NFS loopback mount hangs for upto 20 minutes when HP ServiceGuard NFS package is failed-over

[Andrew Prolisko]

Hardware: HP DL385 and EVA8000
OS: SLES9.3 2.6.5-7.283-smp
Cluster SW: serviceguard-A.11.16.07-0, nfs-toolkit-A.01.04-0

1) Cluster package A, running on NODE-A provides NFS file system X to NODE-A.
2) Cluster package A is failed to NODE-B.
3)All access to NFS file system X on NODE-A hangs for upto twenty minutes. (Other NFS client access to NFS file system X is only disrupted for the time it takes to fail-over the NFS package from NODE-A to NODE-B.)

Why does it take so long? Is there a way to reduce this time?

[Steven E. Protter]

Shalom,

Check syslog /var/log/messages (may be different in your distribution) for messages.

Note that if you are using names instead of IP addresses for your NFS mounts you could be having trouble with slow name resolution. Consider faster /etc/hosts entries for quicker response.

NFS open connections need to be reset when the package fails over and when there is a lot of data going back and forth this can take time.

SEP

[Asha Yarangatta]

This is a known behaviour and the reason is unknown. Thats why loopback mount is not supported for NFS toolkit on Linux. Please see the NFS toolkit's user guide for supported configurations. This type of configuration is not mentioned in it.

I suggest you not to use NFS loopback mounts unless you are fine with 20 minutes delay.

[Martin Sedlacek_4]

Hi,
I think that reason is that the client is using loopback device for accessing NFS share once the NFS pkg is running on the same node. I do not know how to bypass loopback on Linux (on HPUX it's possible)

Martin