Clustered file system / NFS Share

[Legrandmeitre]

Hi all,

I encountered a problem whyle trying to use NFS share resource, based on clustered file system. The configuration of the cluster is: two servers (Solaris 9 is installed on them) with the same three luns from clariion. Two luns are in one disk group, one volume, with shared file system; and one lun is in the second disk group, the second volume - simple vxfs on it. Two service groups were created - Clusterservice (default) and cvm group (parallel) for CFS. This configuration works fine. Then I created the third service group for NFS Share. The resources in this group are correspondingly: NIC - IP - DiskGroup - Mount - Share - NFSRestart + NFS. The situation is as following: The Share resource is placed on CFS volume which is mounted with cvm service group!!! NFSRestart LocksPathName was set to the second volume, which is mounted in this service group (NFS SHARE) before (NIC - IP - DiskGroup - Mount - Share - NFSRestart + NFS).

Then on the third server (not in the cluster) I mounted this NFS share and manually switch this service group (NFS SHARE) to the another server in the cluster.

The problem is that after failover the NFS share on the client (the third server) is not responding, though it is still mounted and dfshares shows the share on the remoute server in the cluster.

It seems to me that smth is wrong with NFSRestart or with dependencies between service groups...

Any advice will be appreciated!!

Thanks,

Pavel

[TomerG]

Hi Pavel,

Not sure here, but one issue around NFS clients not able to use a share after failover is stale NFS handles.

This is often caused by a mismatch between the major number of the file systems' underlying device. In this case you are using Volume Manager for the underlying devices (volumes) so you need to ensure that vxio and vxspec major numbers (in Solaris that's in /etc/name_to_major file) match on the multiple NFS servers. If this number doesn't match, clients won't work correctly after failover without a reconnect.

Don't ask me why NFS records major numbers... it probably has no such need to do so, but it does, so we have to deal with it.

Note: those major numbers are assigned during software installation time, so it is very possible that Volume Manager (Storage Foundation) on the various servers do not have the same numbers. 

[Gaurav_S]

Hello,

Also, could you double check the dependency tree of resources ? You can find this information in sample configuration in VCS Bundled Agents guide....

Gaurav

[mehers]

In the VCS Bundled Agent reference guide, there is a section on the usage of the NFSRestart resource and the dependencies that need to be setup within the service group.

You can set the NFSLockFailover attribute to 0 to disable the NFSRestart functionality and retry the test. If the problem persists, then it is not a problem in NFSRestart, but probably in the dependencies.