On Tuesday 12 August 2008 00:53:21 Todd, Conor wrote: > No, it's not a clone resource: the configuration is different for each > node's riloe configuration, so there's one stonith resource for each node. > Do I need to set up stonith resources as cloned for them to work?
Makes sense for a five node cluster as a stonith resource to stonith a node should run at least on the other four nodes which - in case - should stonith the dead node e.g. : the stonith resource to stonith node1 should run on node2/3/4/5 > > > -----Original Message----- > > From: [EMAIL PROTECTED] > > [mailto:[EMAIL PROTECTED] On Behalf Of Andreas Kurz > > Sent: Monday, August 11, 2008 5:24 PM > > To: General Linux-HA mailing list > > Subject: Re: [Linux-HA] STONITH, default fencing time, forced-fencing > > > > > Hi! > > > > > > I've set up a five-node HA cluster running a bunch of services, and > > > all > > > > seems to be going well, although one node insists on telling > > everyone else that its running any new resource which is > > set-up but not activated. I use crm_resource -C to deal with > > this situation, so it's not so bad. > > > > > Anyway, these nodes are HP DL380-G5s, and so I'm using the riloe > > > STONITH > > > > script. I tested it by hand, and it works like a charm. > > I've set up the STONITH resources so that they never run on > > the same machine as the one they control. The other day, I > > artificially caused a situation in which one of the nodes > > should have been fenced. The cluster realized this and > > "scheduled" it for fencing, but the fence never happened. > > I'm wondering what this "scheduling" is, and what parameters > > are available to control it? > > > > You mean: you configured one clone resource for every rilo > > device and there runs a clone on every node except the one it > > controlls? ... then it should have worked ... if not: log > > files and your cib would be handy. > > > > I think there was even a post of Dejan to the list ... some > > times ago.. > > that said it is also save to run a stonith script on the node > > it controlls as a node does not stonith itself. > > Right, but if a node is running its own stonith resource and then the node > disappears, it will take even longer to fence it because the cluster is > going to have to figure out who's going to run the node's stonith process > before it can actually get around to doing the stoning -- er, stonith'ing. Is that so? I mean ... it always has to be one of the "surviving" nodes that shoot the other == dead node in the head ... Regards, Andreas > > - Conor > _______________________________________________ > Linux-HA mailing list > [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems -- : Andreas Kurz Tel +43-1-8178292-64 : : LINBIT Information Technologies GmbH Fax +43-1-8178292-82 : : Vivenotgasse 48, A-1120 Vienna/Europe http://www.linbit.com : _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
