HA Configuration for primary and failover sites

[fcapizzo]

I am currently working with a client that would like to use HA in a specific manner. Let's say that we have 5 ESX 3 servers at the datacenter and another 5 running at a DR site, and the sites have a fast connection between each other. Now let's say I add all 10 ESX hosts to a cluster with HA enabled. Is there any way to make rules for HA that define a primary set of servers to use for HA? For example, if a host (or even 2 hosts) go down at the primary datacenter, we wouldn't want the VMs to start up at the DR location.

I know there's the das.defaultfailoverhost advanced option for HA, but I don't know if you can configure it for a group of servers.

[surferdave]

HA only works within data centers and will only work if ALL ESX servers can see the shared VMFS LUN, so unless you are running iSCSI or Fabric across a WAN, HA wouldn't work at the DR location.

[Eddy]

No shared storage, no HA...

[fcapizzo]

Yes, I realize that shared storage is a necessity for HA. Perhaps I should have clarified in my example. So now I will; let's also assume that shared SAN storage is available to ESX servers at both sites.

[fcapizzo]

Anyone have any ideas on this? Personally I don't think that the current VI3 configuration allows for what I'm asking, but I just wanted to verify with everyone here. Also, if anyone is trying to do this kind of thing between 2 datacenters and has their own solution I'd like to hear what you've done.

[frankdenneman]

I was asked the same question by a client.

Ít's not exactly what you want, but maybe it will give you a few ideas.

I "solved" this by creating two cluster, one in each datacenter.

Only host failover occurs inside the cluster, this way the VM load of host b from cluster 1 will be picked up by host a or host c when a failover occurs.

Each HA cluster has disposal of a "local" SAN. The LUNs presented to cluster a are being replicated to the SAN at the other datacenter and vice versa. Cluster 1 uses primarily SAN 1, Cluster 2 uses primarily SAN 2.

When a SAN fails, the customer can issue a failover command on the san. This way all ESX hosts from both clusters will access one SAN. When a site\datacenter fails a failover will be issued and the other cluster can startup VM's, which ran on the failed cluster.

It's not fully automatic, which was a requirement for my customer. You don't want to accidentally failover a complete san and ESX cluster load.

I left lots of details out, PM me if you want some more info.

[fcapizzo]

Frank_D, I've done something similar to your setup at a previous client's site. It's always going to be a manual process to fail over a replicated LUN, and we had also documented the steps needed to have the ESX servers at the other site recognize the failed-over LUN and subsequently the VMs running on those VMFS volumes.

At this point I'm mostly certain that the idea I had proposed in the first post is simply not doable with the way HA works right now. So for the time being, failover methods with manual intervention are required.