Failing back a file serving node
After automated or manual failover of a file serving node, you must manually fail back the server,
which restores ownership of the failed-over segments and network interfaces to the server. Before
failing back the node, confirm that the primary server can see all of its storage resources and
networks. The segments owned by the primary server will not be accessible if the server cannot
see its storage.
To fail back a file serving node, use the following command. The
HOSTNAME
argument specifies
the name of the failed-over node.
<installdirectory>/bin/ibrix_server -f -U -h HOSTNAME
After failing back the node, determine whether the failback completed fully. If the failback is not
complete, contact HP Support for assistance.
NOTE:
A failback might not succeed if the time period between the failover and the failback is
too short, and the primary server has not fully recovered. HP recommends ensuring that both servers
are up and running and then waiting 60 seconds before starting the failback. Use the
ibrix_server -l
command to verify that the primary server is up and running. The status should
be Up-FailedOver before performing the failback.
Using network interface monitoring
With network interface monitoring, one file serving node monitors another file serving node over
a designated network interface. If the monitoring server loses contact with its destination server
over the interface, it notifies the management console. If the management console also cannot
contact the destination server over that interface, it fails over both the destination server and the
network interface to their standbys. Clients that were mounted on the failed-over server do not
experience any service interruption and are unaware that they are now mounting the file system
on a different server.
Unlike X9000 clients, NFS and CIFS clients cannot reroute file requests to a standby if the file
serving node where they are mounted should fail. To ensure continuous client access to files, HP
recommends that you put NFS/CIFS traffic on a user network interface (see
“Preferring network
interfaces” (page 60)
), and then implement network interface monitoring for it.
Comprehensive protection of NFS/CIFS traffic also involves setting up network interface monitoring
for the cluster interface. Although the management console will eventually detect interruption of a
file serving node’s connection to the cluster interface and initiate segment failover if automated
failover is turned on, failover will occur much faster if the interruption is detected via network
interface monitoring. (If automated failover is not turned on, you will begin to see file access
problems if the cluster interface fails.) There is no difference in the way that monitoring is set up
for the cluster interface and a user network interface. In both cases, you set up file serving nodes
to monitor each other over the interface.
Sample scenario
The following diagram illustrates a monitoring and failover scenario in which a 1:1 standby
relationship is configured. Each standby pair is also a network interface monitoring pair. When
SS1 loses its connection to the user network interface (
eth1
), as shown by the red X, SS2 can no
longer contact SS1 (A). SS2 notifies the management console, which then tests its own connection
with SS1 over
eth1
(B). The management console cannot contact SS1 on
eth1
, and initiates
failover of SS1’s segments (C) and user network interface (D).
Cluster high availability
29
