Closed Bug 1396479 Opened 7 years ago Closed 7 years ago

multi-second jank/spinner when switching to certain tabs

Categories

(Core :: Graphics, defect, P1)

defect

Tracking

()

RESOLVED WORKSFORME
Performance Impact low
Tracking Status
firefox57 --- affected

People

(Reporter: heycam, Unassigned)

References

Details

(Keywords: regression, regressionwindow-wanted, Whiteboard: [gfx-noted])

For the last week or so in Nightly I've seen multi-second jank and tab spinner when switching to particular tabs.  The ones I've noticed are about:memory tabs (jank, since I guess they're run in the chrome process) and my web.whatsapp.com tab (spinner).

Here's a profile of switching to these tabs:

* https://perfht.ml/2iUjglq (from a treeherder tab to an about:memory tab)
* https://perfht.ml/2ewJ9T4 (from a treeherder tab to the web.whatsapp.com tab)
Some interesting profiles, Bas, can you take a look?
Flags: needinfo?(bas)
So of these two profiles, the second one is just spending time in Skia drawing a single SVG.. This could either be because you're system's memory bus is super-busy, or just because the SVG is super complicated.

The first one is spending time in a little weirder stuff.

Having said that, one thing that's noticeable about this profile is that during these rasterizes the sampler is only able to get -very- few samples, meaning the profiler isn't even getting scheduled, most likely neither is firefox. Suggesting the 'your system is really busy' hypothesis to be most likely.
Flags: needinfo?(bas)
Hrm, what's sort of odd is that we -are- getting samples for other threads, I wonder how that works, that might be some linux specific oddity. Markus, what do you think?
Flags: needinfo?(mstange)
Whiteboard: [qf] → [qf:p3]
We are getting samples for other *processes*.

I have no explanation for this phenomenon. Cameron, do you also see this slowness if the profiler is not running?
Flags: needinfo?(mstange)
Yes, I didn't have the profiler enabled when I noticed the problem.
Component: General → Graphics
Jamie, maybe your recent SVG+WhatsApp workwould apply here as well?

Cameron, if this is obvious, and recent (what Jamie is fixing is not recent) can you do a regression range?
Flags: needinfo?(jnicol)
Flags: needinfo?(cam)
Priority: -- → P1
Whiteboard: [qf:p3] → [qf:p3][gfx-noted]
This could indeed be the same issue for the whatsapp profile, hard to see exactly what's going on but all the time spent drawing svgs looks suspicious. The important thing for that one (bug 1398697) is that it should only happen at non-100% zoom, which I don't know was the case or not.
Flags: needinfo?(jnicol)
I now no longer seem to see this.  I don't think I would've had the about:memory pages zoomed.  Also I don't think those pages have any SVG in them.
Flags: needinfo?(cam)
Status: NEW → RESOLVED
Closed: 7 years ago
Resolution: --- → WORKSFORME
Performance Impact: --- → P3
Whiteboard: [qf:p3][gfx-noted] → [gfx-noted]
You need to log in before you can comment on or make changes to this bug.