NFS Cluster Failover

[Serge Dubrouski]

Sorry I wasn't able to answer to the old thread.

I had exactly the same problem. umount cant' unmount those filesystems
while NFS is up. To bypass it I had to modify
/usr/lib/clumanager/svclib_filesystem script and add 2 following
functions there:

stopNfs()
{
        perl -n -e 's/(.*)\s(.*)\((.*)\)/exportfs -o $3 $2:$1/; print
$_;' /var/lib/nfs/etab >
/tmp/exportfs.$$
        service nfs stop
}

startNfs()
{
        service nfs start
        sh /tmp/exportfs.$$
        rm /tmp/exportfs.$$
}


Then I changed "stop" part there:

        'stop')
          stopNfs
          stopFilesystems $svcID
          startNfs
          return $?
          ;;

Of course it arises the problem with a short outage for all NFS
resources while you need to stop or move just one but it gives you
better flexibility on the other hand.

Serge Dubrouski.
>Hello,

>We are currently deploying some servers in a failover situation for high
>availability using RHEL AS 3. I have followed the Red Hat Cluster Suite:
>Configuring and Managing a Cluster guide to set up a clustered NFS
>service, but failover just does not happen.
>
>We have two nodes, hydradb4 and hydradb5 with a virtual ip that maps to
>the name hydradb. They both have access to a shared disk /dev/emcpowera3
>with /dev/emcpowera1 and /dev/emcpowera2 being the raw devices for the
>cluster shared state. At this stage I can easily move the service
>between the two nodes with no problem and the disk moves with it with
>ease.
>
>/dev/emcpowera3 is now being mounted on /export/oracle/PRD/920_32 and is
>being shared over nfs as /export/oracle/PRD/920_32. Again, I can move
>the service safely from node to node without a problem.
>
>When any host has hydradb:/export/oracle/PRD/920_32 mounted however, if
>I try to move or stop the service I get an error:
><err> service error: 'umount /dev/emcpowera3' failed
>(/export/oracle/PRD/920_32 ), error=1
><err> service error: umount: /export/oracle/PRD/920_32: device is busy
><err> service error: umount: /export/oracle/PRD/920_32: device is busy
><err> service error: Cannot stop filesystems for hydradb
>
>I have the Force Unmount option checked for the nfsexport in the service
>as recommended.
>
>Does this error sound familiar? Does anyone know how I can get the
>clustered nfs service to work as advertised?
>
>- Thanks in advance,
>Richard Bailey,
>Technical Analyst