Hot Backup Community Edition w/ DRBD

[dmart]

I realize hot backups are not supported in the FOSS version of Zimbra, however this is a requirement for our deployment (cold backups just aren't good enough). I think I have come up with a solution and just wanted to run it by some people here before I start buying more hardware.

I have /opt on it's own LVM volume. Setup DRBD to replicate this volume to some other node also running LVM. A script on the backup node would halt replication, take a snapshot and start replication again (DRBD would automatically get all updates it missed). Then we can do whatever we want with the snapshot.

Any reason this wouldn't work?

[raj]

In theory YES this works
But in my expriance DRBD in itself tough very nice software have a habbit on craping on you if you play with it too mich
i.e halting replication 365 times in 1 year for example for everyday backup. chances of you getting split-brain are more than someone who is just "running it" and dont touch a working (replication) every 24 or 12 hours.

Just my experience..i still find rsync multiple times and stop and start services a best overall method for OSS version Open Source Edition Backup Procedure - Zimbra :: Wiki

Even in very heavy (100 GB + of data) and 24x7 operations with 1000's of users on a system..there is a 2-3 min windows which no-one mind and cold-backup during that time just work fine and have less moving parts

if DRBD will stop on you for some reason..process of recovering from it will be > 2-3 min for sure, may be in hours.

Raj

[dmart]

Replication would be 1 way, the backup server would never actually be brought up. If the live server goes down, it's down. This is purely for backing up the /opt/zimbra directory.

[raj]

split brain problem from what i dont have anything to do with if you want to do 1 way or not.
most of the time its only 1 way failover and in case needed you can sync the other way when primary comes back.
* have you tried zimbra in DRBD environment and made it work or you are in planning stages. before you spend money please make it work and test in LAB first.

* if you buying new HW for this project then my suggestion would be to virtualization route and spend money on buying a good VM backup software and do hot backup of VM itself and leave zimbra runnning as is without DRBD.
or better use HA function of virtualization itself.

Zimbra in VM environment works just fine

raj

[dmart]

If the data replication fails any given night I really don't have a problem with invalidating the backup node and doing a full sync. That being said is that the only thing I have to worry about? I know this isn't ideal but I have explained the downsides (as far as I can see) to my boss and he doesn't seem to care.

edit: before the backup node would be brought back online it would be set as secondary and all changes would be discarded, forcing it to sync up with the primary. the script would be manually recovering from the split brain, the backup node would always be considered the 'incorrect' node.