Service Guard : The package can't unmount File system

[Voldoo]

Hi all

Currently we are implementing SAP Service Guard for SCM (ABAP + JAVA) and then LiveCashe

The package for the SCM starts fine, but the Livecashe can't start for different raison
The first one is due to a XUSER (A call is open with SAP support) , but due to this the startup of package fails and then stop (normal)
But the problem is that the file systems can not be unmounted.

Apparently the ressource is busy

Any idea to bypass this situation and unmount the FS in any case.

Note :
The system is a RedHad Linux
We use hanfs files

Cheers

Yacine

[Emil Velez]

In older days this was due to a missing fuser patch that umounts the filesystem.

What OS and version of serviceguard. Does the serviceguard script fail on startup and what error message are you getting when it tries to umount the filesystem.

If there is a fuser patch that is what is usually executed to kill any processes on a filesystem and umount it.

[Andres Stickar]

In the control script (dbci.cntl) you will find:

# FILESYSTEM UNMOUNT COUNT
# Specify the number of unmount attempts for each filesystem during package
# shutdown. The default is set to 1.
FS_UMOUNT_COUNT=1


Change the number to 3 or 4

If the FS is busy, then an fuser command will be performed to release it.

If this do not fix your problem, probably you will have to review you SGeSAP configuration, ctrun and nfs.

Regards

A.

[Voldoo]

Hi all,

Yes the parameter FS_UMOUNT_COUNT was already set to 3.

I thinks it's more a configuration issue.
Indeed we have 2 packages which using NFS FS.

And we we stop one of them , the nfs deamon is stopped.

I don't know if an NFS package can solve this.

Cheers

Yacine

[Andres Stickar]

Yacine.

If you are using SGeSAP you must use NFS toolkit and automounter to manage NFS.

A.

[likid0]

Yep, SGeSAP uses HA NFS toolkit.

[Doug O'Leary]

Hey;

I run into this problem sporadically on HPUX MCSG clusters running sap. It appears SAP doesn't always shut down cleanly.

Manually, you can run fuser on the mount point to find out what processes are holding that filesystem open, then kill those processes.

On *very* rare occasions, I haven't been able to kill a process because it's waiting on I/O. I think that's happened twice in over three years. Normally, the process dies pretty quickly, after which, you can unmount the filesystem, vgchange -a n the volume group and failover/restart as you desire.

The automatic portion of this is supposed to happen in MCSG already as part of the fuser command sequence when it's unmounting filesystems. For me, this issue happens rarely enough that I haven't bothered tracking it down; I just unmount the filesystems manually and am done with it.

HTH;

Doug O'Leary

[Johnson Punniyalingam]

better if we could use fucer -ku "script" command before we umount the file system . in the pkg.ctnl.file,

Thanks,
Johnson