[Pacemaker] Quorum: a couple of ideas
Andrew Beekhof
andrew at beekhof.net
Mon Feb 21 14:01:30 UTC 2011
On Mon, Feb 21, 2011 at 2:05 PM, Carlos G Mendioroz <tron at huapi.ba.ar> wrote:
> Hi,
> As I understand it, pacemaker decides on having quorum
Let me stop you there... pacemaker doesn't normally decide this at all.
It doesn't for heartbeat based clusters and once the corosync quorum
plugin is functional it wont there either.
Quorum is something that should normally come from the underlying
cluster messaging layer.
> when
> the number of nodes it can talk to is strictly greater than
> half of the existing nodes, or 2 * N > T .
>
> This has the unfortunate consequence of turning it useless for
> N = 2, and less than optimal for any even T.
>
> I would like you to consider having one node a double vote.
Short answer: No.
Medium answer: Thats what no quorum policy is for.
Long answer: See your local encyclopedia for the definition of
quorum, we ain't perverting it here :-)
Historical answer: Vote fudging became common practice because cluster
managers had a meltdown when quorum was lost.
Pacemaker does not suffer this problem and
thus does not wish to encourage the practice :-)
Summary: No :-)
> (And that would also jump T to T+1) The properties of quorum
> would stay (not going schizophrenic, so to say) but if you
> happen to partition a 2 node set, the chosen one would stay up.
>
> An more flexible approach could be to base quorum on the number of
> resources that can stay up in this node, a number that can be
> statically configured per node. As long as the number of connected
> resources is above half of the defined, quorum is mantained.
>
> Forgive me is this has already been discussed and discarded by some
> reason.
>
> Regards,
> --
> Carlos G Mendioroz <tron at huapi.ba.ar> LW7 EQI Argentina
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs:
> http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>
More information about the Pacemaker
mailing list