Closed Bug 1319777 Opened 3 years ago Closed 3 years ago

[E10S a11y] slower dynamic updates or even infrequent hangs of tabs or the whole browser

Categories

(Core :: Disability Access APIs, defect)

53 Branch
x86
Windows 10
defect
Not set

Tracking

()

RESOLVED WORKSFORME

People

(Reporter: MarcoZ, Unassigned)

References

(Blocks 1 open bug)

Details

(Keywords: regression)

Probably related to landing of bug 1270916 and friends, but hard to reproduce/pinpoint. But I've observed several very slow dynamic updates and even two cases of hangs in the past three days where the IRCCloud tab, when loading more scrollback buffer, caused Firefox to completely freeze. This didn't use to happen before the event stuff was rewritten. There, E10S was almost as fast as E10S off when it came to dynamic updates within a page.

And I never experienced freezes like the ones in the past three days where the CPU would spin like crazy, the whole machine would freeze, NVDA would no longer speak, but eventually, an alt+f4 would register and Firefox would shut down without crashing, then releasing the machine back to me after several minutes.
Blocks: 1270916
Alex can you help get a profile here?
Flags: needinfo?(surkov.alexander)
is it NVDA specific or I can try to reproduce it on OS X too? Do I understand right that nothing like that is experienced in single process mode?
Flags: needinfo?(surkov.alexander)
Based on comment 0 I don't expect this is e10s specific. Marco can you confirm (when back)?

Alex can you get a profile in windows?
Flags: needinfo?(surkov.alexander)
Flags: needinfo?(mzehe)
(In reply to David Bolter [:davidb] from comment #3)

> Alex can you get a profile in windows?

of course, this will take longer though
Flags: needinfo?(surkov.alexander)
(In reply to David Bolter [:davidb] from comment #3)
> Based on comment 0 I don't expect this is e10s specific. Marco can you
> confirm (when back)?

I know the major bottlenecks in the single process mode, if you recall then I even sketched out a roadmap to address them. Extra profile won't be useless of course, but I afraid it may not give us new information.
(In reply to Marco Zehe (:MarcoZ) from comment #0)
> Probably related to landing of bug 1270916 and friends, but hard to
> reproduce/pinpoint. But I've observed several very slow dynamic updates and
> even two cases of hangs in the past three days where the IRCCloud tab, when
> loading more scrollback buffer, caused Firefox to completely freeze. This
> didn't use to happen before the event stuff was rewritten. There, E10S was
> almost as fast as E10S off when it came to dynamic updates within a page.
> 
> And I never experienced freezes like the ones in the past three days where
> the CPU would spin like crazy, the whole machine would freeze, NVDA would no
> longer speak, but eventually, an alt+f4 would register and Firefox would
> shut down without crashing, then releasing the machine back to me after
> several minutes.

hm, on linux I can see a hang for a couple seconds, but that's it.  According to perf very little of the time is spent in a11y code, and that seems to mostly be logging stuff which is odd, but that is what it is.
I have not been able to reproduce these freezes in nightlies as of this week. None of the steps and IRCCloud actions I saw this in previously have caused a freeze recently.
Flags: needinfo?(mzehe)
Flags: needinfo?(dbolter)
NI+ Marco: please reopen if this bug is still valid, and not captured by other bugs.
Status: NEW → RESOLVED
Closed: 3 years ago
Flags: needinfo?(dbolter) → needinfo?(mzehe)
Resolution: --- → WORKSFORME
Flags: needinfo?(mzehe)
You need to log in before you can comment on or make changes to this bug.