On Sun, Aug 24, 2008 at 20:37, Ronny Becker <[EMAIL PROTECTED]> wrote: > 2008/8/23 Lars Marowsky-Bree <[EMAIL PROTECTED]> > >> On 2008-08-22T09:20:02, Ronny Becker <[EMAIL PROTECTED]> wrote: >> >> > Hello, >> > I am trying to configure a fencing after failover and cannot find a way >> to >> > do this. >> > I am using heartbeat 2.1.4. >> > >> > This is what I have configured: >> > - resource_stickiness is about 99 >> > - rosource_failure_stickiness is -50 >> > - Operations "on_fail" is restart >> >> on_fail="restart" for which operation? >> >> > So, if the resource fails twice on a node, the resource will migrate to >> > another node. That`s ok. But I need to fence, too. How do I configure >> "try a >> > restart twice and if that fails, migrate and fence the node" ??? >> >> You can set on_fail="fence". But there's no need to fence a node after a >> successful stop, so that is simply not supported. >> >> Instead of describing how you want to achieve something, maybe you could >> try to explain why? >> >> >> > The reason behind is for example a mounted partition. If the cluster > migrates to another node, maybe the mount isn`t really "closed". If there is > SAN used in the background (sure connected to all nodes) that could lead to > data corruption because two server have access a the same time to the > filesystem - that`s no cluster fs. > > But if I think about it. This is not really logical - you are right. If > stonith is enabled and every resouce has configured a "on_fail=restart" > monitoring operation. What would happen, if the restart doesn`t work ? Would > heartbeat shoot the other node ? Or if heartbeat want`s to stop a resource > or group and this action fails - would this node get fenced ??
yes every action can have an on_fail preference. the default, when stonith is enabled, is * for monitors: restart (stop and start the resource) * for stops: stonith the host node * for starts: restart (stop and start the resource elsewhere) > If this is true, then you are really right that on a successfull stop or > restart there is no need to fence a node. > > > regards > Ronny > _______________________________________________ > Linux-HA mailing list > [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems > _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
