1405168 - Memory usage rises (in unknown cases) until system freezes

Status: RESOLVED WORKSFORME

(Core :: Graphics: WebRender, defect)

Description

[Darkspirit]

Attachment: Screenshot_20171003_001119.png

This does not feel like bug 1377120 where everything freezed from the one to the other second.

I don't wanted to pollute bug 1394806 with this case, but if it's identical, please mark this a duplicate of it.

------------------------------------------------------------------
Nightly 58 x64 20171002100134 de_DE @ Debian Testing (KDE / Radeon RX480)
gpu process + layers hw accel + webrender + webrendest + blob-images (7 content processes)

Sometimes my desktop gets very slow. I thought Thunderbird would be the evil one, but most times I just closed it quickly and was to lazy to open ksysguard. 
But this can also happen when Thunderbird is closed. Closing some tabs or restarting Nightly helps.

Today I enabled OMTP (bug 1403935) just because it seems to be possible.
The slowness (that can happen after hours of Nightly usage) didn't feel different than yesterday or days before, so I don't think OMTP is the cause.

15 GB RAM used:
* 4 GB main process
* 5 GB only for whatsapp' content process
* 3 GB for the GPU process
* some other content processes

I made screenshots from ksysguard, about:memory and saved memory reports.
Those three buttons on the right-top of about:memory didn't free memory (GC/CC/Minimize Memory).

Comment 1

[Darkspirit]

Attachment: memory-report.json.gz

Comment 2

[Darkspirit]

Attachment: blob-images.png

Nightly 58 x64 20171003100226 de_DE @ Debian Testing (KDE / Radeon RX480)
layers hw accel + webrender + webrendest (max. 7 content processes)
* without gpu process
* without OMTP
* you can see blob-images enabled & disabled on the screenshot. My perception is that blob-images is the cause, but I don't know what I am talking about.

Comment 3

[Sotaro Ikeda [:sotaro]]

Memory leak related video and canvas was addressed by Bug 1407069 and Bug 1412246.

Comment 4

[Darkspirit]

bug 1414890 comment 4 showed a massive improvement.