1618868 - Intermittent webgl mochitest failure increase when remote canvas enabled.

Status: RESOLVED FIXED

(Core :: Graphics, defect, P1)

Description

[Bob Owen (:bobowen)]

The first patch for bug 1598585 causes an increase in intermittent webgl test failures.

This patch changed the canvas thread handling, so my guess is that this is causing a hang somehow in the GPU process:
https://hg.mozilla.org/try/rev/6b300b00fc72832c6bb183760539c4b059db8479

No idea as to why webgl tests trigger this more than others, perhaps more of those tests also interact with canvas 2D as well.

Try push with remote canvas enabled after that patch:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=3dac49a7bacd9ed57d7eb1f9b5ff86dc8ff01db9

Try push with remote canvas enabled before that patch:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=fb9dcddb8cdde60636c45f161768636217c8a687

Comment 1

[Bob Owen (:bobowen)]

One of the crashes I sometimes see in the webgl tests is the parent process waiting for PCompositorBridgeChild::SendWillClose.
Another I'd missed in a reftest was the parent waiting for PImageBridgeChild::SendWillClose.

So this seems to suggest a hang in the compositor thread of the GPU process.

Comment 2

[Bob Owen (:bobowen)]

I've found the race on Canvas thread shutdown in Debug builds.
Try push with potential fix:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=c60316af852fca77fd19b225652f1c688a46ce25

Comment 3

[Bob Owen (:bobowen)]

https://treeherder.mozilla.org/#/jobs?repo=try&duplicate_jobs=visible&selectedJob=296840222&revision=6bfdacd4031a8ad91d7b7fe824ef7723524a93b3

Comment 4

[Bob Owen (:bobowen)]

Attachment: Bug 1618868 Part 1: Call CanvasTranslators() before CanvasThreadHolder::ReleaseOnCompositorThread to avoid race. r=mattwoodrow!

Comment 5

[Bob Owen (:bobowen)]

Attachment: Bug 1618868 Part 2: Make counts and states in CanvasEventRingBuffer footers SequentiallyConsistent. r=jrmuizel!

This fixes an issue with the AboutToWait check. It is possible that this could
be done without this, but while there might be a very slight performance hit, it
seems to be smaller than the general noise in the tests.

Depends on D70335

Comment 6

[Bob Owen (:bobowen)]

Attachment: Bug 1618868 Part 3: Fix failure test race in CanvasEventRingBuffer::ReturnRead. r=jrmuizel!

This also adds checks for the other side closing during the ReturnRead and
ReturnWrite loops.

Depends on D70336

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/f2ade35d7588
Part 1: Call CanvasTranslators() before CanvasThreadHolder::ReleaseOnCompositorThread to avoid race. r=mattwoodrow
https://hg.mozilla.org/integration/autoland/rev/0ab9d2f38280
Part 2: Make counts and states in CanvasEventRingBuffer footers SequentiallyConsistent. r=jrmuizel
https://hg.mozilla.org/integration/autoland/rev/6d5126a36758
Part 3: Fix failure test race in CanvasEventRingBuffer::ReturnRead. r=jrmuizel

Comment 8

[Cristina Coroiu [:ccoroiu]]

https://hg.mozilla.org/mozilla-central/rev/f2ade35d7588
https://hg.mozilla.org/mozilla-central/rev/0ab9d2f38280
https://hg.mozilla.org/mozilla-central/rev/6d5126a36758

Comment 9

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

Since the status are different for nightly and release, what's the status for beta?
For more information, please visit auto_nag documentation.