On Wed, 2003-02-05 at 17:18, Rafael Nunes wrote: > > Hi... > > I've been searching about how the Heartbeat can assume that main node of > the cluster is dead when one of the Network Adapters has no link (i.e = > when I unplugged the network cable). > > I discovery some information about Marcelo Tossati's patch of the > Heartbeat, however, I read that the best way to reach this requirement > is using the MON, but I still haven't found the module that do this. > > Can anyone send me a link or some clue about this?
I'm not sure I understood... You have a HA cluster and want to know when a node is not dead, but has "only" been unplugged from the cluster network. Well if you use heartbeat, you should *always* have (at least) 2 media for the hb packets, e.g. ethernet and a serial cable (in fact this is my setup for a couple of heartbeat-driven clusters I manage). So when you unplug a nic, the node will *not* be marked "dead", since the serial heartbeat is still going on, and you will find sth in heartbeat logs about udp hb being lost. MON approach is quite different from the heartbeat one, since you have some host where mon runs (which may be outside the cluster -- and should, IMHO); from there you ping/fping/(anyone going to write arping.monitor?)/whatever cluster members and if they don't respond you mark them as dead. Please note that mon won't automagically trigger any sort of IP/resource/whatever takeover -- but you may write some custom alert. Timings are somewhat different too: I feel a reasonable interval for (f)ping.monitor has the order of magnitude of a minute, while a default heartbeat setup will schedule a hb packet every 2 seconds. Hope this helps... Piero _______________________________________________ mon mailing list [EMAIL PROTECTED] http://linux.kernel.org/mailman/listinfo/mon
