1895067 - Detaching a tab with canvas makes it completely blank for long period of time and meory rises continuously

Status: NEW

(Core :: Graphics: Canvas2D, defect, P3)

Description

[Mayank Bansal]

0. Open about:addons in a tab.
1. In another tab, Go to https://agezao.github.io/confetti-js/
2. At the top, see the number of confetti to 8000
3. Scroll down and click on the "Render" button
4. Now detach this tab from the main window

AR: Once you detach the demo tab, it will stay blank for a long period of time, but the CPU/GPU continue to be used.
ER: Not so

Bisection:
This is the regression I got, but it is probably wrong.
Bug 1857618 - Ensure normal tasks get enough time to run, r=mstange
Differential Revision: https://phabricator.services.mozilla.com/D193409

The actual regression may be due to the Canvas work done in Q4'23

Repros with both D2D-canvas and gpu-canvas

Comment 1

[Mayank Bansal]

Attachment: about:support

Comment 2

[Mayank Bansal]

Attachment: bandicam 2024-05-04 18-41-05-766.mp4

Comment 3

[Mayank Bansal]

I did a second bisection with gpu-canvas enabled, and this is the best regression range i got:

https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=e39cc33d235644207d14479a4f952740cca9387d&tochange=7bc9f9c659dd5c04c341cef4f898cb08574d9cdd

cc :smaug and :emilio as they have multiple bugs in this range.

Comment 4

[Mayank Bansal]

Also got a crash one time: https://crash-stats.mozilla.org/report/index/2f3e8379-5d98-4202-9b94-85c950240504#tab-bugzilla

Comment 5

[Olli Pettay [:smaug][[e-mail]](vacation until July 14)]

bug 1857618 was a scheduling change, so that shouldn't matter here - or if it does, it hints that the issue is elsewhere.

Nothing in the range mentioned in comment 3 is really obvious. mozregression doesn't give you a better range?

Mayank, do you think you could try to capture this in a performance profile? Enable IPC profiling and whatever graphics related threads there are.

Comment 6

[Mayank Bansal]

https://share.firefox.dev/3UK3pWq with IPC profiling.
(ni? you back just so that you dont miss my response.)

Comment 7

[Olli Pettay [:smaug][[e-mail]](vacation until July 14)]

I think I saw the issue once. It was a very short temporary issue.
I did see the following in the terminal though
Crash Annotation GraphicsCriticalError: |[C0][GFX1-]: Failed to connect WebRenderBridgeChild. isParent=false (t=37.3914) |[C1][GFX1-]: Failed to connect WebRenderBridgeChild. isParent=false (t=43.7548) |[C2][GFX1-]: Failed to connect WebRenderBridgeChild. isParent=false (t=53.7139) [GFX1-]: Failed to connect WebRenderBridgeChild. isParent=false

Tentatively moving to WebRender

Comment 8

[Mayank Bansal]

I did a bisection today, and this is the range I got:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=70f0875a4aa2093a5fd63434db01d5dfbf3db1a6&tochange=e22abf3976f2653087769023f21c9c1615e9437d

Blaming bug 1871467

Profile with gfx preset logging: https://share.firefox.dev/4b4pN1C

Comment 9

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

Set release status flags based on info from the regressing bug 1871467

:lsalzman, since you are the author of the regressor, bug 1871467, could you take a look?

For more information, please visit BugBot documentation.