pechoi@syr.edu wrote:
> Hi,
> I have been trying setup NFS failover servers for read-only filesystem.
> I have been reading that NFS don't care about filehandle if it is
> exported as read-only. I exported read-only, mounted nfs read-only,
> mounted actual disks read-only, and I am still getting stale file
> handle error when I failover. I read article that two NFS servers got
> to see two partition. So is that has to be exactly like "/dev/sda1"?
> One of my node wasn't plugged in all the way now sees as /dev/sdk1. Is
> that matter?
>
> Thx

NFS clients definitely care about filehandles, readonly or not. It's
almost impossible to do transparent fail-over on a generic Unix/Linux
server. The inodes, generation numbers and fsid have to be the same on
the primary and the secondary.

You may be able to make it work by mirroring the hard drives, since
your filesystem is readonly. E.g. use `dd' to copy your disk, then
install that new disk in the backup server. Then your two servers
should be really identical (in terms of filesystem).

Cheers,
bc