NFS timeout

[Adam Noble]

Hi all,

We have a SAP environment consisting of 4 HP servers. We occassionally carry out re-boot and HA tests. One server presents the other 3 servers with NFS mounts which are mounted with autofs.

In the past we have had issues whereby we cannot get certain filesystems to mount. We then do a reboot and all is fine. Would it be fair to say that the best order to reboot would be the three client servers and then the NFS server. The only issue I have is my NFS server is a superdome which takes 30 mins to reboot. Do you think the client servers will cope with this amount of time being unable to contact the NFS server. The application will not be running at this time on the clients.

I realise we may be able to prevent the other servers from mounting on boot but wondered if neccesary.

Cheers

[John Payne_2 · Accepted solution]

Will they cope with the application being down that uses the disk? The very likely answer is yes. How well might be a different issue, I guess.

Are they hard mounts or soft mounts? Is the app that uses the FS really the only thing that depends on it?

A hard mounts makes the NFS client stay busy trying to reconnect when the server goes away. This could have an impact on the way the 3 clients appear to behave, bu they should reconnect just fine. A soft mount will cause an IO error. The point really is that if something is in the process of writing to the NFS volume in an outage, the hard mount will continue to try until successful, the transaction will be lost with a soft mount.

I guess what I'm trying to say is that either way around it, if your app is down, things should be fine.

Hope it helps
John

[TIMOTHY J. REGAN II]

What OS version are you running?

If you are running HP-UX 11.23 and above take a look at the man page "umount_nfs". From what I have experienced HP-UX 11.23 and above has provisioned a new version of the command "umount" that introduces a new switch "-f". This forces and I/O disconnect at the lowest level freeing up all NFS client connections.

Try making a universal script that gets called in your customer functions section that hammers all client NFS connections including circular mounts then recycle the server, client and core NFS daemons. Do the same for your DI's as well.

For all 11.00 - 11.11 systems I can tell you that I have had good mileage with the following commands in order:

umountall -F NFS -k
clear_locks `hostname`
clear_locks -s `hostname`


Then cycle the server, client and core NFS daemons.