600427 - change passphrase causes far too many deletes

Status: RESOLVED FIXED

(Firefox :: Sync, defect)

Description

[Mike Connor [:mconnor]]

First, we should just delete user/storage, not delete by collection.  Need to set the header, but this will be dramatically faster and simpler.

Second, we should create meta/global before we call sync, so we don't try to call _freshStart and wipe the server again.

Third, we should reset lastSync for each engine, and only wipe the server again if lastSync is a different value.

Comment 1

[Mike Connor [:mconnor]]

Fourth, the GET after the DELETE in _syncStartup is generally quite silly.

Comment 2

[Richard Newman [:rnewman]]

A lot of this changed at the end of last year. Needs verification to see what the current behavior is.

I believe there's a reason we continued to delete collections individually... Philipp, do you remember?

Comment 3

[Mike Connor [:mconnor]]

The only reason I can think of is that we wanted to preserve the pub/priv keypair for migration of passphrases.  I think we can/should do this on an explicit key change.

Comment 4

[Richard Newman [:rnewman]]

(In reply to comment #3)
> The only reason I can think of is that we wanted to preserve the pub/priv
> keypair for migration of passphrases.

Ah yes. Not quite for migration: it was to allow old clients (who check keys before storage version!) to detect that they're outdated rather than erroring.


> I think we can/should do this on an
> explicit key change.

You're implying that once a user generates a new sync key, we should no longer allow older versions of Sync to detect that they're outdated, and will instead raise some other error about failing to fetch keys. (That's the consequence of deleting the whole storage collection.)

We still have about 30,000 active daily users who are using versions prior to 1.6. These users won't get the "please update your client" message on their second and subsequent clients once they begin the long journey towards modernity.

Your call on whether to break them...

Comment 5

[Richard Newman [:rnewman]]

We switched the one-delete-per-collection to DELETE storage in Bug 684798.