993673 - Uplift Bug 988437 to Firefox Beta (29)

Status: RESOLVED FIXED

(Firefox for Android Graveyard :: Android Sync, defect)

Description

[Nick Alexander :nalexander [he/him]]

This ticket is MoCo-only because I want to discuss detailed Google Play Store numbers.  I downloaded the Google Play Store data [1] using Firefox Beta > Statistics, Export as CSV, and then checking only "Installs > Current" and "Dimensions > by app version".  I used https://mail.mozilla.com/home/[e-mail]/Releases%20Scheduling.html to determine dates.

In https://bugzilla.mozilla.org/show_bug.cgi?id=988437#c2, I outlined a plan to update some of the technical details around Firefox Accounts.  The immediate goal of the plan is to address Bug 975625, which prevents Sync + FxA working on Android 2.2 devices.  The longer term win is to allow safely updating Firefox Account internals.  We really want Bug 988437 to be present in the very first build of Firefox Release (29).

To be clear: the changes discussed will only have ramifications for existing Firefox Accounts users, and the worst possible outcome is that they silently lose their Firefox Account on the Android phone, and need to re-create it.  Of course, most (presumably? I cannot get any numbers) Firefox Beta 29 users *do not have a Firefox Account*.  So this issue will never affect them.

Overview of changes and repercussions
-------------------------------------

Bug 988437 changes the "Android Account type" defining a Firefox Account.  A device that changes this "Account type" will lose any existing Firefox Account unless special arrangements are made.

To not lose an existing Firefox Account, we write a marker to the device, and then use that marker to restore the Account that would be deleted (but with the updated "Account type").  The wrinkle is that the code to write the marker needs to *run* before the "Account type" is changed: in practice, this means that a Sync needs to be attempted before the package is updated to include Bug 988437.  We expect any device that gets updated to Sync very shortly thereafter, so we'll take sufficiently updated to mean the marker has been written.

Current state
-------------

We have uplifted the code to write the marker (Bug 957894) to Beta; it landed on March 28 and is present in the builds on or after 20140331.  The code depending on the marker (Bug 988437) landed on Nightly about a week ago.

To decide whether enough devices have got the Bug 957894 code, and have had time to write the marker, I looked to Google Play Store data.

State of Beta
-------------

The only Beta builds that even have FxAccounts at all are those that are Beta 29; that is, those built on or after 20140318.

Looking at Google Play Store data from today, I see the following totals:

Build         Device count
----------    ------------
20140331..    360648
20140324..    51396
20140318..    24714

That is, we have three Beta 29 builds in the wild; and 82.6% percent of our Beta population are running the newest released code.  If we were to uplift 988437 to Beta today, it would go out in the build scheduled for *next* Tuesday, on or around 20140415.  (That is, there would be a Beta release on or around 20140408 that would not have the new code.)  I have no idea if more Beta releases significantly changes the portion of our Beta user base running the newest code; pessimism suggests no, so we expect to be exposing about 20% of our Beta users to lost Accounts.  We really can only delay one week (until the 20140422 Beta release) and still get this into Release on day 1.

My perspective is that we should uplift to Aurora now, (which has had the marker writing code for longer than Beta), verify that there is no catastrophic unexpected fallout, and uplift to Beta by the end of this week, so that we hit the 20140415 Beta build.  This gives a scant week to back out in the case of catastrophic fallout, while still getting into Release 29.  But I want release drivers and Responsible Parties to be aware of what I intend to do and raise concerns, questions, and alternatives as appropriate.

Comment 1

[Nick Alexander :nalexander [he/him]]

needinfo to folks who should be aware of this decision.  sylvestre, I meant to flag you in Bug 988437 weeks ago, but I seem to have not done so.  Sorry for the late notice.

Comment 2

[Nick Alexander :nalexander [he/him]]

I'm guessing elan cares about this too :)

Comment 3

[Richard Newman [:rnewman]]

+1 for uplifting ASAP. IMO it's a safe change, it really really ought to go out before 29 hits the release channel, and thus it's better to be in an earlier beta than a later one.

Comment 4

[Kevin Brosnan [Ex-Mozilla]]

957894 was included in 29b4 which is 20140331.. in your table. 

So you are saying people who would need to set up Sync again is the intersection of the set of people who have setup sync and have not updated to 29b4 plus the set of people who have upgraded to 29b4 and have not had a successful sync? That seems like a small number to me.

Landing this now would be released in about a week which would be 29b8 which would give 2 weeks for people to get either 29b4 or 29b6.

Comment 5

[Mark Finkle (:mfinkle) (use needinfo?)]

I have no problem with uplifting to Aurora and Beta. It seems like small numbers and you have covered the risks well.

You are being careful to make sure the marker is in before we change account types. How is Release affected? Both the marker and the account type change will occur at the same time.

I assume this is fine because Release has not had any FxA exposure yet.

Comment 6

[Nick Alexander :nalexander [he/him]]

(In reply to Kevin Brosnan [:kbrosnan] from comment #4)
> 957894 was included in 29b4 which is 20140331.. in your table. 

You are correct.

> So you are saying people who would need to set up Sync again is the
> intersection of the set of people who have setup sync and have not updated
> to 29b4 plus the set of people who have upgraded to 29b4 and have not had a
> successful sync? That seems like a small number to me.

Correct.  It is in fact a slightly smaller set: ... plus the set of people who have upgraded to 29b4 and never tried to sync at all.  (You don't need to sync successfully; just need to try to sync once.)

> Landing this now would be released in about a week which would be 29b8 which
> would give 2 weeks for people to get either 29b4 or 29b6.

Correct.

Comment 7

[Nick Alexander :nalexander [he/him]]

(In reply to Mark Finkle (:mfinkle) from comment #5)
> I have no problem with uplifting to Aurora and Beta. It seems like small
> numbers and you have covered the risks well.
> 
> You are being careful to make sure the marker is in before we change account
> types. How is Release affected? Both the marker and the account type change
> will occur at the same time.

We really want this in so that Release will be unaffected.  If it doesn't get into Release on day 1, then we do the same dance, but with the significantly larger population and slower cycle.

> I assume this is fine because Release has not had any FxA exposure yet.

Correct.

I think engineering and QA is in agreement that we should uplift sooner rather than later; let's wait for sylvestre's perspective.

Comment 8

[Sylvestre Ledru [:Sylvestre]]

Sure, let's go for beta.
If you consider it is critical and QA agrees, we have the possibility to release a beta7 (Thursday) for Mobile.

Comment 9

[Nick Alexander :nalexander [he/him]]

Thanks for weighing in, everybody, and sorry for the unusual step of creating an additional bug.  I'm going to uplift to Aurora and Beta now; due to the staged releases, this should give us Aurora feedback soon enough to take action on Beta -- if any is needed.

Comment 10

[Nick Alexander :nalexander [he/him]]

mfinkle and I discussed, and agree this ticket can be opened to non-MoCo folks.