>>> Lars Marowsky-Bree <[email protected]> schrieb am 23.05.2011 um 13:06 in >>> Nachricht <[email protected]>: > On 2011-05-20T08:16:23, Ulrich Windl <[email protected]> > wrote: > > > > Well, yes. I'm not quite sure why you'd want to use sfex though if you > > > have sbd fencing anyway. > > SBD is for node fencing only. If I need to ensure exclusive assignment of > shared storage resources (well you never know what the cluster stuff tries to > do) to avoid data corruption (e.g. through MD-RAID), I feel the need for > cluster-wise mutex-locks. > > This is not quite correct. > > SBD is fencing "only", but fencing already ensures that no node will > acquire resources unless possible competitors have been fenced. > > If you doubt that this - very fundamental and heavily tested - part of > the cluster logic doesn't work, it is much more reasonable to assume > that sfex generates a false positive or that a sfex dependency is > ignored. > > Neither will protect you against admins fumbling and accidentally > corrupting the data. > > Combining sfex and sbd does not create any real benefit, it just > complicates your configuration, making it more likely to fail.
Hi Lars, so you are saying that the CRM/LRM will never try to start a resource on more than one node concurrently, and they will never try to start a resource on a node of the cluster when it cannot be guaranteed that the resource is definitely down on every other cluster node (all if stonith/fencing works)? That would be good. Unfortunately the documentation on all of that is not very clear, and my belief is better to check twice rather than loose data... Regards, ulrich > > > Regards, > Lars _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
