> This locking is much more elegant in NFSv4. So in clustered NFS Server you > should use v4. See the resource agent description for locking timeouts.
OK, I follow your advice. > Obviously you'll need to tune this for your use case. > > So a floating service IP address. Client mount options are just > "rw,hard,intr", with a DNS name associated to this service IP address. Thank you Colin for your example. A comment about NFSv4, with RHCS on CentOS 6 I have error when I use ressource "nfsexport". It seems that "nfsserver" is better in that case. > How are you managing the failover? If you are doing failover of the > file system (e.g., via SAN), then I'd expect the HA service manager > (rgmanager, pacemaker, etc.) to handle the export, since you don't > want to do the export until the file system is mounted. If you are > exporting a replica file system, then I believe it has to be a block > level replica (e.g., DRDB)--a file system replica via rsync or such > will have different inode numbers, and AIR the inode number shows up > in the NFS file handle. I was on the second case (file system replica) so that's why I have the error about the ID. Thank you very much for your light ! I use a centralized file system to resolve this. I think it's the only way to permit transparent NFS failover ?
-- Linux-cluster mailing list Linux-cluster@redhat.com https://www.redhat.com/mailman/listinfo/linux-cluster
