1838200 - 737.04 - 7% perf_reftest style-sharing-style-attr.html / perf_reftest_singletons abspos-reflow-1.html + 18 more (Linux, OSX, Windows) regression on Thu June 8 2023

Status: RESOLVED FIXED

(Core :: CSS Parsing and Computation, defect)

Description

[Andrej (:aglavic)]

Perfherder has detected a talos performance regression from push b78c7333f0fdf9378eaf33b8ff1b60f4a9a69378. 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)
737%	perf_reftest style-sharing-style-attr.html	macosx1015-64-shippable-qr	e10s fission stylo webrender-sw	0.14 -> 1.13
616%	perf_reftest style-sharing.html	macosx1015-64-shippable-qr	e10s fission stylo webrender-sw	0.15 -> 1.06
616%	perf_reftest style-sharing-style-attr.html	macosx1015-64-shippable-qr	e10s fission stylo webrender	0.15 -> 1.07
533%	perf_reftest style-sharing.html	macosx1015-64-shippable-qr	e10s fission stylo webrender	0.16 -> 1.03
411%	perf_reftest style-sharing.html	macosx1015-64-shippable-qr	e10s fission stylo webrender	0.18 -> 0.93
279%	perf_reftest style-sharing.html	windows10-64-shippable-qr	e10s fission stylo webrender	0.30 -> 1.14
267%	perf_reftest style-sharing-style-attr.html	windows10-64-shippable-qr	e10s fission stylo webrender	0.35 -> 1.27
258%	perf_reftest style-sharing-style-attr.html	windows10-64-shippable-qr	e10s fission stylo webrender-sw	0.37 -> 1.33
241%	perf_reftest style-sharing.html	windows10-64-shippable-qr	e10s fission stylo webrender-sw	0.34 -> 1.15
34%	perf_reftest_singletons style-sharing.html	windows10-64-shippable-qr	e10s fission stylo webrender	3.40 -> 4.54
...	...	...	...	...
27%	perf_reftest_singletons style-sharing.html	macosx1015-64-shippable-qr	e10s fission stylo webrender	3.54 -> 4.49
25%	perf_reftest_singletons tiny-traversal-singleton.html	macosx1015-64-shippable-qr	e10s fission stylo webrender	587.12 -> 734.66
23%	perf_reftest_singletons tiny-traversal-singleton.html	windows10-64-shippable-qr	e10s fission stylo webrender	600.34 -> 739.67
9%	perf_reftest_singletons abspos-reflow-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	65.53 -> 71.58
7%	perf_reftest_singletons abspos-reflow-1.html	windows10-64-shippable-qr	e10s fission stylo webrender	49.34 -> 52.79

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
67%	perf_reftest style-sharing-style-attr.html	linux1804-64-shippable-qr	e10s fission stylo webrender-sw	1.76 -> 0.59
58%	perf_reftest style-sharing.html	linux1804-64-shippable-qr	e10s fission stylo webrender-sw	1.46 -> 0.61
57%	perf_reftest style-sharing-style-attr.html	linux1804-64-shippable-qr	e10s fission stylo webrender	1.46 -> 0.62

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 1835923

Comment 2

[Emilio Cobos Álvarez (:emilio)]

I still don't quite understand this regression (I would've expected the macOS and Windows behavior pattern to match Linux) but bug 1836948 should go back to at least the previous performance if not better. Will push to try to verify that.

Comment 3

[Timothy Nikkel (:tnikkel)]

I wonder if paint suppression is having an effect here? In another bug we found that some of these perf_reftest tests were being run before paint suppression ended.

Comment 4

[Emilio Cobos Álvarez (:emilio)]

Attachment: Bug 1838200 - Flush layout before running perf-reftest measurements. r=tnikkel

Otherwise we might not measure what we want, if we haven't unsuppressed
painting or flushed layout yet flushing style might be a no-op.

Comment 5

[Emilio Cobos Álvarez (:emilio)]

That is a great hint. Yeah, flushing style may not do what we want if we haven't flushed layout earlier.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/b040fff76ed7
Flush layout before running perf-reftest measurements. r=tnikkel,perftest-reviewers,kshampur

Comment 7

[Cristian Tuns]

https://hg.mozilla.org/mozilla-central/rev/b040fff76ed7