On Tue, 22 Jan 2013, Alex Sudakar wrote: > Hi Dima, > >>> With only two nodes comprising the cluster I believe a quorum is >>> impossible, so I've set no-quorum-policy to 'ignore'. However I was >>> wondering if there is a possibility of using one or more 'tie breaker' >>> devices/resources to determine a proper quorum? I _think_ I've seen >>> mention of such a thing in passing in this list; I'm not sure. >> >> I simply grep for 'link detected' in the output of `ethtool eth0` -- if >> it's not there, the node is off-line and can shoot itself. (Although I >> don't do that: it's off the net, it's not bothering anyone, and I have >> nagios elsewhere monitoring the hosts. If the link comes back on its >> own, that could cause a problem, but that hasn't happened to me yet.) > > I imagine other things could block network connectivity for one of the > nodes, outside of the machine itself - a switch or hub failure? Which > wouldn't be detected by running ethtool locally. > > But it's a good check to run regardless, thanks!
I've also had network connectivity restored (switch got rebooted, someone noticed a loose cable and plugged it in, etc) What I would do would be to look into defining a ping node that could be used as the tie-breaker (but that ping node needs to be HA as well) when you get a more complex arrangement (two nodes plugged into two different switches that are joined together, multiple interfaces per node, etc) this gets 'interesting' The real question that you need to answer before going down this road is how much damage you suffer in a split-brain situation. If you have shared nothing HA nodes, the odds are that you really aren't damaging much, the worst case tends to be the dup IP issue, and if one is off the network, that really doesn't matter. David Lang _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
