[Linux-cluster] any timeline for supporting software RAID?

[Jeff Wasilko]

On Wed, Jan 23, 2008 at 02:20:44PM -0500, Lon Hohberger wrote:
> On Tue, 2008-01-22 at 16:18 -0500, Jeff Wasilko wrote:
> > On Tue, Jan 22, 2008 at 03:59:47PM -0500, Christopher Hawkins wrote:
> > > If you don't need GFS... Then drbd + heartbeat is a good way to go if you
> > > want to mirror the local storage on each machine.
> > 
> > True, except Zimbra has explicit support for RHCS, and we'd be off
> > rolling our own solution with drbd + heartbeat....
> 
> As I recall, you can use LVM mirroring.  In recent versions of RHCS (4.6
> maybe?), there's an agent to handle assembly/disassembly of LVM
> (non-clustered) volume groups before/after failover.

I found the 4.6 docs, and it does look like all of the previous
warnings about using mirrors under cluster are gone.

Plus, I saw this note in the 4.6 release notes:

#
A known device mapper issue causes unexpected cmirror deadlocking. This issue prevents reliable use of cluster mirrors configured with three or more legs. As such, it is advised that you avoid configuring cluster mirrors with three legs or more until this issue is resolved.
#

Is cmirror the layer below lvm?

We generally don't do 3-way mirrors, other than when we're transitioning
between arrays on the SAN.

-j