1495162 - 668MB of heap-unclassified in GPU Process

Status: RESOLVED WORKSFORME

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

Description

[Mayank Bansal]

Attachment: memory-report.json.gz

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:64.0) Gecko/20100101 Firefox/64.0
Build ID: 20180928220433

Steps to reproduce:

Enable WR
1.I used my regular profile for about 1 hour, mostly bugzilla pages, and profiler pages
2. Closed all tabs, and checked memory use is task manager. IT showed 800MB in one process
3. I opened about:memory, minimized the memory, and saved a report



Actual results:

GPU (pid 8988)
Explicit Allocations

701,607,936 B (100.0%) -- explicit
├──668,156,016 B (95.23%) ── heap-unclassified
├───15,560,000 B (02.22%) -- threads


Expected results:

not so, specially after closing all tabs, and minimizing memory manually

Comment 1

[Bobby Holley (:bholley)]

Hm, that seems bad. Given that it's unclassified, memory reporters aren't going to help here.

If you're willing to help debug further, I think there are basically two approaches:
(1) Narrow down a reduced STR, or
(2) Run your session with instrumentation.

If you can reduce it down to a small set of pages and/or actions, that's obviously ideal. I think the best approach for option (2) would be to use WPT, via the instructions at [1], and send me a trace.

[1] https://wiki.mozilla.org/Tracing_VirtualAlloc_With_Xperf

Comment 2

[Mayank Bansal]

I was able to repro this under WPT. But the resultant file was 13GB uncompressed, and 1GB+ compressed.
There is no way I can upload that much data

I opened a ton of bugzilla tabs, 1 gmail, a couple of github tabs. And I scrolled a lot on the bugzilla pages.
Is that information useful in adding some logging or something?

Comment 3

[Mayank Bansal]

Attachment: memory-report.json.gz

with image.mem.debug-reporting enabled

Comment 4

[Bobby Holley (:bholley)]

Might this be related to bug 1495661?

Comment 5

[Mayank Bansal]

I dont see such large heap-unclassified now. In fact, memory use is much better now. I wouldnt mind calling this bug as FIXED, specially as I dont have exact STR

Comment 6

[Bobby Holley (:bholley)]

Sounds good. Thanks for retesting!