[Linux-cluster] RHEL cluster 5 fencing problem

Branimir bpejakov at fina.hr
Tue Mar 3 10:22:32 UTC 2009


Hi list,

I am testing RHEL 5 cluster for production (two-node cluster). So far, 
so good :) I simulated network outage and fencing works just fine here. 
But there is a problem. When I simulate power outage - literally I pull 
the plug from one of the nodes - in that moment, the node that is still 
alive tries to fence the node that is powered down.. and it tries, and 
it tries but it can't do that because the node that is powered down is 
powered down :) The living node cannot acquire cluster resources until 
it fences the dead node (which is powered down) and those retries go to 
infinity.

I was told to use tie breaker IP address to solve this. But as I 
understand this (thread "[Linux-cluster] two node cluster with IP 
tiebreaker failed"), I have to use quorum disk to do that - or better to 
say to use heuristics with ping, for example). Do I really have to use 
quorum disk for this case? Is there some other solution?

I would appreciate any hint!

Thanks in advance,

Branimir




More information about the Linux-cluster mailing list