Replication sync-server and Delayed Delete

[Bron Gondwana]

On Wed, Sep 15, 2010 at 12:29:18PM +0100, Gavin Gray wrote:
> Hi there,
> 
> We have a cyrus murder using replication and we have a few questions
> about the behaviour we are seeing on our system.
> 
> 1. cyr_expire on the master doesn't cause any replication to happen.
> Is that 'correct'? In other words if we want to delete folders from
> the DELETED heirarchy on the replicant then we need to also run
> cyr_expire on the replicant?

Yeah, pretty much.
 
> 2. We're also a little unclear about replication vis a vis the
> delayed expunge and the unexpunge facility. Could you explain what
> ought to happen in terms of replication when email is expunged and
> then possibly unexpunged if anything?

It's a bit messy.  Unexpunge is a sin against IMAP by the way, and
has been replaced with "generate new UID and promote" in 2.4.  In
which case it's just like a new append wit the same flags, and
replicates like an append :)

2.3 replication ignores expunges - it's as if they don't exist!  When
the mailbox syncs, it nukes the records that aren't "alive" on the
master from the replica.  If you re-inject them with unexpunge, it
should find them and sync_combine_commit() the result.  I don't know
if unexpunge inserts replication events though - somewhat doubt it.

> 3. We are seeing a strange anomaly on the replication of deleting a folder.
>    e.g a user deletes a folder
>        the folder goes into the DELETED heirarchy of the partition
> the user's mailbox is on
>        the folder is also deleted from the replicant as we would expect
>        however the folder on the replicant goes into the DELETED
> heirarchy on a different partition(the default     partition as
> specified in cyrus.conf). Is this normal?

Replication and partitions is broken in some ways in 2.3.  It should
be better in 2.4 I believe, though I haven't tested it.  I'm going to
be releasing an alpha super-soon (it's been pushed to git.cyrusimap.org
now!)
 
Bron.