Backup of Cyrus

[Craig Ringer]

Doug Koobs wrote:
> I've been doing some reading of the list archives, and have learned that my
> current backup process is not sufficient. I'm hoping to get some feedback on
> my new proposed process. I only have about 20 users, total mail store is
> under 300 MB. I can stop the services. I want to be able to recover from
> total disaster, and from user errors. For now, I will do a full backup
> nightly
> 
> I'm using RH9 and Simon Matter's RPMs. Here are the beginnings of a script
> that I hope to use:
> 
> #######################################
> postfix stop				# Stop MTA
> etc/rc.d/init.d/cyrus-imapd stop		# Stop Cyrus
> su cyrus -c "ctl_mboxlist -d" > mailboxes-`date '+%m%d%y'` # List mailboxes;
> #file makes restore easier???

It's not quite as critical as I originally thought (as the ACL 
information _is_ stored somewhere else as well), but it can make 
recovery from db corruption easier, and can make restoring to a 
different machine a _lot_ less fuss. That said, I think Simon Matter's 
RPM builds take care of making these backups for you anyway. According 
to a recent post by Simon it makes backups in /var/lib/imap/backups - so 
verify that they're there and look right, and you should be fine.

> #Backup the following directories (I  haven't decided on which tool yet):
> #/var/spool/imap
> #/var/lib/imap
 >
> etc/rc.d/init.d/cyrus-imapd start		# Start Cyrus
> postfix start				#Start MTA
> ########################################
> 
> Will this be sufficient? Thanks,

It's pretty similar to what I do. Though I use snapshots instead of 
stopping the daemons, I'd /prefer/ to be able to just stop the daemons. 
Unfortunately, my mail spools are growing - fast - and I have users who 
keep weird hours (not least myself). At last check the gzipped backup 
files were 3gb, from mail spools of 4.6GB. This would mean too much 
downtime if I stopped the daemons, especially since I'm doing network 
backups to (slow) tape.

IMHO the best way to make sure your backups are OK is to restore them. 
Preferably to a test machine ;-) - ideally the one you'd be restoring to 
if there was a hardware failure right now (and maybe another, different, 
one as well). Alas, you're proabably in a situation like me - "what cold 
spare?!?" - where this isn't possible. I did my restore testing on a 
hapless debian 3.0 box - and this is why I'm so paranoid about 
converting the mailboxes.db . Everything else about the restore went 
perfectly, but it couldn't use the mailboxes.db because of a different 
berkley db version. I'm very glad I learned that then, and not after a 
server failure.

Craig Ringer