[Dovecot] Dovecot altmove questions
sven at svenhartge.de
Mon Feb 27 04:26:11 EET 2012
Maria Arrea <maria_arrea at gmx.com> wrote:
> 3º We have separate raids for indexes (raid 1+0) and mdboxes (raid 5).
> We have almost 3 TB of gzipped mdboxes, and our backup (bacula) is
> taking almost 16 hours to make a full backup. Our main problem is that
> indexes are backupd hours before mdboxes, and that bring us some
> inconsistencies. What is the "right" way of backing up for separate
> indexes and mdboxes?
If you put your indexes and mdboxes on a LVM (seperate volume group
obviously, because of the different storages both reside on), you can
snapshot both volumes at the same time, using an FD-based script, mount
them somewhere else (like /backup/data and /backup/index) and then you
have (nearly) all the time you want to backup them. And because of the
snapshots taken at the same time, no inconsistencies can happen.
(OK, not entirely true, since both snapshots are taken with a minimal
gap between them, but if you start the backup during off-hours the
inconsistencies should be minimal.)
If you don't use a LVM but have a storage which is able to do snapshots
on its own, you can also use this.
If you don't have a LVM and no storage with such functionality, then you
are kind of lost, as far as I can see.
My additional advise on the long time of full backups: don't do them!
Switch to accurate backups and only do incremental and differential
backups, replacing the full backups with virtual full ones. This will
greatly reduce the load on your mail systems and speed up the backup
operation at the same time.
But you have to use "accurate" type backups, or your virtual full backup
will grow and grow and grow, since Bacula without "accurate" is unable
to know about deleted files and will include them in every virtual full
Sigmentation fault. Core dumped.
More information about the dovecot