1368745 - Create whitelist of "known good" extensions for multi

Status: RESOLVED WONTFIX

(Firefox :: General, enhancement)

Description

[Jeff Griffiths (:canuckistani) (:⚡︎)]

The current cohort of mpc=true extensions in beta are less crashy than other cohorts:

https://sql.telemetry.mozilla.org/queries/4355#8687

Based on this I'd make an educated guess that if the most popular extensions in Beta are add-blockers, this improvement of stability is related to showing less ads.

What I'd like to move forward with is a plan something like:

1. create a whitelist of top/"known good" extensions from beta
2. identify the number of users in release that have only those extensions installed
3. potentially ship multi to those users in a staged roll-out controlled by the system extension after 54 is released.

Comment 1

[Jeff Griffiths (:canuckistani) (:⚡︎)]

Ben - where is this issue in your current list of priorities? It's pretty important to me.

Comment 2

[Ben Miroglio [:bmiroglio]]

(In reply to Jeff Griffiths (:canuckistani) (:⚡︎) from comment #1)
> Ben - where is this issue in your current list of priorities? It's pretty
> important to me.

My apologies--I've been on PTO / sans internet the last week. 

To clarify-- top/"known good" extensions would be the top N WebExtensions and MPC=true add-ons having no previous issues with e10s/Multi (such a whitelist should be reviewed by the add-ons team). It will be useful to know how many clients in release *only* have add-ons that belong to this whitelist, so I can provide sample size estimates for different values of N.

Let me know your thoughts.

Comment 3

[Jeff Griffiths (:canuckistani) (:⚡︎)]

(In reply to Ben Miroglio from comment #2)
> (In reply to Jeff Griffiths (:canuckistani) (:⚡︎) from comment #1)
> > Ben - where is this issue in your current list of priorities? It's pretty
> > important to me.
> 
> My apologies--I've been on PTO / sans internet the last week. 

o need - I'd forgotten about your PTO, hope it was a nice break :)

> 
> To clarify-- top/"known good" extensions would be the top N WebExtensions
> and MPC=true add-ons having no previous issues with e10s/Multi (such a
> whitelist should be reviewed by the add-ons team). It will be useful to know
> how many clients in release *only* have add-ons that belong to this
> whitelist, so I can provide sample size estimates for different values of N.
> 
> Let me know your thoughts.

Yes exactly.

Comment 4

[Ben Miroglio [:bmiroglio]]

I've estimated that ~3.8% of clients in release only have Adblock Plus and/or uBlock Origin (among any system or hotfix add-ons).

I can develop an expanded whitelist from Beta once the multiprocess_compatible field is uplifted in Beta 55.

Comment 5

[Jeff Griffiths (:canuckistani) (:⚡︎)]

(In reply to Ben Miroglio from comment #4)
> I've estimated that ~3.8% of clients in release only have Adblock Plus
> and/or uBlock Origin (among any system or hotfix add-ons).

Cool!

> I can develop an expanded whitelist from Beta once the
> multiprocess_compatible field is uplifted in Beta 55.

Sounds good.

Comment 6

[Jeff Griffiths (:canuckistani) (:⚡︎)]

(In reply to Ben Miroglio from comment #4)
> I've estimated that ~3.8% of clients in release only have Adblock Plus
> and/or uBlock Origin (among any system or hotfix add-ons).
> 
> I can develop an expanded whitelist from Beta once the
> multiprocess_compatible field is uplifted in Beta 55.

Ben, to be clear can we get a list of add-ons on beta sooner than 55? This would be by using the same method we seem to have used to create the 'addons-set50allmpc-multiBucket4' in the first place.

Comment 7

[Ben Miroglio [:bmiroglio]]

(In reply to Jeff Griffiths (:canuckistani) (:⚡︎) from comment #6)
> Ben, to be clear can we get a list of add-ons on beta sooner than 55? This
> would be by using the same method we seem to have used to create the
> 'addons-set50allmpc-multiBucket4' in the first place.

tl;dr:
Getting a list *is* doable with some gymnastics if we can't wait till beta 55 (which is next week correct?). If we can wait till 55, then let's :). 

In telemetry, no. The system add-on responsible for cohort assignment looks at client data that was just recently added to telemetry (it is only in nightly at the moment). There is one option: We can generate a whitelist now from add-ons in nightly that are marked as mpc=true, comparing them to frequently occurring add-ons in Beta. Hacky and a one-off, but will work. 

We could simply look at add-ons among clients assigned to the addons-set50allmpc-multiBucket{1, 4} cohorts however, the problem here is that we cannot verify if these add-ons are all truly mpc=true. It has been the case that additional filtering is required for all cohorts as some add-ons/settings leak in that shouldn't.

Let me know the priority for this.

Comment 8

[Jeff Griffiths (:canuckistani) (:⚡︎)]

(In reply to Ben Miroglio from comment #7)
...
> We could simply look at add-ons among clients assigned to the
> addons-set50allmpc-multiBucket{1, 4} cohorts however, the problem here is
> that we cannot verify if these add-ons are all truly mpc=true. It has been
> the case that additional filtering is required for all cohorts as some
> add-ons/settings leak in that shouldn't.

This is what I want. I will work with the add-ons team to trim the list down to something we're all comfortable with.

> Let me know the priority for this.

I still think it's high, happy to discuss what this would displace though.

Comment 9

[Mark Banner (:standard8)]

We no longer support legacy extensions, and multi-process has shipped, so I don't think we need this now.