Hi, On Wed, Aug 25, 2010 at 05:03:46PM +0200, Andre Timmermann wrote: > Hello List, > > I am maintaining a cluster where both nodes are in different > datacenters. Therefore, I don't have a serial connection for heartbeat > and must rely solely on the network connection (GB-fiber-ring) > > A few days ago, we had a problem with one of the fibers so that the > spanning-tree was rebuilding. This caused an interrupt of ca 30 seconds > so that we ended up with a split-brain :(.
I guess that there's no fencing configured, right? > Can someone point me to an approach to avoid this? Is it a good idea to > play with ha.cf to have higher timeouts? > > This is, what we have now: > # Thresholds (in seconds) > keepalive 1 > deadtime 3 > warntime 2 > initdead 30 The warntime/deadtime are a bit too tight. Depending on the failover timing requirements, I'd rather go with something like 5/10 or 10/20. > Do you recommend other technical solutions? AFAIK, none right now. The safest way would be to deploy the meatware stonith plugin and let the operator decide which side to let run the resources. Thanks, Dejan > Best regards, > Andre > > _______________________________________________ > 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
