On Sat, Oct 09, 2010 at 09:59:44AM -0500, Patrick Goetz wrote: > On 10/9/2010 7:23 AM, Michael D. Sofka wrote: > > > > I can either rsync just the mail messages to the warm backup, and then reconstruct the index files. Or, I can take a slight detour, and upgrade the warm backup server to 2.3.16. > > > > This is something I've been wondering, too. The problem with not > copying the indexes is that presumably you lose all the metadata. And > if you rsync /var/spool/cyrus/mail it would seem to me that you run the > risk of the index being out of sync with the mail spool; i.e. the only > way to safely use rsync is > > stop cyrus > rsync > restart cyrus > > Which isn't practical for some environments. If you have delayed expunge it should be safe to rsync the metadata first and then the spool afterwards. It's a bit racy of course. Better is to fcntl lock the cyrus.header and cyrus.index files (or lock the foldername.lock file in 2.4 of course... when we release that in OMG 2 days!) But best of all is to not fiddle with the underlying filesystem at all. I want to replace our old and creaky backup system pretty soon (it doesn't grok namelock yet - though it is safe due to the use of index exlusive locks - just inefficient). The replacement will use a cyrus tool that can do a full or partial backup of either a mailbox or an entire user. This will also replace mbdump/undump for XFER (murder mailbox moves) and be identical in syntax to the replication subsystem. I haven't had time to make it all work just how I want in Cyrus 2.4, but it's sitting on my TODO list and slowly rising up it, to the point where it gets written on whiteboards a LOT (including discussion of how to implement it). FastMail/Opera has 3 new programmers who started in the last week - I met them all during interviews, but I'm currently travelling (sitting in Newark airport right now waiting for my flight upstate - 16 hours from Oslo to Syracuse!) on my way to meet Ken and Dave. We'll have more resources to throw at Cyrus internals pretty soon. Some of it specific to our environment of course - but we like to simplify the bits we have to deal with, and that has benefits for everyone. Bron. ---- Cyrus Home Page: http://www.cyrusimap.org/ List Archives/Info: http://lists.andrew.cmu.edu/pipermail/info-cyrus/
