Relationship of IPMP on a node to cluster public IP services
I've setup a new two node cluster on a pair of T1000 boxes, Solaris 10 U3 (11/06), latest patch cluster, Cluster 3.2 software.The boxes are interconnected with a pair of cables between two of the four built in gigabit ethernet interfaces. The new how-to guides are great - almost everything seems happy in this simple configuration. The log files messages from the cluster product seem to indicate a happy system.
However, the cluster doesn't seem to be working as I expect for IP failover. I've configured a local IPMP setup with one public interface on each node - no test setup, but the docs indicate that's not required for a one interface "group". I defined a public IP address, and can ping it and telnet successfully to whatever node it automagically directs. I can also access the direct IP of each node without issue. However, if I pull the network cable on that node, the cluster is very unhappy in the message log, and does not seem to failover for those simple IP services to the second node.
What I'm missing, conceptually and technically, is the glue that connects IPMP on a local node with some sort of cluster IPMP group that will handle the failover as expected. The Cluster docs seem to mostly point to the Solaris 10 admin docs for IPMP - but setting it up locally seems fine. Do I need to use two interfaces on each node for the local IPMP group with the test addresses for success? What's the secret magic cluster command that I'm missing in this setup?
