3.7.0 HP StorageWorks HP Scalable NAS File Serving Software administration guide - HP Scalable NAS 3.7 for Linux (AG513-96002, October 2009)

of the NFS servers that once held locks for the client. If the power fails, these files
are used by the statd process when it restarts to send an SM Notify command to
each of these servers, releasing all locks previously held by this NFS client.
If an NFS client is powered down while holding a lock and cannot come back up,
the locks for this client will continue to be held on the server. This is expected behavior
for the NFS/NLM/SM protocols. These locks will persist until either the NFS server
is rebooted or, in the case of a held lock obtained via a virtual host, the virtual host
is failed over to another NFS server node. With either approach, clients accessing
that NFS server via virtual hosts will be transitioned gracefully to other servers.
However, any other NFS client accessing the NFS server using its physical IP address
will need to remount if the NFS server should need to reboot.
There are known NFS client issues where a client reboot does not cause locks to be
released. In this case, use the same method of virtual host failover (or NFS server
reboot) to clear the locks.
After any server reboot or virtual host failover, all lock holding NFS clients have a
45-second grace period to reclaim their locks. Should a new client machine be used
to host the application, no locks will be available for this 45-second grace period.
You will need to determine which NFS server was affected by the outage to know
which Virtual NFS Server to rehost. You may not have access to the /var/lib/
nfs/statd/sm directory on the NFS client to view the IP addresses of the affected
servers. At this point, the lock failure itself should point you to the VNFS host that is
holding the lock. For example, if you try to startup your Oracle database and you
get a error indicating failure to obtain the locks, you will need to rehost the Virtual
NFS Server used for $ORACLE_HOME and those used for your Oracle datafiles. In
the case of Virtual IP addresses, check the HP Scalable NAS Management Console
to determine which NFS server is the current primary server for that address.
Virtualized NFSD RPC reply caches
FS Option provides support for virtualized NFSD RPC reply caches. When the feature
is enabled, the contents of the NFSD RPC reply cache are written out to a file when
a Virtual NFS Server (a virtual host) is removed from the node. By transferring the
cache file to another node before the virtualized server is reactivated there, the cache
entries can be preloaded on the destination node, giving a more transparent transition.
Without an RPC reply cache, clients that retry non-idempotent operations (operations
whose success invalidates their pre-conditions, such as create, remove, link, unlink,
rename, and so on) may see erroneous error returns. The conventional reply cache
prevents this problem. It is kept in memory on each node, with no provision to transfer
it between nodes or save and restore it. With the addition of virtualized NFS servers
HP Scalable NAS File Serving Software administration guide 197