1532749 - [Direct3D 11 (Advanced Layers)] Loading a website causes an insane spike in RAM (15GB) even with a clean profile, when it works fine on other browsers

Status: RESOLVED FIXED

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

Description

[Peter]

Attachment: Task manager loaded with a clean profile attempting to load the site

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:65.0) Gecko/20100101 Firefox/65.0

Steps to reproduce:

Go to http://notabilia.net, and watch the RAM usage hit 15gb on a clean profile, or >32gb on an existing heavier profile.

Actual results:

I noticed that website was freezing up my PC when attempting to load. I checked my RAM and all 32gb had been eaten up. I made a clean profile to check it works before figuring out which extension was causing it, and was pretty surprised to see the spike happen in a completely fresh profile too.
The usage was less with a clean profile but still extremely high

Expected results:

Website loads without burning all your RAM.

Comment 1

[Kanchan Kumari QA]

I could reproduce this issue on

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:65.0) Gecko/20100101 Firefox/65.0

Comment 2

[Olli Pettay [:smaug][[e-mail]]]

So far I haven't reproduced on Nightly/Linux.

Comment 3

[Alice0775 White]

On Nightly67.0a1 Window10,
it works if WebRender enabled.
It works if HWA disabled.

But, the problem is reproduced if HWA(Direct2D) enabled.

Comment 4

[Alice0775 White]

There are at least 2 regressions.

#1 regression: Browser hang up, but memory usage is small.
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=25d4e570f1d983ecc112e835d8a23eeafa8cd684&tochange=a9ec5effe35f607793514fe3688efb3ba8096689

Regressed by: Bug 1248913

#2 regression: Browser and Windows hang up, and memory usage is full.
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=81ea83beadf846f0d8f691162e5ef74cddf3c47a&tochange=13fe99de32f645984008a2ddc31923093a9fa43a

Regressed by: 13fe99de32f6 Bas Schouten — Bug 1300338: Do not attempt to preserve layer content when we're going to override it in a blend operation anyway. r=jrmuizel

Comment 5

[Olli Pettay [:smaug][[e-mail]]]

This could be qf or memshrink, but rather bad anyhow.

Comment 6

[Denis Palmeiro [:denispal]]

Hey Bas, could you help take a look at this one? Thanks!

Comment 7

[Pascal Chevrel:pascalc]

Marking as fix-optional for 67 as this is a long standing issue and the bug is unassigned. If an safe patch materializes, we can evaluate uplifting it.

Comment 8

[Patricia Lawless]

Bulk change to wontfix for 68 (P3+ carryover with needinfo).

Comment 9

[Bas Schouten (:bas.schouten)]

We shouldn't be using this anymore with WR.