[Linux-cluster] Fencing problem on Cluster Suite 3.0.12: fenced throws agent error when invoking fence_xvm

[Victor Ramirez <victor ramirez prhin net>]

I cannot get fence_virt in multicast mode to fence automatically even though I can do it manually with the fence_node command.

Lemme start from the beginning. I have a 2 node cluster, each node is a kvm guest running on a different physical host. All machines are RHEL 6 x64 and the cluster suite version is 3.0.12. 

Like I mentioned before, cluster is configured correctly since I can fence manually with the fence_node command, but when I trigger fenced to call fence_xvm automatically, fence_xvm fails silently with error status 1 and no multicast packet is sent. fence_xvm command does not write its output anywhere when invoked by fenced so I cannot know why it fails, but I suspect that it may be trying to use serial communication instead of multicast.

During troubleshooting, I made a script to run in place of fence_xvm in order to write the piped arguments into a log file and the arguments seem to be correct:

domain=prhin01-vm01

nodename=prhin01-vm01

agent=fence_xvm

debug=5

Also, I used tcpdump to determine that no multicast packet was being sent by fence_xvm.

I downloaded the fence-virt code but I am not too keen on debugging linux C code as I am a lowly Java webapp developer.

More info can be found here: https://access.redhat.com/discussion/fencing-problem-cluster-suite-3012-fenced-throws-agent-error-when-invoking-fencexvm

What else can I try to coax fence_xvm to work?

How can I make fence_xvm write to its output to a log file?

Can I call fence_virt instead and use a parameter to force multicast mode?

Should I give up multicast and try to use some vmchannel scheme?