I do the same for several TB of production data here as well, but have found
that very active filesystems (like mail) can cause rdiff-backup to complain a
bit. As you can expect, it doesn't like it if you start backing up a
(source) mailbox, and before it can finish doing all the reverse-diff/update
of the (backup) mailbox you let the mail server stuff another message into
the (source) mailbox. I speak from the experience of mbox formatted mail - I
don't know if other formats would have quite the same issues.
I've used rdiff-backup for nightly backups in a mail-server scenario and it
works great. What I found is that I would sometimes miss a single night's
backup of certain mailboxes, but they would be caught the next night. The
randomness of which user's mailbox was updated at exactly the wrong time was
enough that I had success for any given mailbox nearly 100% of the time,
while in aggregate I had several mailboxes fail each night.
If you can, I'd look into snapshots to make this problem go away. If you are
using LVM it is fairly easy to create a snapshot volume:
lvcreate --size 10G --snapshot --name snapshot /path/to/lvm/volume
(where /patch/to/lvm/volume is your own LVM volume), mount that new snapshot
volume, and then do your backups from the snapshot. Once you are done with
the rdiff-backup, you can remove the snapshot with:
lvremove /path/to/lvm/snapshot-volume
For "mission critical" data, I also push a copy of my rdiff-backup archive to
an offsite server using rsync. (We're lucky enough to have a fat pipe to
another site). That way I have 3 disk copies, on 3 different hardware
devices, in 2 physical locations, where the data is within 24 hours of
current. (The live data, the local rdiff-backup copy, and the rsync'ed copy
of the rdiff-backup copy). I can bring up either of the two rdiff-backup
versions as a primary repository in case of a nasty hardware failure, like a
raid controller going up in smoke and taking out an entire disk array (like
our last power failure...). This is particularly nice since the data is as
current as tape backup, but your time to recovery is MUCH shorter. I this
case, under an hour to reconfig and bring back 3+TB of data on about a dozen
logical volumes.
True, the multiple copies are expensive, but for us it was cheaper than tape
solutions.
Do keep in mind, though, that rdiff-backup does not give you easy (if clunky)
archives like tape backups do. If you are required to retain e-mail records,
you will still need a tape solution or some other offline system.