Inexplicable disappearance of NFS exports

[Ralph Grothe]

Hello,

one of the systems I have to look after is a bunch of SAP/R3 servers, i.e. the main server is running on an SG cluster, and 9 other servers are some kind of replication, development, training, testing, or who knows what else servers.
SAP seems to require 10 standby servers at minimum for a sole production system.

The downside is that all the standby servers need to run endless transports (don't ask me what is concealed behind this term, it's SAP lingo)

Therefore the production server (i.e. the one that currently hosts the HA package) needs to export an nfs filesystem /usr/sap/trans to all the standby servers.
This all seems to be handled by the additional SAP kit for MC/SG.

On the standby/replication servers the NFS export has to be mounted through automounter.
Here is a sample dump of one of such auto_direct file:

# cat /etc/auto_direct | sed 's/[a-z]*:/sg_server:/'
/sapmnt/Z01 sg_server:/export/sapmnt/Z01
/usr/sap/trans sg_server:/export/usr/sap/trans
/Z01 sg_server:/export/Z01
/usr/sap/profiskal sg_server:/export/usr/sap/profiskal

In the past it happened a couple of times on a particular replication host that somehow the exported /usr/sap/trans became inaccessable or immountable to automounter.
When I had a look at the dump of exportfs on the exporting server at such incidents, I noticed that somehow miracally the host that was encountering the problem had disappeared from the list of exportees.
When I checked the
/etc/cmcluster//sap.conf
- just to make sure -
there was nothing wrong there, and all importing hosts correctly appeared as entries in the IPFS, OPFS, SXFS arrays.

Also a look at NFS statistics through nfsstat revealed no significant errors to me (although I have to admit maybe not to fully understand the statements of nfsstat)

Has anyone of you a hint what the cause of this could be?

Regards
Ralph

[Stefan Farrelly]

Ralph, this is going to be almost impossible to diagnose and resolve. I cant understand why you are using nfs in an SG cluster - talk about unreliable. Bill Hassel wrote this in a reply just a month ago;

"NFS on production systems, as you've seen, is not very stable. Essentially, your production system is critcally dependent on the NFS servers as well as the network to these servers to always be up and running. "

Basically your SG cluster is reliant on NFS ! great. And using the automounter is even more unreliable. I would investigate another way of achieving what you want without using nfs and automounter.

[Ralph Grothe]

Stefan,

I totally share your views regarding HA and NFS.
I also regard NFS mounts as too unreliable and HA defeating.
But, as with so many constraints imposed on us by our customers, I have absolutely no influence on the design and implementation of applications.
We sysadmins are only the idiots who are always blamed when things don't work for reasons we already knew beforehand.
This SAP stuff *must* have NFS exports and mounts.