HP StorageWorks XP Cluster Extension Software Administrator Guide (T1656-96035, April 2010)

3. Restart the node that was shut down.
NOTE:
The time to detect a storage outage due to failure of all paths to storage depends on the
setting for no_path_retry in the multipath software configuration. A value of fail does
not queue I/O in the event of a failure in all paths and returns an immediate failure. For
information about the recommended value for your environment, see the DM-Multipath
documentation.
Some resource agents, such as LVM, offer a mechanism called self_fence to take themselves
out of a cluster through node reboot when an underlying logical volume can no longer be
accessed. For supported options, see the RHCS documentation.
A storage replication link is down (RHCS)
If an Cluster Extension configuration uses DR groups with failsafemode enabled, the array disables
access to the disk when it cannot replicate the I/O to the remote array.
In this situation, if a replication link is broken, the resource agents of configured resources, such as
lvm or fs, may be able to detect and take appropriate actions. The stop operation on a service might
fail due to the inability to stop individual resources cleanly because the disk is no longer accessible
for read/write operations. When the stop operation on a service fails, RHCS marks the service as
failed and the service does not automatically fail over to another node.
To recover from this situation, use the following procedure:
1. Remove the node that lost access to the storage by shutting down the node.
2. Follow the steps required to bring up a service in a failed state, as documented in the RHCS
administration guide. This process involves disabling the service, and then enabling it on the
node where the service is allowed to come online.
3. Restart the node that was shut down.
NOTE:
The time to detect a storage outage due to failure of all paths to storage depends on the
setting for no_path_retry in the multipath software configuration. A value of fail does
not queue I/O in the event of a failure in all paths and returns an immediate failure. For
information about the recommended value for your environment, see the DM-Multipath
documentation.
Some resource agents, such as LVM, offer a mechanism called self_fence to take themselves
out of a cluster through node reboot when an underlying logical volume can no longer be
accessed. For supported options, see the RHCS documentation.
A data center is down (SLE HA and RHCS)
RHCS and SLE HA expect an acknowledgement from the fencing device before services are failed
over to another node. In the event of complete site failure, including fencing devices, clusters do not
automatically fail over services to surviving cluster nodes at the remote site. Manual intervention is
required in this situation. For instructions on bringing a service online, see the cluster software
documentation.
Troubleshooting158