983733 - Add telemetry probe to get performance data on customize mode transition

Status: RESOLVED INVALID

(Firefox :: Toolbars and Customization, defect)

Description

[Mike Conley (:mconley) (:⚙️)]

We currently collect performance data on tab animation frame paint times. It might be worth doing the same with the customize mode transition, just to get a sense of how many people out there are experiencing jank with it.

Comment 1

[Avi Halachmi (:avih)]

So what exactly do we want to measure? Telemetry values are expensive (local/remote storage, upload time, etc), so we should use a minimal set which gives us enough info.

Potential values to measure IMO:

- Average interval of the CSS animation part.
- Duration of the CSS animation part

- Average interval of the "second" part.
- Duration of the second part.

- Average interval of "Everything".
- Duration of "Everything".


For reference, here are the tab animation telemetry numbers (for which we only collect average interval) for nightly:
open-newtab-blank  2nd/3rd tab: mean=21.0, median=17, 75th%=19, 95th%=43
open-newtab-thumbs 2nd/3rd tab: mean=26.4, median=18, 75th%=24, 95th%=68
tab-anim - everything*:         mean=26.7, median=18, 75th%=24, 95th%=62

* "everything" is any tab animation of open/close for any reason and regardless of how many other tabs exist.


mconley? what say you?

Comment 2

[Mike Conley (:mconley) (:⚙️)]

We also need to deal with both entering _and_ exiting, so how about:

-Average interval of CSS animation part for entering
-Duration of the CSS animation part for entering
-Average interval of CSS animation part for exiting
-Duration of the CSS animation part for exiting

-Average interval of "everything" for entering
-Duration of "everything" for entering
-Average interval of "everything" for exiting
-Duration of "everything" for exiting.

I imagine from these measures, we can derive the durations a rough estimate on average for the "second" parts. How does that sound?

Comment 3

[Avi Halachmi (:avih)]

After discussing it on irc, Mike, Vladan and myself agreed that the following 4 histograms would give us a good coverage at a reasonable cost:

- CSS animation only - average frame interval on enter/exit.
- Overall duration of enter/exit.

The reason we left out average interval/duration of overall enter/exit is that it's not very visible to the user beyond the actual CSS animation (which we will measure). The pure animation FPS and the overall duration satisfy us good enough.

I'll take it.

Comment 4

[Avi Halachmi (:avih)]

Vladan thinks this should get lower priority due to lack of explicit need. mconley agrees this is not high priority right now.

Unassigning myself for now, ping here if the priority changes.

Comment 5

[:Gijs (he/him)]

We removed the transition.