I've started playing with replication for courier. The objective is to maintain synchronization between totally separate mail systems for availability purposes. The underlying structure of maildirs lends itself to this quite nicely, and the synchronized state tracking for IMAP (uid/uidvalidity) isn't too bad, although there's a number of pathological cases that don't work terribly well.

At any rate, it's something I intend to build into a functional system. Currently there's a patch and a replay script available at http://www.rcpt.to/courier/ that is a first tottering step in that direction. The patch instruments various places in courier (maildir operations) and writes a transaction log.

The replay script can parse the transaction log and build a maildir elsewhere that looks more or less similar to the real one. The script is fairly stupid at the moment; it doesn't do locking, conflict fixups exist as comments rather than code at best (and, in fact, mostly in my head right now), and there's a number of assumptions to do with directory structure.

This is an FYI rather than an announcement - I'm soliciting comments from anyone that has anything to say about the idea :)

m.