668653 - Event loop telemetry

Status: NEW

(Core :: XPCOM, defect)

Description

[Boris Zbarsky [:bzbarsky]]

It would be nice to add event loop info (how much time we're spending away from it, say) to telemetry...

Comment 1

[Jesse Ruderman]

Does the code in bug 606574 help?

Comment 2

[(not currently active) Ted Mielczarek]

The code in bug 606574 could measure this, but it wasn't written with the intent of running in release builds on end-user systems. As written it would be terrible for power consumption, since it wakes up every 10 milliseconds to ping the event loop.

I think the code in bug 429592 would be more suited to measuring this, since it's already designed to run all the time in release builds.

Comment 3

[(not currently active) Ted Mielczarek]

Also, what sort of numbers would you want to send here? A histogram of event loop delays?

Comment 4

[Andrew McCreight [:mccr8]]

I'm not sure if this is redundant with other Snappy telemetry work.  I think that has mostly focused on measuring specific sources of pauses.

Comment 5

[(dormant account)]

(In reply to Andrew McCreight [:mccr8] from comment #4)
> I'm not sure if this is redundant with other Snappy telemetry work.  I think
> that has mostly focused on measuring specific sources of pauses.

I think this is highly relevant. Assigning this a P2 until we can find someone to work on this. This is not a P1 because it's not clear how this would help us improve responsiveness.

Comment 6

[(dormant account)]

... in near future.