Zimbra NE Redundancy and Disaster Recovery

[akertis]

I know something like this has been asked many times. Right now we are running Zimbra on a physical server and I want to see if there are good options to make sure there is minimal downtime if there is hardware failure or data corruption on the server itself.

How does everyone make sure that they can get backup quickly if there is hardware failure. What about data corruption ? My understanding is the full backups will backup everything including LDAP and the mysql database. Is this enough to get up and running on a new server?

I think virtualizing zimbra would allow us to survive hardware failure. My only concern is using something like NFS or iSCSI can it handle the I/O that zimbra needs to operate without any problems?

My understanding is that something like Zimbra clustering requires RHCS and that is primarily for load balancing. Is that correct?

Thanks!

[bdial]

zimbra NE backups are enough to do a disaster recovery, although it can take some time to doa full restore of all the data depending how many users you have

zimbra runs fine virtualized with iscsi/nfs environments, you just need to make sure it's robust enough to handle the iops. we run it on vmware esx + hp/lefthand iscsi san.

zimbra cluster is only 'officially supported' on rhel, but i hear it works fine in centos & ubuntu

[akertis]

OK. I have a question about backups.

I was looking at the backups and I noticed that the last full is about 35 GB but the previous Full backups were like just over 2 GB. I would think that 35 GB sounds right if it is backing up everyone's mailbox and other stuff like LDAP and MYSQL. Is there a reason why the previous fulls would seem to be so small? Does Zimbra touch the full backups after it gets the next full?

[bdial]

zimbra does a full backup every time you add a new user. so if you do a full backup on friday, add a new user on tuesday it will do a full backu pof just that user tuesday night. so for tuesday you'll see both a full and incremental backup for that day, and the full will be small (unless you imported a ton of e-mail).

this command will give you a good overview of how much space each backup is taking up

du -m --max-depth=1 /opt/zimbra/backup/sessions

[akertis]

Well from looking at the sessions it seems it is doing a full backup every friday morning. I can't believe all the mailboxes are only 2 gb. The whole directory is only like 46 gb with the last full backup being 36 gb. That just seems odd to me.

[ewilen]

Something to be aware of about full backups: if you're using ZCS 5 or lower, then by default the full backups will hardlink to files contained in earlier fulls. This makes the backups a lot faster and the marginal disk space utilization per-backup is kept down. As of ZCS 6, the default has been changed so that fulls are each zipped into a single file, without cross-backup hard links. You can change that back to the old behavior by tweaking the zimbra cron file directly. A forum search will turn up more background on this.

This thread summarizes and links the info I was able to gather about setting up a cold backup: Setting up cold backup server

A key point is that you can greatly reduce the time it takes to recover, by pre-restoring the full as soon it's copied to your backup machine. Then you just restore the subsequent redologs contained in incrementals, /opt/zimbra/redolog/archive/ and /opt/zimbra/redolog/redo.log, and this can be done very quickly using zmplayredo instead of zmrestore. Again, look at the notes & links.

[mfvm]

A key point is that you can greatly reduce the time it takes to recover, by pre-restoring the full as soon it's copied to your backup machine. Then you just restore the subsequent redologs contained in incrementals, /opt/zimbra/redolog/archive/ and /opt/zimbra/redolog/redo.log, and this can be done very quickly using zmplayredo instead of zmrestore. Again, look at the notes & links.

As far as I understand BUG#31536 it is necessary to do a full restore if there is a new user. Is that correct?

[ewilen]

Basically, yes. But you have to (ought to) do a full restore anyway. If users have been added since the last full backup (of everybody) then they'll be in an "ad-hoc" full performed when the the following incremental ran. So you may need to restore more than one full to get everyone.

If you never have any new users you might be tempted to just restore once from a full and then do zmplayredo from then on, but I'd rather work from a recent full. My opinion might change if that bug is fixed.

[iway]

We have found so far that restoring LDAP daily from the daily full/incr beckups is suffcient. Then zmplayredo can populate the new accounts (without full backups) with the redolog data.

[ewilen]

We have found so far that restoring LDAP daily from the daily full/incr beckups is suffcient. Then zmplayredo can populate the new accounts (without full backups) with the redolog data.

Using zmrestoreldap?