1772638 - Do eager major GCs after draining the microtask queue if approaching a GC threshold

Status: NEW

(Core :: JavaScript: GC, enhancement, P2)

Description

[Steve Fink [:sfink] [:s:]]

This is a forked version of bug 1661293 since it triggered some regressions, and I wanted to land everything in the stack that doesn't regress. So this bug is for actually running eager major GCs when the microtask queue is empty.

Comment 1

[Steve Fink [:sfink] [:s:]]

Attachment: Bug 1772638 - make nsJSContext::MaybePokeGC trigger EAGER_ALLOC_TRIGGER major GCs

Comment 2

[Steve Fink [:sfink] [:s:]]

The original regression doesn't seem to be happening anymore, even when retriggering the old versions. Looking closer, it seems like it was always somewhat bimodal. I'm re-landing at the beginning of a cycle, fully expecting some regressions to pop up, so that I have more to go off of (and may end up accepting any regressions if they appear to be test artifacts.)

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/8f85cd030fce
make nsJSContext::MaybePokeGC trigger EAGER_ALLOC_TRIGGER major GCs

Comment 4

[bszekely]

https://hg.mozilla.org/mozilla-central/rev/8f85cd030fce

Comment 5

[Acasandrei Beatrice (needinfo me)]

Perfherder has detected an AWSY performance regression, but I was unable to determine correctly the culprit as jobs fail to run on push 8f85cd030fce. Can you let me know if it is possible that this push could have caused the regression?

Comment 6

[Steve Fink [:sfink] [:s:]]

(In reply to Acasandrei Beatrice (needinfo me) from comment #5)

Perfherder has detected an AWSY performance regression, but I was unable to determine correctly the culprit as jobs fail to run on push 8f85cd030fce. Can you let me know if it is possible that this push could have caused the regression?

It definitely could.

Comment 7

[Noemi Erli[:noemi_erli]]

Backed out changeset 8f85cd030fce (Bug 1772638) for causing size and perf regressions

https://hg.mozilla.org/integration/autoland/rev/af7f9589f48c497f0dab327c0de70b73d9325c6e

Comment 8

[Iulian Moraru]

Backout merged to central: https://hg.mozilla.org/mozilla-central/rev/af7f9589f48c

Comment 9

[Acasandrei Beatrice (needinfo me)]

== Change summary for alert #35023 (as of Sat, 06 Aug 2022 02:33:26 GMT) ==

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
10%	Explicit Memory	windows10-64-2004-shippable-qr	fission tp6	592,107,090.38 -> 534,423,324.35
8%	Explicit Memory	macosx1015-64-shippable-qr	fission tp6	613,495,821.88 -> 563,866,854.06
8%	Resident Memory	windows10-64-2004-shippable-qr	fission tp6	798,654,561.25 -> 736,476,142.33
5%	Heap Unclassified	macosx1015-64-shippable-qr	fission tp6	141,412,503.95 -> 134,305,710.70
5%	Resident Memory	macosx1015-64-shippable-qr	fission tp6	988,970,125.55 -> 940,276,018.06
5%	Heap Unclassified	windows10-64-2004-shippable-qr	fission tp6	92,835,584.23 -> 88,621,576.70

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=35023