[Linux-cluster] Configuring NFS on Red Hat Enterprise Linux 4AS / Cluster Suite / GFS 6.1
Robert S Peterson
rpeterso at redhat.com
Wed May 24 16:43:35 UTC 2006
On Wed, 2006-05-24 at 10:40 -0400, Kovacs, Corey J. wrote:
> Bob, in looking through your document, first I'd like to say "Thanks". Job
> well done.
> Second, I have a question about how RHCS (either version) handles the rmtab.
> At a glance it appears that the cluster simply keeps a copy of the rmtab for
> a given NFS instance sync'd wioth a copy on the exported filesystem itself.
> For a single point of entry for NFS failover, it's a no-brainer and simply
> However, for a multi server config like you describe where all servers have a
> piece of the pie so to speak, how do does the cluster keep from killing the
> rmtab file on the failover target.
> That is if I have serverA and serverB, both running clusterred NFS services,
> and serverA
> buys the farm, it seems that when the rmtab from serverA gets sync'd over to
> serverB, then
> serverB's rmtab gets squashed.
> That's a very simplistic view of things but I've not taken the time to go
> code diving
> to ge the answer of "How, at a low level, is the rmtab handled"...
> Thanks again for a good read.
Your question hinges on whether your NFS servers are set up with
virtual IPs, or real "NFS services" managed by rgmanager.
In the case of virtual IPs, as far as I know, nothing is managed except
the virtual IP itself, and therefore all the management of fstab and
other nfs-related files like rmtab are all on the user's shoulders.
With rgmanager managing an nfs service, however, things are better.
There the rmtab stuff is all handled by an rgmanager daemon known as
clurmtabd whose job is to keep the rmtab information in sync
between nodes configured for the nfs service. It tries to figure out
clustered resources and keep them separate from non-clustered
resources. BTW, the daemon is in user-space, so the kernel knows
nothing about all this.
I'm afraid I don't know much more about the clurmtabd daemon than that
because I haven't studied that particular area of code much. I'm more
of a gfs guy than an rgmanager guy. But hopefully this helps.
Red Hat Cluster Suite
More information about the Linux-cluster