[Linux-HA] Heartbeat, GFS, Samba, and NFS

[Brandon Allhands]

Nick Couchman wrote:
> Hey, Everyone,
> I'm running SuSE Linux Enterprise Server 10 and trying to get a sort of clustered file server setup going.  SLES10 uses heartbeat for the clustering/HA setup, and I've chosen GFS for the filesystem to back the system.  I'd like to use GFS because I want both file servers (and our backup server) to have access to the file system simultaneously (the backup system can access it read-only).  If one of the servers fails I'd like to avoid the issue of having the backup server remount it.
>  
> Anyway, I'd like to set up heartbeat with NFS and Samba shares for the directories on the file system.  I'm trying to use hb_gui to set this up, and I have a few questions.
> 1) First, can I have both servers share out the same NFS volumes simultaneously?  I know Samba doesn't work well (at all) in this scenario because Samba doesn't (currently) have a way of sharing file locking information between multiple instances of Samba.  I'm wondering if that's also a limitation of NFS or if NFS can handle the file locking?
> 2) I need to know how to set up the resources, exports files, and Samba configurations.  Most of the configurations seem to point to having "normal" filesystems (ext3, etc.), that are mounted when the heartbeat configuration starts and move between systems as necessary.  A lot of them also seem to have the NFS service running on one system or another and not NFS servers on both systems, which is what I'm going for.  
>  
> Thanks,
> Nick
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>   
Hello Nick,

Are you referring to the Global File System that is attached to the 
Redhat Cluster Suite? If so, GFS relies on cman and the redhat 
distributed locking manager as well as the clustered volume manager, all 
of which are part of their clustering implementation.

As for your setup, have you looked at OCFS2 (Oracle Clustered 
Filesystem)? We have recently migrated a few arrays away from GFS to 
OCFS2, and OCFS2 seems much faster. OCFS2 would allow you to mount your 
filesystem on both of your nodes as RW, which would assist in a failover 
scenario, as well as OCFS2 integrates nicely with heartbeat.

Also, as long as you have a cluster aware filesystem, NFS can be used to 
serve the same files from multiple machines. This is the way we have it 
set up, and we have not experienced any locking problems. Each nfs 
server handles it's own locking, as the underlying filesystem handles 
locking for the nfs servers.

Brandon