topic create mount point in cluster in Operating System - HP-UX

create mount point in cluster

joseph51 — Tue, 24 Feb 2009 12:27:46 GMT

hi
I am configuring a MCSG with vxvm.the customer want /sap/tmp in storage disk & /sap/tmp/run in local disk.
i tried a lot but the package is not running. finally i caeated same mount point in local and in storage also.. now the package is running.but when the primary node is halted i am not able to view the file(i created one test file) in local mount poing.
in adoptive ,i am able to view the file because i creted the same mount point here.

is there any solution to make this proper.
customer want /sap/tmp as a shared filesystem(in cluser) and /sap/tmp/run as a local file system

please help me

Re: create mount point in cluster

Duncan Edmonstone — Tue, 24 Feb 2009 15:13:10 GMT

"That dog won't hunt" as A Clay used to say...

You can't have a local filesystem nested in a clustered filesystem.

Use a symbolic link instead . Make your local filesystem something like /local/run then create a link nin your clustered filesystem thusly:

ln -s /local/run /sap/tmp/run

Incidentally are you using Serviceguard Extension for SAP? This takes care of all SAP clustering issues:

http://www.hp.com/go/sgesap

HTH

Duncan

Re: create mount point in cluster

Matti_Kurkela — Tue, 24 Feb 2009 15:33:45 GMT

Your problem is that you must unmount the local /sap/tmp/run in the package configuration before unmounting the shared /sap/tmp, or the umount operation fails. If the umount fails, the VG cannot be deactivated and the package failover cannot happen. The order of the mount and umount operations is important.

There are many ways to fix this.

I would recommend using symbolic links.

First, mount the disk that would be the local /sap/tmp/run to some other location, for example /var/saprun. Do this on all nodes of the cluster. You can choose any path for this mountpoint, but the path must be the same in all nodes of the cluster.
Do not include /var/saprun to the package configuration at all; make it a normal local filesystem, and add it to /etc/fstab as normal.

Halt the package. As the package halt unmounts the shared disk, /sap/tmp will now be an empty directory.

Create a symbolic link pointing from /sap/tmp/run to /var/saprun on all nodes of the cluster:

ln -s /var/saprun /sap/tmp/run

Now mount the package disks (e.g. by disabling the software startup scripts and starting the package).
As the shared disk is mounted over the existing /sap/tmp, it will hide the symbolic link from view. Now you'll have to make one more symbolic link on the shared disk:

ln -s /var/saprun /sap/tmp/run

This creates another symbolic link which is exactly the same as the link hidden by the shared disk.

Now, when the package is running, the local disk /var/saprun is accessible as /sap/tmp/run through the symbolic link included in the package... but because it is a symbolic link, it does not prevent the package from failing over. When the package fails over to another node, another symbolic link is uncovered and /var/saprun is still accessible locally as /sap/tmp/run.

MK