NFS / Automount with MCSG for SAP systems

[Lai Nee Shyang_1]

Hi there,

I've this "stranged" NFS / automount structure implemented for MCSG on SAP system.

It goes something like this: I've a this shared SAP directory called /sapmnt/SAP. The filesystem resides in /dev/vgsap00/lvsapmnt, and this is mounted on /export/sapmnt/SAP, which in turn is exported by the hosts "SAPA0" (MCSG floating host name) and mounted back onto the same host "SAPA0" at /tmp_mnt/sapmnt/SAP which is in turn soft link to /sapmnt/SAP.

I figured that this is for the purpose of retaining the filesystems exports on the failover server during failure of the primary server.

I find this rather inefficent. When any file is read or write to this particular directory, it has to "jump" 4 times before it reaches its destination. Is this a normal way how MCSG for SAP is implemented?

Lai

[melvyn burnard]

This is correct.
The main reason behind this is to overcome the problem that if a node fails, the umount/mount commands for the NFS file system will fail, due to Server Not Responding errors.

By implementing this, the umount and then the mount commands are able to succeed which ensures the SAP package will survive a failover.

This is a very simplified answer, of course, and you may find it worthwhile attending the ServiceGuard Extensions for SAP course

[MARTINACHE]

Hi,

I don't work with SAP but problems are the same.

I have a cluster with 3 nodes and 5 packages.

One of these packages is a NFS server (NFS Toolkit).

Supposing that PKGNFS is running on node A and export the filesystem /FSEXPORT.

When I start a package on node which needs the /FSEXPORT filesystem, here is the start procedure :

- check if /FSEXPORT is mounted (mount | grep FSEXPORT ...)

- if it's OK, I can start the package (this means that /FSEXPORT is already mounted, local or NFS).
- if not, try to mount via NFS the filesystem (mount -F nfs -osoft,intr PKGNFS:/FSEXPORT /FSEXPORT)
- if it's OK, I can start the package
- if not, I test if the NFS package is running (I wait 120 seconds for the package and I retry to mount /FSEXPORT)
- if this don't work, I stop the package.

Regards,

Patrice

[Lai Nee Shyang_1]

Hi Melvyn,

I understand how it works, but I do not agreed why it is done this way.

It is ineffiecient, having to propagate thru four reference points for reading and writing to the directory. It has multiple points of failures, ie softlink, NFS. It only serve the purpose of failover.

Isn't there a better way around this,getting rid of the soft links and mounting filesystems exported by the same host. I've asked the HP's MCSG implementors why it was done that way and they counldn't give me a firm and assuring answer.

Another question, is it a new trend to go for autonfs rather than the conventional nfs?


Lai

[Lai Nee Shyang_1]

Hi Patrice,

Thanks for sharing your method. Does the algorithm reside in the MCSG startup/failover scripts? If so, I believe this maybe one alternative to avoid my situation.

Thanks

Lai

Lai