OCFS2 Looks Interesting

There was a discussion on the CentOS mailing list the other day about Oracle Cluster File System 2. I went and checked it out. Well, I read some docs about it anyway.

It's released under the GPL and it's not really just for Oracle. It looks to be a share nothing type file system layer. Oracle uses it for RAC, I believe. Of course RedHat doesn't ship it -- it would eat into their GFS sales. Oracle does distribute RPM's for RedHat/CentOS though.

I've played around with DRDB for replication/fail over before. It works pretty well, but you definitely feel blind on your secondary machine since you don't have the shared partitions mounted. With MySQL starting to push DRDB for MySQL failover, I wonder if this would work better. You'd still want to use heartbeat to control where MySQL is running, but at least you'd be able to see everything on each node.

It also may be a decent alternative to replication if you're just using MyISAM. This may be good for scaling out data warehouses -- having only one master updating files and many slaves reading those files.

Another interesting use would be for apache document roots. It gets around having to sync each web server node's document root (if you take a share nothing approach), or using an NFS server (if you use NFS).

I haven't really looked into how well OCFS2 handles node additions (I'm guess you have to restart the cluster). Nor what happens when a node gets out of sync.

Sooner or later I'll install and test it out.