nical

Comment 6

[Natalia Csoregi [:nataliaCs]]

https://hg.mozilla.org/mozilla-central/rev/b0450657a485

Comment 7

[pstanciu]

Backed out for causing multiple regressions.

• Backout link

Comment 8

[(PTO June 19-July 7) Ryan VanderMeulen]

Backed out from Beta for 125.0b2 also.
https://hg.mozilla.org/releases/mozilla-beta/rev/1f815868a897599dc211b55fffec34305bd5c020

Comment 9

[Brad Werth [:bradwerth]]

Resolved by backout.