1851578 - 26 - 19% nytimes ContentfulSpeedIndex / pinterest FirstVisualChange + more (Windows, Linux, Android, OSX ) regression on Tue August 29 2023

Status: RESOLVED INVALID

(Core :: DOM: Core & HTML, defect)

Description

[Andra Esanu (needinfo me)]

Perfherder has detected a browsertime performance regression from push 0b374bcf3727e39fb3a5bbcc66005c9dc4690f8d. 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)	Performance Profiles
26%	nytimes ContentfulSpeedIndex	windows10-64-shippable-qr	fission warm webrender	281.34 -> 355.82
19%	pinterest FirstVisualChange	linux1804-64-shippable-qr	cold fission webrender	696.89 -> 830.78
19%	pinterest fcp	linux1804-64-shippable-qr	cold fission webrender	642.70 -> 763.83
13%	imdb ContentfulSpeedIndex	android-hw-a51-11-0-aarch64-shippable-qr	cold webrender	2,472.08 -> 2,798.38
13%	nytimes fcp	windows10-64-shippable-qr	cold fission webrender	329.27 -> 371.17
12%	nytimes FirstVisualChange	windows10-64-shippable-qr	cold fission webrender	363.39 -> 407.69
12%	pinterest SpeedIndex	macosx1015-64-shippable-qr	fission warm webrender	331.65 -> 371.17	Before/After
10%	pinterest ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	1,105.71 -> 1,213.02
9%	buzzfeed fcp	windows10-64-shippable-qr	cold fission webrender	569.00 -> 618.82
9%	buzzfeed fcp	linux1804-64-shippable-qr	cold fission webrender	714.04 -> 775.02
...	...	...	...	...	...
5%	linkedin fcp	macosx1015-64-shippable-qr	cold fission webrender	395.43 -> 416.65	Before/After
5%	expedia LastVisualChange	linux1804-64-shippable-qr	fission warm webrender	1,366.68 -> 1,432.20
4%	netflix SpeedIndex	linux1804-64-shippable-qr	fission warm webrender	855.97 -> 893.63
4%	netflix PerceptualSpeedIndex	linux1804-64-shippable-qr	fission warm webrender	736.14 -> 767.90
2%	nytimes SpeedIndex	windows10-64-shippable-qr	fission warm webrender	1,005.02 -> 1,027.34

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)	Performance Profiles
31%	linkedin loadtime	windows10-64-shippable-qr	cold fission webrender	2,177.42 -> 1,513.38
25%	twitch loadtime	windows10-64-shippable-qr	cold fission webrender	1,662.34 -> 1,241.32
24%	linkedin loadtime	windows10-64-shippable-qr	fission warm webrender	1,303.12 -> 986.86
11%	linkedin loadtime	macosx1015-64-shippable-qr	fission warm webrender	1,034.76 -> 917.82	Before/After
11%	linkedin ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	2,188.23 -> 1,941.16
...	...	...	...	...	...
3%	facebook-cristiano loadtime	android-hw-a51-11-0-aarch64-shippable-qr	warm webrender	667.65 -> 650.78

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) 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.

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 1845668

Comment 2

[Tooru Fujisawa [:arai]]

These patches can change the ordering of JS execution and layout.

Most of regressions/improvements come from the bug 1845668 patch part 1, which changes how main thread task gets resumed when off-thread compilation finishes:
https://treeherder.mozilla.org/perfherder/compare?originalProject=try&originalRevision=6451393187384a4f56faf8024e8c22c2b54e7dde&newProject=try&newRevision=f235bdba461a5fe1515a43d74fbc23f5c2dcbee1&framework=13&page=3&showOnlyImportant=1

The regressions are mostly in visual part, which frequently happens if there's race between JS and other resources. I'll verify that this and next week.
The improvements are in loadtime, which also need investigation why that happened.
Anyway I think these are not serious things that needs backout.

Comment 3

[Tooru Fujisawa [:arai]]

I don't observe the improvement or regression on local run in any suite.

Then, in nytimes case, there's certainly race between resources (JS, font, image, etc), and the visual score depends on which one becomes available first.

(apart from the benchmark score, I noticed that it's better separating the [A,B] and [C,D] range explained in https://phabricator.services.mozilla.com/D184897#6113651 from the off-thread compilation/decode marker, given [C,D] part dominate the time, while it's mostly just waiting on the main-thread's other tasks. of course it's still nice to visualize the range tho)

Comment 4

[Donal Meehan [:dmeehan]]

Tracking for Fx119 for a resolution on the investigation - RE: comment 2 and comment 6

Comment 5

[Denis Palmeiro [:denispal]]

Attachment: Screenshot from 2023-09-12 22-05-53.png

I'm looking at the time series results and I don't see any notable regressions at all. I would have expected to see the 30% regression in CSI for nytimes but I don't see any real movement from the attached graph. I think perhaps this was incorrectly marked due to noise.

I also tried to reproduce this on try and could not. I think I'm going to mark this as invalid, but please reopen if there is more evidence. Thanks!