Preferred and Possible Owners conundrum

Essendon

This preferred and possible owners thing has me going in circles. Cant get my head around this, the problem being clustering can be labbed only on SCSI drives or Fiber Channel and I dont have either. (I have seen the link that astorrs posted about using VMware's virtual SCSI capability).

I know that Possible Owners are those on which a resource can be brought online. Preferred Owners are those on whom a failover can be initiated. I know that Active/Active (A/A) should have 2 EVS's running on either node (in a 2 node cluster) and Active/Passive (A/P) should have 1 EVS.

Now, let's say a node (Node1) fails on a A/A and for the other node (Node2) to take over the EVS running on Node1, BOTH nodes need to be Possible Owners, correct? In this scenario, what node should be the Preferred Owner?

Similarly, in a A/P configuration, say Node1 is running EVS and when it goes down, Node2 takes over. In this config too, BOTH nodes need to be Possible Owners, correct? If only one node was a Possible Owner then the resource would not failover. I am unsure what a Preferred Owner would be in an A/P config. Would both nodes need to be Preferred nodes? This Preferred Owner thing is tripping me up.

Please help.

royal

Possible Owners are on resources. Preferred Owners are on Groups. If you have 2 nodes in an Exchange Cluster, you would have each resource in the Exchange CMS (Virtual Server) be a possible owner of all resources so either one of those nodes can own it. If you wanted to have automatic Failback to Node1, you'd have Node1 on the top of the list of Preferred Owners.

You can get pretty fancy with all of this in large Clusters with Multiple Nodes such as if you had Node1, Node2, Node3, Node4, Node5. You can have 5 Cluster Groups where all 5 Nodes are Possible Owners so they can own every resource in all Cluster Nodes. But you'd have the Preferred Owner for Node 1 be Node1 then Node2 then Node3 then Node4 then Node5. Then for Node2 you'd have Preferred Owner as Node 2 then 3 then 4 then 5 then 1. So if Node1 went down and Node2 was also down, it'd go to Node 3. If Node3 went down, it'd first try to failover to Node4.

Does that help?

Essendon

royal wrote: »

Possible Owners are on resources. Preferred Owners are on Groups. If you have 2 nodes in an Exchange Cluster, you would have each resource in the Exchange CMS (Virtual Server) be a possible owner of all resources so either one of those nodes can own it. If you wanted to have automatic Failback to Node1, you'd have Node1 on the top of the list of Preferred Owners.

That's a big help royal. There was a question in Transcenders that went something like:

You have an A/A setup with both nodes running EVS. When Node1 fails, Node2 should take over both EVS's. When Node1 comes back up, it should be returned its EVS.

So in this question, both nodes should be Possible Owners so that EVS can failover to the working node. But there should be only one Preferred Owner so that when Node1 came back, it should get back its EVS. I think I am right?? Also by your explanation royal, let's say there was a 3rd node too. So if I wanted that Node1's EVS to failover to Node3, I would make Node3 the Preferred Node. Am I right??

Essendon

Anyone?

royal

Ya you're right by the 3 node. You'd need all 3 nodes to be possible owners of all the resources. But if you wanted Node 1 to have the priority and then Node 3 and then Node 2, you'd have in the Preferred Owners list Node1 then Node 3 then Node 2. If it's 2 Actives and you wanted failback, you'd have Node1 as the Preferred Owner and then enable failback.

Essendon

Thank you for clearing it up.

Cant wait to tackle the test. I do it on Thursday.

royal

No problem. Another thing to take into consideration is that if you have a bunch of Nodes but no Preferred Nodes, the Clustering Service will choose a Node to failover to randomly.

Essendon

royal wrote: »

No problem. Another thing to take into consideration is that if you have a bunch of Nodes but no Preferred Nodes, the Clustering Service will choose a Node to failover to randomly.

I'll remember that. There were two questions in transcender on clusters with no Preferred Nodes which helped drill this into my already overloaded brain.