[Pacemaker] 2-node clusters, who's the master now.

[David Coulson]

If you are running two nodes, you need to tell pacemaker you don't care 
if it can't get quorum, by only having 1 of 2 nodes available. Neither 
node which take over in this event know if there is split brain or not, 
so you will need to make sure you have sufficient infrastructure between 
the two to ensure you don't have a split brain due to a single issue 
(e.g. switch, cable, port).

The simplest solution is to add a third node to the cluster, so you 
hopefully end up with at least two nodes available during an issue to 
retain quorum. If your network dies, then you're going to still have 3 
nodes that won't run any resources, but at that point you have different 
problems.

I've many two node pacemaker clusters with no-quorum-policy set to 
ignore, and it works fine. We have multiple interfaces for the systems 
to communicate on, which talk via diverse switch environments. It is a 
known and accepted risk that if the entire network fails we will have 
split brain, but depending on your workload (e.g. haproxy in this case), 
split brain probably doesn't matter too much.



On 6/7/12 7:35 AM, Christian Parpart wrote:
> Hey all,
>
> regarding 2-node clusters for a single service (e.g. HAproxy, in my case),
> when now one node goes down, by server crash, or cable unplugged,
> how does the other still functioning node actually know, that this is
> not a split brain and can take over?
>
> I also wonder how keepalived tool is managing that, which is magically 
> working well so far,
> but I'd like to migrate to pacemaker/corosync for the new services.
>
> Many thanks,
> Christian.
>
>
> _______________________________________________
> 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://bugs.clusterlabs.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20120607/f5d6234e/attachment-0003.html>