ServiceGuard and NFS

[Tony Romero]

When I failover from host1 to host2 one of the many NFS filesystems does not unmoun, which causes the package to fail when it tries to come up on host2. Any ideas or familiarity with this type of problem.

The landscape is as follows:

host1 - prod db server
host2 - qa db server (failover server)
app1 - app server host
app2 - app server host
app3 - app server host

The NFS mount point on the app# servers becomes stale, which also prevents the app from coming up. However when the nfs.server and nfs.client is stopped or servers rebooted everything works well again.

Thanks...

[Robert Gamble]

I have a few questions to ask that may clear up your issue..

1. Is the filesystem to be unounted in /etc/exports ? It shouldn't be ...
2. Are you using the current NFS Toolkit for MC/SG ?
3. In the pkg.cntl script does the unmount_fs function include a 'fuser -ck /mount_point' ? It should ...
4. The mount point isn't listed twice is it (in the pkg.cntl) ?
5. What type of mount are you using from app# server, hard or soft ? probably should be soft ...

Good Luck!

[Robert Gamble]

Tony,

I just thought of one other question.

Are the app# servers pointing to the package name / IP address and not the cluster node that is down ? They should be pointing to the package name / IP of the package, not the hostname or IP of the server.

Good Luck!

[Tony Romero]

Thanks for your questions:

1) Filesystems are being mounted using automounter.
2) Using the native NFS nsf.server and nfs.client not the SG NFS.
3) fuser being used to kill processes using the mount points.
4) I don't believe pkg.cntl has duplicate entries.
5) App server are mounting using automounter as well.
6) Apps point to the cluster's hostname and IP. (i.e. hostname goes away once package goes down).