Zimbra/RedHat Cluster and LVM

[Rich Graves]

Related to my query at One + One and a half clustered NE (RHCS+VMWare)

RedHat Enterprise 4+RedHat Cluster Suite appears not to support clustering LVMs. ZCS 4.5.5 only supports RHEL4, not RHEL5. Even in RHEL5, CLVM is somewhat experimental.

Zimbra Cluster appears not to support multiple /opt/zimbra/storeN partitions. The choice is one partition or 10.

This appears to make it really difficult to expand storage over time. I can't simply add extents to a LV and ext2online to enlarge, because RHEL4 doesn't do CLVM. I can't simply add an additional Zimbra store partition, because it's not fuly supported by Zimbra Cluster.

How do people resolve this?

[Rich Graves]

Originally Posted by Rich Graves

RedHat Enterprise 4+RedHat Cluster Suite appears not to support clustering LVMs. ZCS 4.5.5 only supports RHEL4, not RHEL5. Even in RHEL5, CLVM is somewhat experimental.

OK, so 1.6.Â*Cluster Logical Volume Manager tells me that CLVM *is* supported in RHEL4 Update 5. Groovy. Following the bouncing documentation has been difficult.

Do other people rely on this working in a clustered Zimbra environment?

[ari]

Zimbra uses a simple failover model: only one mailbox store is using the partition(s) at a time. GFS and CLVM are only required where multiple cluster nodes require concurrent access.

So the answer is "not applicable" in Zimbra environment.

To grow/manage volumes on the fly, you could use SAN tools, zmvolume, and or LVM. CLVM/GFS are not required.

- Ari

[Rich Graves]

To grow/manage volumes on the fly, you could use SAN tools, zmvolume, and or LVM. CLVM/GFS are not required.

Here's the case I'm concerned about:

/opt/zimbra-cluster/mountpoints/mail.example.com is a LVM partition mounted on mail1.example.com.

It is running low on disk space, so mail1.example.com, using non-cluster-aware LVM tools, expands the volume.

mail1 fails. The service fails over to mail2.example.com, which is not aware of the LVM changes done on mail1. Assuming the volume mounts at all, massive corruption ensues.

How do you protect against that?

Hmm, now that I know that zmvolume or equivalent can and in fact do work (I just need to add another disk volume resource manually or with system-config-cluster), I guess I don't see a need to use LVM at all. I guess I can just create one ext3 partition on the LUN and be done with it.

[punit.jain]

You can use zmvolume but how about dbdata ? mysql cannot span multiple disks.