[Linux-cluster] quorate error

[David Teigland]

On Wed, May 30, 2007 at 10:12:41AM -0400, Neil Watson wrote:
> On Wed, May 30, 2007 at 09:08:19AM -0500, David Teigland wrote:
> >>> During a disaster recovery test three nodes of a three node cluster
> >>> were caused to fail.  A single node was rebooted to attempt a
> >>> recovery.  The node failed to bring back the service and reports these
> >>> errors:
> >
> >Sorry, I failed to read that properly.  In a three node cluster you need
> >two nodes for quorum, so you need to bring two back before anything will
> >happen.
> 
> So this plan will fail :(.  We would need to have four nodes: two at
> the main data centre and two at the DR centre?  That is unexpected.
> There is no way around this?

Quorum is a voting scheme.  Each node has one vote by default and the
cluster needs a majority of all possible votes to do anything.  So, with a
four node cluster (each node with one vote) your cluster would require
three three of them for quorum (3 of 4 votes for a majority).

Two node clusters are a special case where we can override the normal
quorum rule and allow just one of the nodes to run on its own.

Qdisk (a quorum disk) is a system you can set up to assign votes to a
shared disk.  A cluster that can access the disk gets the votes from the
disk.  That's the standard way to work around quorum restrictions.

Dave