[Linux-cluster] Converting from dlm to lock_nolock?

[Ray Van Dolson]

On Thu, Feb 09, 2012 at 07:14:50AM -0800, Ray Van Dolson wrote:
> I am in the process of migrating a lot of data off of a 2-node GFS2
> cluster (RHEL5) via rsync.  There isn't a lot of data (only about
> 500GB), but there are a *lot* of files (about 5.4 million inodes).  As
> such, secondary rsync runs still take a lot of time walking the GFS2
> file system tree looking for differences.
> 
> I'm exploring some options for speeding that up -- the main one being
> dropping my cluster to only one node.  Is this doable for a file system
> that was greated with the dlm lock manager instead of lock_nolock?
> 
> Perhaps simply failing one node out of the cluster would be sufficient?
> I imagine some cluster.conf changes might be necessary?
> 
> I am less concerned about maintaining resiliency in the cluster than I
> am in reducing the downtime when I shut the current live server off, do
> a final rsync and bring the new service online.

I didn't do a thorough enough job Googling.  Sounds like as long as I
bring down the other node (or at least stop gfs2 services), I can then
tell the primary node to remount the gfs2 file system with lockproto
set to lock_nolock.

May give this a shot.

Ray