Hi Everyone,

I am (trying) to use a combination of drbd, lvm, unionfs, and nfs
to provide a HA cluster of diskless nodes. With the myriad of
technologies in use, this may not be a unionfs issue, so apologies
for the noise if not.
 
I have it all setup, failover works correctly in that the drbd
primary gets switched to the failover host, vgs get exported,
and reimported, then mounted correctly, and the unions get
created/mounted just fine. The NFS daemons get stopped and started
correctly as well.

So from the failover pair of hosts, everything is good. However
on my diskless nfs root client I get a load of NFS "error: fileid
changed" messages post failover.

Even failing back to the *original* NFS host doesnt fix it - I
just get a different "expected <such>, got <some>" where <such>
remains the same. File/Directory accesses on the diskless client
give "Stale File handle" errors.

I've checked that I have a fsid in /etc/exports (that is identical
on both NFS hosts). Also major/minor numbers for the device mapper
seem identical pre and post failover, although I am running udev.

Also checked that I have the -n flag on statd, yep, that points
at my virtual IP, and during failover I get (the expected) NFS
server not responding.

mountd is configured with --no-tcp -N 2 to prevent NFSv4 and
NFSv2 being offered, and the client has v3 in the nfsroot string
(Full string is nfsroot=192.168.39.39:/mnt/union/%s,v3). Ive
tried using both soft and hard mounts, same difference.

kernel is 2.6.16 from debian sid on all hosts.
unionfs is 20060530-0122 on the two NFS hosts (from debian sid)
drbd is Version: 0.7.19 (api:78/proto:74) (from debian sid)

Anyone know what I might have missed, or if I need a later kernel
or unionfs, or a patch, or does this just not work ? I'd been led
to believe that it should work...

I can post configuration files if folks want, just didnt want to
clutter up the list even more, if its something as simple as "Doesnt
Work", or "Needs <blah> patch, or later kernel...

All the Best

Iain
_______________________________________________
unionfs mailing list
[email protected]
http://www.fsl.cs.sunysb.edu/mailman/listinfo/unionfs

Reply via email to