1776668 - 23.55 - 4.09% expedia ContentfulSpeedIndex / google-mail loadtime + 18 more (Linux) regression on Fri June 24 2022

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect, P2)

Description

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

Perfherder has detected a browsertime performance regression from push a411b158a51ee4b5a165402b863f758ef51f070c. 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)
24%	expedia ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	450.42 -> 556.50
11%	google-docs-canvas loadtime	linux1804-64-shippable-qr	cold fission webrender	1,733.42 -> 1,931.54
11%	cnn fcp	linux1804-64-shippable-qr	fission warm webrender	316.29 -> 351.75
11%	espn loadtime	linux1804-64-shippable-qr	fission warm webrender	601.67 -> 666.33
10%	cnn fcp	linux1804-64-shippable-qr	cold fission webrender	774.75 -> 856.04
10%	expedia ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	449.17 -> 496.08
10%	google-mail loadtime	linux1804-64-shippable-qr	cold fission webrender	1,083.23 -> 1,189.25
8%	cnn PerceptualSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	910.33 -> 985.42
7%	google-docs fcp	linux1804-64-shippable-qr	fission warm webrender	390.25 -> 418.00
7%	google-docs fcp	linux1804-64-shippable-qr	cold fission webrender	751.17 -> 801.79
...	...	...	...	...
5%	google-docs FirstVisualChange	linux1804-64-shippable-qr	cold fission webrender	756.67 -> 793.33
5%	google-docs SpeedIndex	linux1804-64-shippable-qr	cold fission webrender	1,142.75 -> 1,196.92
4%	google-docs PerceptualSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	1,405.33 -> 1,465.75
4%	google-docs ContentfulSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	1,524.00 -> 1,587.58
4%	google-mail loadtime	linux1804-64-shippable-qr	fission warm webrender	399.00 -> 415.33

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.

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

[Hsin-Yi Tsai (she/her) [:hsinyi]]

Given the suspicious bug 1776205 was backed out, is this issue still valid? Thank you.

Comment 2

[Nicolas B. Pierron [:nbp]]

(In reply to Hsin-Yi Tsai (Fx104 REO) [:hsinyi] from comment #1)

Given the suspicious bug 1776205 was backed out, is this issue still valid? Thank you.

I presume this is no longer an issue, the back-out should have restored the performance as they were previously, as mentioned in the bug as well.
I suspect this bug can be closed now.

Comment 3

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

Set release status flags based on info from the regressing bug 1776205

Comment 4

[Ryan VanderMeulen [:RyanVM]]

Bug 1776205 re-landed and AFAICT the regressions are back (looking at the graphs from the alert).

Comment 5

[Nicolas B. Pierron [:nbp]]

Before this patch inline scripts are supposed to use on-demand delazification.

After this patch inline script are supposed to use on-demand delazification, but if one of the condition ahead fails, then it will use on-demand delazification, and if the preference which is set to parsing-everything eagerly is set then we use on-demand delazification. Only in the corner case where the preference is set to some value such as concurrent delazification or on-demand delazication, then we will use this delazification strategy for inline scripts.

So … either the inline script are playing too much of a role in the code budget, while one might expect them to be small, and are causing eagerly parsed script to be processed on-demand. Or there is something I do not understand …

I will make a patch which moves the check for inline script sooner, before counting inline script as part of the source text content.
As well as adding a preference to later skip this check.

Comment 6

[Nicolas B. Pierron [:nbp]]

Attachment: Bug 1776668 - ApplyDelazifyStrategy: Skip inline script sooner.

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/f3cd67fc1a50
ApplyDelazifyStrategy: Skip inline script sooner. r=arai

Comment 8

[Nicolas B. Pierron [:nbp]]

If the current patch does not work, then this would mean that I my initial assumption, that this code path is only used by inlined scripts, is incorrect.
In which case I could make another fix to identify the caller more precisely and not just filter out inline scripts as implemented today.

Comment 9

[Cristian Tuns]

https://hg.mozilla.org/mozilla-central/rev/f3cd67fc1a50

Comment 10

[Dianna Smith [:diannaS]]

Backout for causing (bug 1780426) in fx104, it remains in fx105

https://hg.mozilla.org/releases/mozilla-beta/rev/74616c53a1187d7865f9b40fe1d36fdaea9c69f3

Comment 11

[Dianna Smith [:diannaS]]

Backout for fx105
https://hg.mozilla.org/mozilla-central/rev/624e6f2a2af30c9689c56a77eb0bf01a0a085c7d

Comment 12

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

== Change summary for alert #34949 (as of Thu, 28 Jul 2022 12:27:55 GMT) ==

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
5%	expedia LastVisualChange	linux1804-64-shippable-qr	cold fission webrender	1,773.33 -> 1,860.00
4%	expedia PerceptualSpeedIndex	linux1804-64-shippable-qr	cold fission webrender	1,054.38 -> 1,096.00

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
10%	google-docs loadtime	linux1804-64-shippable-qr	fission warm webrender	733.12 -> 659.25
9%	cnn PerceptualSpeedIndex	windows10-64-shippable-qr	bytecode-cached fission warm webrender	509.38 -> 465.25
9%	google-docs fcp	linux1804-64-shippable-qr	fission warm webrender	640.25 -> 584.96
6%	google-docs fcp	linux1804-64-shippable-qr	cold fission webrender	790.04 -> 741.58
5%	google-docs loadtime	windows10-64-shippable-qr	fission warm webrender	675.00 -> 639.29
...	...	...	...	...
4%	google-docs SpeedIndex	linux1804-64-shippable-qr	cold fission webrender	1,263.83 -> 1,210.42

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=34949

Comment 13

[Nicolas B. Pierron [:nbp]]

This issue is fixed by backing out patches.