1379940 - Hovering some hyperlinks causes 100% CPU usage, memory leaks and UI hangs after landing patch from bug #1356317

Status: VERIFIED FIXED

(Core :: Graphics: Layers, defect)

Description

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

[Tracking Requested - why for this release]:

STR:
1. Open this website page - https://bugzilla.mozilla.org/show_bug.cgi?id=1357486#c41
2. Scroll down to " Treeherder Bug Filer" comment
3. Hover bug 1379721 (or do it here)
and enjoy 100% CPU usage, memory leak eating nearly all memory and GUI hang

Comment 1

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

FYI - Also set to "Off" "Use modal user interface" preference in Bugzilla General Preferences ( https://bugzilla.mozilla.org/userprefs.cgi?tab=settings )

Comment 2

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

"Speedy" Regression window (mozilla-central)
Good:
https://ftp.mozilla.org/pub/firefox/nightly/2017/07/2017-07-05-03-02-06-mozilla-central/

Bad:
https://ftp.mozilla.org/pub/firefox/nightly/2017/07/2017-07-05-09-59-41-mozilla-central/

Pushlog:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=0893f6685e154aaa3252ed091abb16a2feb2806d&tochange=11755fd63168581e194258d04bb6a7337779ec78

Comment 3

[Alice0775 White]

Regression window:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=41b421e98b77575451f5769f89af788fe68f183b&tochange=d1181db727669d3f0902c9ace520c377f67ae46a

Regressed by: Bug 1356317

Comment 4

[Alice0775 White]

Attachment: reduced html

Comment 5

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

Thank you Alice0775 White very much for finding specific regression range!

Comment 6

[Stephen A Pohl [:spohl]]

I'm able to reproduce and I'm looking into it.

Comment 7

[Stephen A Pohl [:spohl]]

Kris, could you take a look at this as well? I've observed a continuous increase in memory consumption when displaying the tooltip when hovering links, but the memory consumption reverts to its initial state once the mouse is moved and the tooltip disappears. At this point, I don't believe we're dealing with a memory "leak", but rather some kind of unnecessary, continuous memory allocation. Unfortunately, I've been unable to isolate where or why this is occurring so far.

Comment 8

[Stephen A Pohl [:spohl]]

At the risk of possibly stating the obvious, a tooltip (which is displayed when hovering a link) is a popup and gIsPopupCompositingEnabled is set to false, which is why previous to attachment 8881870 [details] [diff] [review] landing, we would not use OMTC.

Comment 9

[Stephen A Pohl [:spohl]]

Attachment: Patch

This basically restores attachment 8881870 [details] [diff] [review], but changes nsWindow::ShouldUseOffMainThreadCompositing() to enable OMTC for remote content. This fixes the issue reported in this bug. However, I'm not sure if this change is satisfactory because I don't know if we're also experiencing excessive memory consumption with OOP WebExtension popups.

Comment 10

[Marcia Knous [:marcia]]

Tracking 56+ for this regression.

Comment 11

[Kris Maglione [:kmag]]

I think we should probably just disable OMTC for "small" popups, which would include things like tooltips and menus:

http://searchfox.org/mozilla-central/rev/31311070d9860b24fe4a7a36976c14b328c16208/widget/nsBaseWidget.cpp#882

Comment 12

[Stephen A Pohl [:spohl]]

Attachment: Patch

As suggested. I've confirmed that this fixes the memory consumption issue for me locally.

Comment 13

[Stephen A Pohl [:spohl]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=96b4247c585f8345a974b2070c6c80dcd3559806

Comment 14

[Jim Mathies [:jimm]]

On Windows tool tips in bugzilla currently aren't displaying. Possibly a dupe of this bug?

Comment 15

[Jim Mathies [:jimm]]

(In reply to Jim Mathies [:jimm] from comment #14)
> On Windows tool tips in bugzilla currently aren't displaying. Possibly a
> dupe of this bug?

never mind, appears this is caused by level 3 sandboxing.

Comment 16

[Stephen A Pohl [:spohl]]

I'm setting this to checkin-needed to get a second pair of eyes on it before landing. The try run seemed to show a lot of failures, but I was unable to determine if those were all just known intermittents, or if this patch would introduce new failures. For comparison, here is a try run with this patch applied:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=3710fb78cf3deca4828ffd3d21af4f5e81d03fc2

And here is a try run with a patch that should just behave as nightly does right now:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=166dda3e730dd66bc1cd53042ba1ad56fc32812c

Rather than just introduce some whitespace for the second patch, I wanted to run a try run with a patch that touches the same files as the first patch. This was to make sure that we rebuild the same files and run the same tests. I've noticed before that a no-op patch with just some added whitespace does not necessarily trigger the same tests to run. It appears to me that both try runs show almost exactly the same failures, which leads me to believe that these are known intermittents. Is this correct?

Comment 17

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/c623242b62da
Fix high memory consumption when hovering links on Windows. r=kmag

Comment 18

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/c623242b62da

Comment 19

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

I'm confirming that this issue is fixed, starting since Mozilla Firefox Nightly 56.0a1 (2017-07-15).
Thank you very much! \o/