Alain.Moulle wrote: > Thanks Andrew, > > 1. So my understanding is that in a "more than 2 nodes cluster" , if > two nodes are failed, the have_quorum is set to 0 by the cluster soft > and the behavior is choosen by the administrator with the no-quorum-policy > parameter. So the question is now : what is the best choice for > no-quorum-policy > value ? My feeling is that "ignore" would be the best choice if all services > can run without problems on the remaining healthy nodes.
That's not the only case this can happen. If you run into split-brain, each node may be healthy but the network connections may be broken. With "ignore", you will end up with resources running multiple times. That's a problem sometimes ;) Don't use ignore in >2 node clusters. > "suicide" or "stop" : my understanding is that it will kill the > remaining healthy nodes or > stop the services running on them, so it does not sound good for me ... > "freeze" : don't see the difference between "freeze" and "ignore" ... ? > > Am I right ? > > 2. and what about the quorum policy in a two-nodes cluster ? You need working stonith and policy=ignore, as no node can have >50% on its own. When the connection is lost, one node will shoot the other. The cluster software should not be started at boot time, otherwise you will end up in a stonith death match. There was quite a nice explanation on the pacemaker list some time ago. Look for STONITH Deathmatch Explained in the archives. Regards Dominik _______________________________________________ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems