1725270 - Fission warmload regression from null principal isolation

Status: VERIFIED FIXED

(Core :: DOM: Navigation, defect)

Description

[Chris Peterson [:cpeterson]]

Jesup says:

I see major regressions in warmload in the last day or two for fission; if you look a the graphs at https://drive.google.com/drive/folders/1UvXyaWTMmrf8s6h0ZxGH7fkOnkeb_p2F?usp=sharing. You can see spikes for fission at the end. The average moves more slowly (of course), but the spikes are large enough I see it even in the averages (in the graphs, and at https://docs.google.com/spreadsheets/d/1dVXTypCkosetd5KyBbNP5vgWzxm4fr6fxhzbicx4Dyo/edit#gid=1965470853 (and see the summary tab as well, though the graphs are a work-in-progress)

quite visible in https://arewefastyet.com/linux64/warm-page-load/overview?numDays=90&series=Firefox,Firefox-Fission as well.

Nika says:

the regression is caused by an extra process switch for every intermediate about:blank load.

I think there was a plan to potentially add a workaround pref which could be set in perf tests to make these about:blank loads act in the way they used to so that the test continues running as expected

The about:blank is loaded in the web content process, rather than the microsoft.com content process or whatever, so there will be an extra process switch there

Comment 1

[Nika Layzell [:nika] (ni? for response)]

Attachment: Bug 1725270 - Add testing pref to get old about:blank process selection behavior, r=jesup

Adds a pref to perform about:blank document loads which have system
triggering principals within the existing process, rather than doing a
process switch. This can be used in performance tests which load
about:blank and do not expect process switches.

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/50435d7ca25a
Add testing pref to get old about:blank process selection behavior, r=jesup

Comment 4

[Noemi Erli[:noemi_erli]]

https://hg.mozilla.org/mozilla-central/rev/50435d7ca25a

Comment 5

[Chris Peterson [:cpeterson]]

This fix landed on 2021-08-13. Randell confirmed that the page load regressions went away on 2021-08-14,

Comment 6

[Andra Esanu (needinfo me)]

== Change summary for alert #30939 (as of Mon, 16 Aug 2021 11:17:31 GMT) ==

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
12%	about_preferences_basic	macosx1014-64-shippable-qr	e10s stylo webrender-sw	111.54 -> 98.56
11%	about_preferences_basic	windows10-64-shippable-qr	e10s stylo webrender	107.29 -> 95.17

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=30939

Comment 7

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

== Change summary for alert #30945 (as of Mon, 16 Aug 2021 17:11:47 GMT) ==

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
50%	welcome FirstVisualChange	windows10-64-shippable-qr	warm webrender	80.00 -> 40.00
36%	welcome loadtime	macosx1014-64-shippable-qr	warm webrender	37.33 -> 24.00
25%	welcome dcf	windows10-64-shippable-qr	warm webrender	27.71 -> 20.83
25%	welcome loadtime	windows10-64-shippable-qr	warm webrender	30.96 -> 23.29
23%	welcome dcf	linux1804-64-shippable-qr	warm webrender	35.06 -> 27.12
...	...	...	...	...
2%	welcome SpeedIndex	linux1804-64-shippable-qr	warm webrender	1,241.58 -> 1,211.67

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=30945