709971 - reflow telemetry

Status: RESOLVED FIXED

(Core :: Layout, defect)

Description

[(dormant account)]

Bug 681535 added xul telemetry. Lets also add html reflow telemetry. This time do not report 0ms reflows. We need this to see how often we encounter pages that stall the browser in layout.

Comment 1

[Boris Zbarsky [:bzbarsky]]

I'd appreciate it if you build on top of the patches in bug 709256!

Comment 2

[Nathan Froyd [:froydnj]]

Is it worth doing something like what we did in bug 699051 by sending back URLs that take longer than $MAX_TIME in reflow?  (Possibly with MAX_TIME calculated based on some timing cleverness.)

Obviously there are significant privacy implications to doing that, but I'm not sure knowing "hey, 5% of pages stall the browser for longer than 5s" is going to help us that much, as we don't have STR from that data.  Willing to be corrected by the layout guys!

Comment 3

[(dormant account)]

(In reply to Nathan Froyd (:froydnj) from comment #2)
> Is it worth doing something like what we did in bug 699051 by sending back
> URLs that take longer than $MAX_TIME in reflow?  (Possibly with MAX_TIME
> calculated based on some timing cleverness.)

It would be cool, but sending back URL(or anything related to URL) is not going to fly

Comment 4

[(dormant account)]

Also good point about 5% of pages, can't skip 0ms reflows in this case.

Comment 5

[Nathan Froyd [:froydnj]]

(In reply to Taras Glek (:taras) from comment #3)
> (In reply to Nathan Froyd (:froydnj) from comment #2)
> > Is it worth doing something like what we did in bug 699051 by sending back
> > URLs that take longer than $MAX_TIME in reflow?  (Possibly with MAX_TIME
> > calculated based on some timing cleverness.)
> 
> It would be cool, but sending back URL(or anything related to URL) is not
> going to fly

OK, then what can we do once these numbers are in hand?  Plot them for Talos (or similar measures) tests?  I'm not seeing how having just reflow ms in isolation is going to help us.

Comment 6

[Nathan Froyd [:froydnj]]

Attachment: HTML reflow telemetry

Simple enough, I guess.

Comment 7

[Nathan Froyd [:froydnj]]

Attachment: HTML reflow telemetry

Comment 8

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 587365 [details] [diff] [review]
HTML reflow telemetry

r=me

Comment 9

[(dormant account)]

Boris, how hard would it be to track reflow in background tabs?

Comment 10

[Boris Zbarsky [:bzbarsky]]

In what sense?  The presshell knows whether it's in a background tab....

Comment 11

[(dormant account)]

(In reply to Boris Zbarsky (:bz) from comment #10)
> In what sense?  The presshell knows whether it's in a background tab....

that's what I wanted. I would like to know how much jank is caused by reflows in background tabs

Comment 12

[Boris Zbarsky [:bzbarsky]]

The mIsActive member of presshell should be what you want, then.  Presumably just log reflows when !mIsActive to a separate histogram?

Comment 13

[Dão Gottwald [:dao]]

http://hg.mozilla.org/integration/mozilla-inbound/rev/103460516db6

Comment 14

[Matt Brubeck (:mbrubeck)]

https://hg.mozilla.org/mozilla-central/rev/103460516db6