1734558 - 36 - 3.64% google-slides dcf / google-slides PerceptualSpeedIndex + 12 more (Android, Linux, OSX, Windows) regression on Wed September 29 2021

Status: RESOLVED WONTFIX

(Core :: Layout, defect)

Description

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

Perfherder has detected a browsertime performance regression from push e5a8c52f0f187ba8c5e70b060b4b19a39d7f193b. 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)
36%	google-slides dcf	linux1804-64-shippable-qr	cold webrender	1,085.29 -> 1,476.00
36%	google-slides dcf	linux1804-64-shippable-qr	cold fission webrender	1,070.56 -> 1,451.21
21%	youtube dcf	linux1804-64-shippable-qr	warm webrender	968.10 -> 1,171.38
15%	netflix dcf	linux1804-64-shippable-qr	warm webrender	408.02 -> 470.88
11%	youtube ContentfulSpeedIndex	linux1804-64-shippable-qr	cold webrender	1,270.67 -> 1,413.00
11%	youtube ContentfulSpeedIndex	linux1804-64-shippable-qr	warm webrender	780.67 -> 866.42
10%	web-de dcf	android-hw-g5-7-0-arm7-shippable-qr	warm webrender	732.29 -> 804.75
6%	google-docs SpeedIndex	linux1804-64-shippable-qr	cold webrender	1,796.88 -> 1,910.58
6%	youtube ContentfulSpeedIndex	macosx1015-64-shippable-qr	cold webrender	1,475.79 -> 1,568.33
6%	google-docs SpeedIndex	linux1804-64-shippable-qr	cold webrender	1,790.92 -> 1,900.33
4%	google-docs SpeedIndex	macosx1014-64-shippable-qr	cold webrender	2,067.58 -> 2,158.92
4%	youtube ContentfulSpeedIndex	windows10-64-shippable-qr	cold webrender	1,252.50 -> 1,305.83
4%	youtube loadtime	linux1804-64-shippable-qr	warm webrender	1,235.67 -> 1,285.62
4%	google-slides PerceptualSpeedIndex	linux1804-64-shippable-qr	cold webrender	1,018.96 -> 1,056.08

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
25%	google-docs-canvas LastVisualChange	windows10-64-shippable-qr	cold webrender	1,789.67 -> 1,339.92
20%	google-docs-canvas LastVisualChange	linux1804-64-shippable-qr	cold webrender	1,800.00 -> 1,433.33
9%	yahoo-mail ContentfulSpeedIndex	macosx1014-64-shippable-qr	fission warm webrender	537.12 -> 486.42
9%	google-docs-canvas PerceptualSpeedIndex	windows10-64-shippable-qr	cold webrender	1,049.08 -> 952.67
7%	google-docs-canvas LastVisualChange	macosx1015-64-shippable-qr	cold webrender	1,705.00 -> 1,580.00
...	...	...	...	...
4%	google-docs-canvas SpeedIndex	linux1804-64-shippable-qr	cold webrender	1,071.42 -> 1,032.83

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

[Chris Peterson [:cpeterson]]

Randell, do you think Daniel's "Don't tick during the process-switch period" fix (bug 1712123) could be responsible for these regressions? Daniel's patch landed on September 29. On the same day, Nika landed her bfcache process switch fix for bug 1731792, which was blamed for similar Google Slides regressions in bug 1734118.

Comment 2

[Daniel Holbert [:dholbert]]

I would bet cpeterson's theory (that this is actually fallout from bug 1712123) is correct.

But I'd also be happy to have my bug 1712123 patch backed out, to test the theory of whether it caused this regression. Nothing will break if my patch is backed out; it's just a minor tweak to disable an eager "try-to-paint-content-one-frame-sooner" behavior, during pageload.

Comment 3

[Chris Peterson [:cpeterson]]

Randell thinks we can accept these performance results and resolve this bug as wontfix. Some sites are faster, some are slower. He said the YouTube regressions are a little concerning, but he looked at test videos and the issue appears to be slower text loading, not affecting video loading.