On Mon, Jul 31, 2006 at 01:20:07PM +0000, danwest@xxxxxxxxxxx wrote: > The out-of-band IPMI is basically a crossover cable to each others > IPMI interface. > > A simple test of pulling the public interface one the node with an > active resource results in each node being fenced. My understanding is > that in the 2-node configuration the fencing is a race to see who can > fence each other first. My concerns are the following: > > 1) If it is indeed a race it seems like both nodes successfully > issue a /sbin/fence_ipmilan (ipmi_tool) resulting in both nodes being > fenced. Because of this race you should use a network power switch that forces the two fencing operations to be serialized, then one will will. > 2) Is there a way for this algorithm to determine which of the > nodes was the one that had the problem? A simple race to fence could > result in the faulty server winning. No, that's another problem with the two node option. I think the new quorum disk feature that's in the next release should address both of these problems. > Also is there a way to configure fence_ipmilan in cluster.xml to reboot > rather than stop the server? fence_ipmilan by itself takes the –o > option (on,off,reboot) Someone else can probably answer that. Dave -- Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
