HP ServiceGuard support

[fmodola]

Hi everybody,

we're looking at the best architecture for our Zimbra integration.
As we have 2 sites, 1 HP EVA 4000 SAN bay on each site, I thought about using RedHat Cluster Suite with the EVA CLX extension in order to move the cluster node to the second site while a volume isn't available anymore ...

But it isn't supported by HP, which supports only HP ServiceGuard as a cluster service.

Is it plan to support ServiceGuard as you're very near from HP (other threads seem to indicate that you've done many works with HP teams !) ???

Another solution is to create scripts to make the same operations that the CLX can do. It has already been done for the VMWare ESX3 servers, but the switching time is really noticeable (about 2 minutes), but that's not the best solution ...

Moreover, it's possible to have one mailstore instance on each site, but is it possible to have the same configuration on the both instances, with one instance sleeping, until the above scripts launch it while it stops the first one ?

I think my post isn't really simple to read, I don't know how to express myself about this problem !

If anyone wants to comment my post, all constructive answer is welcome.

Regards

[Klug]

Franck, as I told you, the the official answer is "not supported" even if this setup is "theoretically possible".

There's currently no supported way (except the Red Hat Cluster Suite) to build a fully redondant, fail-over and disaster recovery solution with Zimbra.

Maybe we should look into using the EVA with no CLX but with GFS : using the two EVA to provide two storage volumes and access to this volumes through GFS. GFS would be the software that "syncs" the data between the two EVA (no CLX required) and handles the failover, etc.

Red Hat Cluster Suite is supported by Zimbra.
I don't know for GFS.

[jholder]

Originally Posted by fmodola

Is it plan to support ServiceGuard as you're very near from HP (other threads seem to indicate that you've done many works with HP teams !) ???

Klug is right, not at this time. If you wish, you can file an enhancement on this, and we can see if we can work something out.

Typically, we don't seek partnerships, they seek us out

-john

[fmodola]

Ok, I looked for solutions with GFS, but I remember that two Zimbra nodes can't write onto the same /opt/zimbra directory, can it ?

If it can, I think it could be interesting to use GFS to do that, but, I think that's not possible for the moment.

So, if I wait for the mailstores' sync between two nodes (it would be really nice to have it in version 5.0), there's no need to have a GFS volume ... isn't it ?

[Klug]

Unless you use the GFS not to have "two servers writing on the same volume at the same time" but to have GFS "write the data on the two SAN at the same time"...

I'm speaking of GNBD, from what I read, it's made for this (Versaflex).

[fmodola]

Doing a replica of the first volume on the seconde SAN storage is really simple with HP EVA mechanisms, so, no need to use other replication machanisms.

Is it planned to permit several cluster nodes to write on the same volume ???
It would allow them to have only one volume, read and written by all maistores, and replicated by HP EVA replication mechanisms.

[fmodola]

About mailstores' synchronization feature, if a cluster node shuts down, will the second node be able to write on the data written originally by the first node ???

[Klug]

Originally Posted by fmodola

Doing a replica of the first volume on the seconde SAN storage is really simple with HP EVA mechanisms, so, no need to use other replication machanisms.

GNBD is not doing replica, it's writing on the two SAN at the same time.

If you "really simple system" needs the CLX extension, you know this is "not compatible" with RedHat Cluster Suite, so not supported by Zimbra.

If it's doing an snapshot/replica from time to time, then it's just like using the suggested by Zimbra solution, ie using the backup/restore/redolog solution to sync the two datacentres.

Originally Posted by fmodola

Is it planned to permit several cluster nodes to write on the same volume ???

No, AFAIK.

Originally Posted by fmodola

About mailstores' synchronization feature, if a cluster node shuts down, will the second node be able to write on the data written originally by the first node ???

Mailstore sync feature (not available yet) is not to be seen (AFAIK once more) as a "cluster of two nodes" (in the RHCS definition of cluster) but as a replicated system somewhere else (more like LVS with failover/replication without load balancing).

[fmodola]

Mailstore sync feature (not available yet) is not to be seen (AFAIK once more) as a "cluster of two nodes" (in the RHCS definition of cluster) but as a replicated system somewhere else (more like LVS with failover/replication without load balancing).

If one server fails and shuts down, will the users automatically be switched transparently to the other synchronized servers ?

[Klug]

I don't know the planned feature for "replication".

But what I understand in "replication" is that it just provide "replication", not "failover" (so no automagic switch to the other server).
I can be wrong of course.