[Linux-cluster] Preventing clvmd timeouts
linux at alteeve.com
Thu Jan 26 14:41:03 UTC 2012
On 01/26/2012 09:02 AM, Corey Kovacs wrote:
> My first thought is that since a two node cluster is at best a misnomer,
> then clvmd won't work as expected since it might require the operational
> semantics of a proper quorum based cluster (three nodes/votes or
> better). That's how I understand it anyway so if anyone has info the
> contrary, please chime in and correct me. If it work as I think it does,
> then perhaps the devs could insert a check to ensure all pre reqs are
> met before even attempting this and spitting out the appropriate message?
> One thing you could take advantage of is qdiskd in order to attain a
> quorum based cluster with two nodes. It's not an ideal solution but it
> can work quite nicely if you are limited to two nodes.
> If it will in fact work properly with a two node "cluster" then it could
> be the result of a node being fenced and not acknowledged properly?
> Not much help but some things to think about.
DLM works fine on 2-node as quorum is allowed with just one node, thanks
to the 'expected_votes="1" two_node="1"' attributes.
As for qdisk, you can't use it on DRBD, only on a SAN (as it is possible
to have a split-brain condition where both nodes go StandAlone and
Primary, this allowing both nodes to think they have the qdisk vote).
E-Mail: digimer at alteeve.com
Papers and Projects: https://alteeve.com
More information about the Linux-cluster