1744207 - 82.93% reddit-billgates-post-2.hot LastVisualChange (Windows) regression on Sat November 27 2021

Status: RESOLVED WONTFIX

(Core :: Graphics: WebRender, defect)

Description

[Alexandru Ionescu (needinfo me) [:alexandrui]]

Perfherder has detected a browsertime performance regression from push 3fe9e1643aafa1f8f5ba19b012b90b0d39a422b5. 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)
83%	reddit-billgates-post-2.hot LastVisualChange	windows10-64-shippable-qr	cold webrender	1,008.29 -> 1,844.50

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 1

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

Set release status flags based on info from the regressing bug 1711061

Comment 2

[Andrew Osmond [:aosmond] (he/him)]

It is worth noting that Chrome and Chromium regressed as well on Windows, although not as much.

The post-2.billg and post-2.top variants are unchanged. The difference for images between top and hot are minimal, so I find it somewhat hard believe I could have reasonably regressed that particular variant that much.

The results began varying a lot for post-2.hot on mozilla-central on November 25, two days before my patch on Windows, but for Linux and Mac it lines up better.

What is strange is that visiting that reddit post / comments does not suggest it is particularly image heavy.

Comment 3

[Andrew Osmond [:aosmond] (he/him)]

I do wonder if bug 1743761 will help here, as it will cause us to repaint sooner when we get a substitute surface for an image. It could be something as simple as that.

Comment 4

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

Set release status flags based on info from the regressing bug 1711061

Comment 5

[Ryan VanderMeulen [:RyanVM]]

(In reply to Andrew Osmond [:aosmond] (he/him) from comment #3)

I do wonder if bug 1743761 will help here, as it will cause us to repaint sooner when we get a substitute surface for an image. It could be something as simple as that.

I don't see any change in the graphs since bug 1743761 landed. Can you please take another look at this, Andrew?

Comment 6

[Andrew Osmond [:aosmond] (he/him)]

I spent some time a while ago looking at the profiler traces and could not ascertain any reason besides bug 1743761 for the regression. Most of the metrics for this test case did not change. I think we have to live with this unless a better explanation appears.