722678 - Lots of jank on recent nightlies

Status: RESOLVED DUPLICATE of bug 722188

(Firefox :: General, defect)

Description

[(not currently active) Ted Mielczarek]

I'm seeing really horrible jank on recent nightlies (perhaps about a week or so). This is on all three platforms: Windows, Mac and Linux. My browser hangs for several seconds at a time every couple of minutes. (I haven't pinned down the exact intervals here.)

I installed about:jank on Mac and Win, the Windows report looks like:
about:jank results (22845 samples)

6721 - c-GC::GarbageCollectNow
5948 - c-bookmarks::RunInBatchMode
2530 - c-bookmarks::RemoveFolderChilder
2371 - c-storage::Statement::ExecuteStep
1242 - c-Timer::Fire
964 - c-gfx::DrawThebesLayer
773 - c-event::nsViewManager::DispatchEvent
576 - c-CSS::ProcessRestyles
281 - c-network::nsHttpChannel::OnDataAvailable
270 - c-layout::FlushPendingNotifications
264 - c-layout::DoReflow
226 - c-JS::CallEventHandler
164 - c-network::nsHttpChannel::OnStopRequest
149 - c-GC::CycleCollectNow
145 - c-content::nsXMLHttpRequest::OnStopRequest
103 - c-JS::EvaluateString
39 - c-Paint::PresShell::Paint
29 - c-AnnotationService::SetItemAnnotation
24 - c-nsInputStreamPump::OnStateStart
12 - c-Input::DispatchMouseEvent
8 - c-image::imgFrame::Draw
2 - c-html5::RunFlushLoop
2 - c-nsObjectFrame::InstantiatePlugin
1 - c-JS::EvaluateStringWithValue
1 - c-layout::nsLayoutUtils::DrawBackgroundImage

Comment 1

[Dietrich Ayala (:dietrich)]

cc'ing marco for the bookmarks stuff. that's not usual.

ted: do you run sync? have you made any recent changes lately? running sync on mobile?

Comment 2

[(not currently active) Ted Mielczarek]

I do run sync. I have two desktops and a laptop syncing. I don't run sync on mobile currently, I have in the past (with XUL Fennec, a while ago).

Comment 3

[(not currently active) Ted Mielczarek]

I don't recall making any changes lately. I do most of my browsing on the Windows machine, but a fair amount on my Mac, and occasional browsing on Linux.

Comment 4

[Marco Bonardo [:mak]]

So, this is a dupe of bug 722188. that said, there must have been some threading/mutex change lately. My best bet is the SQLite upgrade.
Would be nice to fetch which query is on the async htread holding the mutex when this happens, if possible, if we don't find that information we can backout the sqlite upgrade from aurora, at least.

Comment 5

[Jeff Muizelaar [:jrmuizel]]

(In reply to Marco Bonardo [:mak] from comment #4)
> So, this is a dupe of bug 722188. that said, there must have been some
> threading/mutex change lately. My best bet is the SQLite upgrade.
> Would be nice to fetch which query is on the async htread holding the mutex
> when this happens, if possible, if we don't find that information we can
> backout the sqlite upgrade from aurora, at least.

We should be able to do this when bug 719777 is done.

Comment 6

[Jeff Muizelaar [:jrmuizel]]

Otherwise, the slow sqlite query stuff might show it.

Comment 7

[Marco Bonardo [:mak]]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #6)
> Otherwise, the slow sqlite query stuff might show it.

yes, about:telemetry. Though that presumes you don't kill the process when it hangs.