490661 - (asyncBookmarkBackups) bookmark backups should be async

Status: RESOLVED FIXED

(Toolkit :: Places, defect)

Comment 1

[Dietrich Ayala (:dietrich)]

hm, are these written on idle only?

Comment 2

[Marco Bonardo [:mak]]

nope, a backup may also be created on shutdown if there is no backup in the last 24 hours, this covers those cases where idle may not happen.

Comment 3

[Dietrich Ayala (:dietrich)]

ok, so if only on idle and possibly shutdown, this is a pretty low priority bug.

Comment 4

[Asa Dotzler [:asa]]

Taras says we should not do shutdown IO here.

Comment 5

[Marco Bonardo [:mak]]

No doubts, still, if the user never hits idle, there isn't another place where to backup his data.
Please read above, this only happens if there is no backup in the last 24 hours, and backups are saved on idle wherever possible.

Btw, all the import/export/backup/restore code is pretty old and needs a rewrite (bug 549736 may be a good thing to do, if we read asynchronously and directly from the db and allow to store data in different formats).

Comment 6

[(dormant account)]

Why do we need to do a backup on shutdown. Reducing this to p2 since we don't have data on how often this happens

Comment 7

[Dietrich Ayala (:dietrich)]

Yeah, definitely not P1. See above - we only backup on shutdown if the user never hit idle during previous session for a long enough time do to a backup *and* it's been >24hrs since previous backup.

Comment 8

[(dormant account)]

(In reply to Dietrich Ayala (:dietrich) from comment #7)
> Yeah, definitely not P1. See above - we only backup on shutdown if the user
> never hit idle during previous session for a long enough time do to a backup
> *and* it's been >24hrs since previous backup.

We could use telemetry to see how frequent this is.

Comment 9

[Marco Bonardo [:mak]]

We can consider the main-thread IO issue that was covered by this meta as fixed. Work on backups didn't finish, there's still lot of stuff to cleanup (like shutdown, backups hashes comparison...), but that has a smaller priority compared to the main-thread IO issue tracked here. It can proceed apart.