Serviceguard NFS Toolkit for Linux Version A.02.00 Release Notes, July 2007

Serviceguard NFS for Linux Version A.02.00 Release Notes
Known Problems and Workarounds
Chapter 1 13
Known Problems and Workarounds
The following describes known problems with the NFS Toolkit and
workarounds for them. However, this is subject to change without notice.
For the most current information contact your HP support
representative.
More recent information on known problems and workarounds may be
available on the Hewlett Packard IT Resource Center:
http://itrc.hp.com (Americas and Asia Pacific)
http://europe.itrc.hp.com (Europe)
JAGaf57739: HA NFS and ‘Stale NFS Handle’
What is the problem?
Serviceguard relies on LVM (Logical Volume Manager) to manage its
shared storage, which contain data and files of applications managed by
Serviceguard. Just as with other applications, an NFS instance managed
by Serviceguard means that the specific instance is “active” on one node
at a time, with all resources available to that node only (including
volume groups configured as a resource for the NFS package).
If the package goes down on node 1, the resources are released so that
the second node can “claim” the resources. The package is brought up,
and the instance is now “active” on node 2. NFS clients continue to
connect to the server, unaware that the server has migrated from one
node to another.
The behavior has changed in Linux kernel 2.6 such that the kernel uses
lvm2 with the device mapper to virtualize devices instead of using the
actual physical names.
In this implementation, the actual device node for a logical volume is
dynamically created upon volume group activation, meaning a volume
group that starts out on node 1 but is failed over to node 2 can very easily
end up with a different minor number after fail over. This will result in
clients who connected before the failover getting a “stale NFS handle”.
The volume groups are created with dynamic minor numbers. This
causes NFS problems (‘stale nfs handle’) on the client when the NFS
package is migrated to another node.