1980558 - [meta] Add-on manager functionality has significant impact on start-up

Status: NEW

(Toolkit :: Add-ons Manager, task, P3)

Description

[Justin Link]

This comparison shows a significant improvement in numerous start/page-load metrics as a result of disabling add-on manager (aka extension manager) initialization.

newssite-applink-startup is a test that we've been looking at closely lately and it improves by ~8-14% with add-ons disabled. Various CPU time metrics also show dramatic reductions, likely due to some additional time savings that occur after the applink_startup metric is taken.

This was observed on Android specifically but it would make sense that there would be some impacts on other platforms as well.

Comment 1

[Luca Greco [:rpl] [:luca] [:lgreco]]

Hi Justin,
could you share some more details about this meta bug and next steps related to it? e.g. Are you planning to attach new bugzilla issue to it and a GeckoProfiler data collected by running the newssite-applink-startup with and without the add-on manager or provide details about how to run the test locally so that we can collect GeckoProfiler data to look into?

Comment 2

[Justin Link]

Yes - my apologies for the meta-bug with no linked bugs (yet). I will be creating two or three more specific bugs based on what Markus and I saw in some profiles. I will also grab some profiles from the comparison that I cite above and will post them here and take a also look myself to see if anything interesting jumps out at me.

If you want to run the same experiment locally (on whatever platform you are using) it was just a one-line change here so it should be simple to replicate.

Comment 3

[Rob Wu [:robwu]]

Is there a scenario where extensions should be completely skipped/ignored during startup, for app links?

Comment 4

[Justin Link]

I'm not aware of any motivation to actually skip extensions during start-up. The reason for the initial experiment was just to quickly get an idea of what the total impact of extensions is (on start-up). Since the impact was pretty clear we decided that it was worth making a note of it and looking a bit closer and getting some more eyes on it. (We have a big focus for the year on reducing applink start-up time.)