serviceguard and nfs toolkit

[Eugene Fleischmann]

I have an HA package consisting solely of a filesystem (one lvol in one vg) that is NFS exported for mount on other servers. When I failover the (NFS) package to the other node in the two-node cluster, sometimes the umount hangs. I am not using the NFS toolkit for serviceguard and am wondering what it contains that will make the failover work every time. In my package control script, I simply use umount to un-mount the filesystem in the customer_defined_run_commands. What does the NFS toolkit (an optional ServiceGuard product) provide

[Duncan Edmonstone]

Eugene,

The NFS Toolkit is just a bunch of scripts to save you some work in this process... If you want scripts which are 'supported' by HP then it may be worth looking at... if you are confident in supporting your own script then don't bother.

Re your problem... I think we can hazard a guess at what the issue is - you certainly shouldn't be un-mounting the file system yourself in the customer defined halt commands section (I presume this is where you meant rather than the run command section). Take a look at the package control script, you should find a FILESYSTEMS section in here where you list any filesystems that are part of the package, e.g.:

LV[0]="/dev/vgsg01/lvol1"; FS[0]="/pkg1/mnt1"; FS_MOUNT_OPT[0]="-o delaylog"

As long as a filesystem is detailed in here, ServiceGuard will take care of mounting and unmounting your filesystem as the package is started or stopped. ServiceGuard will also run an fuser -ku on the mount point to remove any processes accessing the file system whcih would prevent the umount from working.

HTH

Duncan

[Robert Gamble]

Eugene,

I have been running a NFS package for about two years. It was initially setup using the NFS toolkit. It was modified recently because it wasn't unmounting Filesystems that were still in use by an application.

In the unmount_fs function of the pkg.cntl script, we had to add another counter for how may attempts at unmounting we had tried. The original NFS toolkit we had (from 1999) did not use 'fuser -ku'. It does now =)

HTH, Good Luck!

[Eugene Fleischmann]

thank you for your answers, they were very helpful. As a followup, during package failover when "fuser -ku" kills all processes on the server accessing that NFS filesystem does the NFS client re-connect to the NFS server transparently to the user process on the client and re-open all files that were open when the package went down?

[Wayne Green]

I hacked the nfs toolkit script around to suit our needs so that all it does now is export and mount the FS to all nodes on start and fuser -k, unmount and un-export FS on stop.

One thing I did do was write a startup script to mount the NFS filesystem if the package is available in case the clients are booted after the NFS server.

On your last point I think if the package is switched the application is stopped and started and users have to start from scratch.