On 04/12/2012 11:18 AM, emmanuel segura wrote:
That's right you'll found your cluster partitioned and if you "<cman two_node="1" expected_votes="1">" as redhat setting our cluster maybe you get data corruption
GFS2 and rgmanager depend on fencing completion prior to service recovery or GFS2's journal recovery.
Most two-node cluster partitions are caused by things like cable pulls. What normally happens is the node that lost its network cable then fails to fence.
No fencing, no recovery.
Forse the fencing problem redhat implement a work around as permanent solution fence delay for some cluster agents For fence_scsi in Redhat 5.X the redhat support says it's ok for production BAAAAA not tre
Delays simply let you figure out which host "wins" if the partition was caused by a temporal network outage or if the fencing devices are on a network that is not used by the cluster for communication.
AND i the redhat technical tells the cluster don't require the quorum disk UMMMMMMMMMMMMM
Quorum (whether augmented by something like qdiskd or not) doesn't prevent data corruption. Fencing does.
A node which does not have quorum can -still- write to disks, which is why you must fence it prior to doing anything else.
As an aside, qdiskd requires write access to storage before it can even make up its own mind about quorum.
-- Lon -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
