NFS Cluster Server Issue

[MikeL_4]

I have two servers, nfs1 and nfs2 that are Clustered together using Red HAt Clustering...

nfs1 being the primary and nfs2 the failover server...

It has about five (5) file systems that are NFS mounted over the VIP IP that moves between servers during a failover to the client servers...

The issue is, when nfs1 drops the file systems and moves to nfs2, it leaves the client nfs servers hung up... If you try to access any of the NFS mounted file systems on the client servers your session just locks up...

Is there some sort of mount issue that I am missing that will help resolve this issue ??

[Sameer_Nirmal]

This sounds like a problem related to remote mount table synchronization.

I would look at the package control log and hanfs.sh.log to check if there are any errors/messages being reported.

[Matti_Kurkela · Accepted solution]

(Sameer: I think hanfs.sh is one of the ServiceGuard toolkits. Not at all related to RedHat Cluster used by MikeL.)

Have you specified the "fsid=" option when exporting the filesystem? If not, the fsid is auto-generated using a default algorithm which may give a different result on different nodes.

When a failover occurs, if nfs2 uses a different fsid for the exported filesystem than nfs1 did, all the currently-open file handles on the clients will become invalid.

If you're using RHEL 5 and have set up your cluster using Conga, it is supposed to set this NFS exportfs option correctly automatically. If you use an older version of RedHat Cluster or have set up your cluster using command-line tools, you should check this.

Also make sure your clients mount the NFS filesystem through the floating IP address. If the clients are using a node-specific IP for NFS connections, the failover will not be successful.

MK

[Ivan Ferreira]

Ensure that your NFS clients mounts the resource using the UDP protocol (udp option). If you use the TCP protocol, the session will hang in TIME_WAIT for 10-15 minutes.

[MikeL_4]

Thanks for replies... Will implement FSID for the NFS mounts.. what I've read on UDP, Red Hat uses it as the default protocol...