[Linux-cluster] Basic question on CLVMD / GFS

[Jonathan Brassow]

On Thu, 2005-12-22 at 15:02 +0100, Michael Weitzel wrote:
> Jonathan Brassow wrote:
> > On Wed, 2005-12-21 at 16:01 +0000, Patrick Caulfield wrote:
> >>Michael Weitzel wrote:
> >>>I am confused about the usage of clvmd. On a blank 80 GB disk I created
> >>>pvcreate /dev/hdb
> 
> >>hdb doesn't sound like a shared disk to me.
> >>clvmd doesn't magically make IDE drives into shared storage I'm afraid.
> 
> I see. I suspected this :-/
> 
> > ... but gnbd would work.  You would have a single point of failure in
> > that case, but you would be able to access a volume from two machines.
> 
> Is there a limit on the number of machines a blockdevice can be exported
> to?

Maybe, but it would be extremely high.

>  I read somewhere on this list that it is possible to use LVM to
> build volume groups of imported gnbd block devices (probably with
> mirroring). 

That would be to get around the single-point-of-failure problem.
However, cluster mirroring is not yet available.

> Can I safely "clone" such a LVM configuration to use it on
> the different nodes of a cluster? ... I have a bunch of nodes with spare
> IDE disks ...

Not sure what you mean by clone...  You would export local drives via
gnbd, so they are visible to the rest of the nodes.  Then, you can put
LVM on one or more of the exported disks.  Without mirroring though,
each exported disk you use in your storage pool becomes a potential
point of failure.  Should be fun to play around with, but until the
stars line up, don't use something like that in production.

 brassow