On Tue, Feb 26, 2013 at 4:44 PM, John R Pierce <pierce at hogranch.com>
wrote:> hey, I have an application for drbd replication between a pair of EL6
> servers, and I just realized that drbd is no longer built in.
>
> googling found me this blog on doing it using ElRepo distributions...
> http://www.broexperts.com/2012/06/how-to-install-drbd-on-centos-6-2/
>
> is that still best practice?
Yes.
> or is there a better distribution?
No
>
>
> or really... my application is a BackupPC server replicating its
> backup volume to a standby server (hey, no backup is really a backup,
> unless there is a backup of the backup, right?). is there a
> better/alternate way of maintaining a mirror of a backupPC volume than
> drbd... the backup volume is initially 8TB and could well grow to 3
> times that over time. its stored on an XFS file system, under LVM, on
> striped raid6 volumes.
DRBD is a distributed block device. It's very useful for offering HA
services, but much like raid, drbd isn't a backup. If I had two boxes
for doing my backups, I'd use a file level replication (like rsync)
to "backup up the backup" (or better yet, ship a snapshot of a COW fs
you ran an fs check on). It's too easy to trash a volume with an rm /
fs bug / split brain. It would make it harder to do an auto failover,
but i would be safer. If I miss a backup because the primary is
offline, I miss one backup cycle. If I screw up and delete a backup
off the drbd volume (or I hit a bug that corrupts the fs), I could
lose my entire backup set. I could even treat the second box as a
monthly "offline" box that I unplug from the network as "added
protection against worms and malware" or some other nonsense like that
if I had to justify it's existence.
That being said, if you have a requirement that your backup solution
is up five nines, then yeah, use drbd / pacemaker, it's just not a use
case I see very often.
Best,
Patrick