HSM SCheduling

[yellowhousejake]

Good morning,

We are preparing to begin using HSM. I have created and mounted a 4t drive in vsphere. The drive has been added as a volume to Zimbra and made active, set to secondary message volume. The policy has been set and all that needs to happen is to run a HSM job.

We use Veeam for our VMWare backups, which creates a snapshot before each backup job. Do we need to take any precautions about scheduling the HSM jobs to avoid running when a snapshot is being taken? We have had server processes that had issues when running during snapshots in the past and we do not want to have such as issue with our email.

Zimbra 8.0.7 Network Edition
VMWare 5.5.0
Veeam 9.5

Thanks,

Dave

[yellowhousejake]

So a short pause during a snapshot will cause no ill effects then?

[L. Mark Stone]

Running a Veeam snapshot in the middle of an HSM job executing seems like something you don't want to do, especially when you can schedule HSM policy executions via cron, or just do them manually.

If it were me I'd do HSM first, and then a Veeam snap. If you are nervous about HSM failing, then do a Veeam snap both before and after HSM executes.

Hope that helps,
Mark

[yellowhousejake]

Yes that does help. We supply email, VPN, Network, and phone for Police and Fire 24/7, so we have a limited time frame we can have services down for maintenance.

I suspected it would not be a wise choice to allow anything else to be accessing the Zimbra volumes during HSM. It just means that we have to run HSM manually or take extra precautions to ensure that Veeam does not start before HSM runs, or does not run over and go into HSM's timespace.

Thanks

DAve

[L. Mark Stone]

Yes that does help. We supply email, VPN, Network, and phone for Police and Fire 24/7, so we have a limited time frame we can have services down for maintenance.

I suspected it would not be a wise choice to allow anything else to be accessing the Zimbra volumes during HSM. It just means that we have to run HSM manually or take extra precautions to ensure that Veeam does not start before HSM runs, or does not run over and go into HSM's timespace.

Thanks

DAve

I wasn't really concerned about Veeam compromising an HSM job in progress. I was more concerned about how your Veeam job performed while HSM was running, if you ever decided to restore from it, you might find errors.

Remember what happens during an HSM job: Mail blobs are moved individually from one disk to another, and as part of each move, MySQL is updated. (Recall that MySQL is what is used to connect specific mail blobs on disk with what the user sees in the web or other client interface).

So if Veeam backed up the HSM volume first in the middle of an HSM job, and then backed up afterwards the primary store volume and MySQL, between those two events more blobs would have been relocated to the HSM volume. Now you do a restore, and you'll find that your users get "random" missing mail blob errors.

Veeam is a great product, but we've found it not appropriate for use cases where much data is stored in RAM (like LDAP, MySQL...) and doesn't get written out to disk instantly. Sure, for Windows VSC-aware applications like SQL Server, Veeam is great. But for Zimbra, not so much in our experience.

The Network Edition backup in 8.6 and 8.7 puts mailboxes, even those as large as 75GB, in maintenance mode individually for at most a minute or so. Inbound email is queued in Postfix and even if the user is logged in the web interface at the time they will notice nothing unless a Compose autosave takes place or they try to manipulate data. The Next Generation backup in Network Edition 8.8 avoids that entirely (it uses a continuous backup methodology).

Perhaps you've found Veeam works for you, but I would urge you to test some restores quite thoroughly to be absolutely sure.

All the best,
Mark

[yellowhousejake]

Excellent information, thank you. How HSM works is a good thing to know.

Currently we have multiple levels of recovery with storage both onsite and remote. We use the Zimbra backups for all "everyday" uses and have had to use them twice in the past several years. We backup our volumes to Veeam in separate jobs. So the HSM volume, the backup volume, and /opt/zimbra are unique jobs to recover failed volumes.

As to restoring a full server, Veeam is our Hail Mary DR, all onsite volumes are failed and we need email now.

We will ensure we have no scheduling overlaps.

Thanks,

DAve