1764502 - 26.88 - 22.51% reddit-billgates-post-1.posts ContentfulSpeedIndex / reddit-billgates-post-1.posts ContentfulSpeedIndex + 1 more (Linux) regression on Fri April 8 2022

Status: NEW

(Core :: Widget: Gtk, defect)

Description

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

Perfherder has detected a browsertime performance regression from push 498e0e6f9b19107c8c17525a3d01a659f89002e8. 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)
27%	reddit-billgates-post-1.posts ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	530.04 -> 672.54
26%	reddit-billgates-post-1.comments ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	494.33 -> 623.50
23%	reddit-billgates-post-1.posts ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	530.79 -> 650.29

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
8%	instagram LastVisualChange	linux1804-64-shippable-qr	cold fission webrender	1,825.00 -> 1,676.67
8%	outlook LastVisualChange	linux1804-64-shippable-qr	cold fission webrender	3,460.80 -> 3,183.33
8%	office LastVisualChange	linux1804-64-shippable-qr	cold fission webrender	1,911.67 -> 1,760.00
6%	buzzfeed LastVisualChange	linux1804-64-shippable-qr	cold fission webrender	1,256.67 -> 1,183.33
6%	wikia ContentfulSpeedIndex	linux1804-64-shippable-qr	fission warm webrender	847.29 -> 799.83

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

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

Set release status flags based on info from the regressing bug 1756903

Comment 2

[Emilio Cobos Álvarez (:emilio)]

Any chance of having the side-by-side video? It's not the first time I've got a ContentFulSpeedIndex regression that happened to be an improvement (bug 1760551 for example).

Comment 3

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

Unfortunately the side-by-side doesn't work on responsiveness tests, only for the pageload ones. :(

Comment 4

[Emilio Cobos Álvarez (:emilio)]

How should I investigate more? Given there's a regression in one page and a progression in 5 I'm tempted to not dig a lot into this, but it'd be nice to confirm that it's not too worse, or that it's an instance of the metric bug mentioned in comment 2...

Comment 5

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

Wait for it... might be more alerts incoming.

Comment 6

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

Attachment: cold-side-by-side.mp4

Comment 7

[Andrej (:aglavic)]

(In reply to Alexandru Ionescu (needinfo me) [:alexandrui] from comment #5)

Wait for it... might be more alerts incoming.

== Change summary for alert #33870 (as of Wed, 20 Apr 2022 19:16:32 GMT) ==

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
1241%	tp5o XRes	linux1804-64-shippable-qr	e10s fission stylo webrender	197,763.34 -> 2,652,787.17
1240%	tp5o_webext XRes	linux1804-64-shippable-qr	e10s fission stylo webrender	197,766.34 -> 2,649,784.62
1240%	tp5o_webext XRes	linux1804-64-shippable-qr	e10s fission stylo webrender	197,783.90 -> 2,649,709.04
27%	tresize	linux1804-64-shippable-qr	e10s fission stylo webrender	16.97 -> 21.51
14%	tp5o_webext XRes	linux1804-64-shippable-qr	e10s fission stylo webrender-sw	6,483,870.96 -> 7,367,183.11
14%	tp5o XRes	linux1804-64-shippable-qr	e10s fission stylo webrender-sw	6,486,447.92 -> 7,366,145.96

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=33870

Comment 8

[Ryan VanderMeulen [:RyanVM]]

The tp5o regressions from comment 7 appears to be holding steady - any chance you can take a look?

Comment 9

[Emilio Cobos Álvarez (:emilio)]

I'm a bit swamped with other things right now, but this code path was specific of some desktop environments; on everything but gnome/unity on X11, we were already using client decorations.

Martin, do you have a strong opinion / any idea of what might be going on there? We can back out to restore the baseline but that'd be unfortunate (and cause some glitches on the corners on GNOME+X11).

Comment 10

[Martin Stránský [:stransky] (ni? me)]

Is there any way how to run the tests locally?

I think it may be related to XWindow content sync. In CSD/X11 mode we may use two XWindows, one for toplevel window (mShell) and one for content (MozContainer) while system mode uses only one XWindow for both (mShell/MozConatiner) - MozContainer is used without XWindow. So there may be a delay in the rendering/presentation of nested MozContainer.

My ultimate goal is to remove MozContainer as windowed widget on X11 but that's a bit long term issue and needs some Wayland updates to attach wl_subsurface to mShell.

Comment 11

[Martin Stránský [:stransky] (ni? me)]

It would help to know if that's SW or HW rendering. But it looks like delayed painting, not a performance issue. Will be interesting to see if there's any video playback / WebGL rendering regression.

Comment 12

[Pascal Chevrel:pascalc]

Alexandru, could you answer Martin's questions please? Thanks

Comment 13

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

there's no recording available for talos.

Comment 14

[Marco Castelluccio [:marco]]

There were a few more questions from Martin:

• Is there any way how to run the tests locally?
• Is it using SW or HW rendering?

Comment 15

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

(In reply to Marco Castelluccio [:marco] from comment #14)

There were a few more questions from Martin:

• Is there any way how to run the tests locally?

https://firefox-source-docs.mozilla.org/testing/perfdocs/talos.html#running-talos-locally

• Is it using SW or HW rendering?

I think the ones that have sw in Options column use software rendering (e.g. webrender-sw) and the others hardware.