Well, you're using iscsi so restarting network will impact access to the fs (a quorum disk wouldn't help either). Maybe timeouts are too short and the machines are queuing fence commands too soon. How is your fencing config'd? (net, same iface or private, serial) describe scenario more accurately: name machines A B C, it's not clear if two nodes are fencing each other or if the third one is.
From: linux-cluster-bounces redhat com <linux-cluster-bounces redhat com>
To: linux-cluster redhat com <linux-cluster redhat com>
Sent: Sun Sep 28 21:04:12 2008
Subject: [Linux-cluster] fencing loop
My cluster nodes are getting fenced when I simply just restart the network
service. What can I do to prevent that happening so quick? Can I even
prevent it since im using GFS?
I have a 3 node cluster running CentOS 5.2. They all are connected to my
iscsi SAN, so CLVMD is being used and GFS as well.
To make things worse, it appears that 2 servers were fenced when I simply
restarted the network service on one server. One server came back online,
the second one did, but then was immediately fenced again. Seems to loop
until I basically stop the cluster and have everyone rejoin.
Any help/ideas would sincerely be appreciated.
Linux-cluster mailing list
Linux-cluster redhat com