1927103 - 11.64 - 11.44% perf_reftest style-attr-1.html / perf_reftest_singletons style-attr-1.html (Linux) regression on Tue October 22 2024

Status: NEW

(Core :: IPC, defect)

Description

[Treeherder Bug Filer]

Perfherder has detected a talos performance regression from push c25e12d0f6699e7fc0b08fe4f31c0bdb36276a6b. 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)
12%	perf_reftest style-attr-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	2.81 -> 3.14
11%	perf_reftest_singletons style-attr-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	3.39 -> 3.77

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
34%	cpstartup content-process-startup	linux1804-64-shippable-qr	e10s fission stylo webrender-sw	75.17 -> 49.75
31%	cpstartup content-process-startup	linux1804-64-shippable-qr	e10s fission stylo webrender	73.29 -> 50.83
7%	tabpaint	linux1804-64-shippable-qr	e10s fission stylo webrender-sw	60.77 -> 56.75

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 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.

You can run all of these tests on try with ./mach try perf --alert 2654

The following documentation link provides more information about this command.

For more information on performance sheriffing please see our FAQ.

If you have any questions, please do not hesitate to reach out to [e-mail].

Comment 1

[:gerard-majax]

You filed mentionniong c25e12d0f6699e7fc0b08fe4f31c0bdb36276a6b but marking bug 1918538, so either it's https://hg.mozilla.org/integration/autoland/rev/9b74bcc1a5a10a584162ef5124e8cd1bffb7c17d and bug 1918538 is the correct one, or it should be against bug 1874689

Comment 2

[Acasandrei Beatrice (needinfo me)]

I am uncertain which of the identified bugs (Bug 1874689 or Bug 1918538 ) has led to this regression and I need some help from the authors to identify the right culprit. Since the push contains two revisions c25e12d0f6699e7fc0b08fe4f31c0bdb36276a6b and 9b74bcc1a5a10a584162ef5124e8cd1bffb7c17d I can't pinpoint for certain the right culprit. Please make any necessary edits to the Regressed by field.

Comment 3

[:gerard-majax]

(In reply to Acasandrei Beatrice (needinfo me) from comment #2)

I am uncertain which of the identified bugs (Bug 1874689 or Bug 1918538 ) has led to this regression and I need some help from the authors to identify the right culprit. Since the push contains two revisions c25e12d0f6699e7fc0b08fe4f31c0bdb36276a6b and 9b74bcc1a5a10a584162ef5124e8cd1bffb7c17d I can't pinpoint for certain the right culprit. Please make any necessary edits to the Regressed by field.

Out of the blue I would say bug 1874689, but who knows what can happen here? Can you backfill only 9b74bcc1a5a10a584162ef5124e8cd1bffb7c17d and check?

Comment 4

[Acasandrei Beatrice (needinfo me)]

Due to both revisions being included in the same push it's impossible to just backfill 9b74bcc1a5a10a584162ef5124e8cd1bffb7c17d, as far as I am aware. We can only retrigger jobs from a push, not choose a particular revision from the push.

Comment 5

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

Unfortunately this seems to reproduce on Try, changing only the fork server pref: style-attr-1.html regresses by 12% and some-descendants-1.html by 6%, but also bidi-resolution-1.html improves by 26% (!) and all of these have “high” confidence after 25 runs of both revisions.

I also tried the perf_reftest_singletons suite, and that has a number of changes listed with “high” confidence, both positive and negative, and mostly different from the perf_reftest results except that style-attr-1 is shown as regressing on both.

I'll note that these are artificial microbenchmarks of what seem to be unrealistic inputs (like a CSS rule with 100000 copies of the same selector). On tests of realistic workloads, from what I've seen, enabling the fork server is at worst neutral and in some cases a significant improvement, especially for RAM use but also for total CPU time consumed, and possibly even user-visible latency of page load given the tabpaint improvement mentioned in this perf alert.

I'm not an expert in browser performance but it seems to me that these perf changes are mostly just random and, to the extent they're not, are an acceptable tradeoff.

Comment 6

[Donal Meehan [:dmeehan]]

:jld based on Comment 5 will there be any further investigation or are you thinking of resolving this as wontfix?