How to manage special 2 node configuration
haberland at altus.de
Thu Aug 9 08:49:23 MDT 2001
Juri Haberland wrote:
> "Poppi, Sandro" wrote:
> > Therefor I configured the cluster in that way that the heartbeat is only
> > sent through eth0, no serial connection since when doing heartbeating on
> > both the serial line and eth0 if eth0 goes down on nserver1 nserver2 does
> > NOT take over until the machine is completely down.
> > The takeover of nserver2 works as expected if eth0 is down for any reason,
> > but if e.g. the problem is a broken cable heartbeat still runs on nserver1
> > and if the cable is replaced nserver1 is "heartbeating" again but no
> > re-takeover occurs so on both servers eth0:0 shows up with the virtual ip.
> Yeah, this is expected. That's why it is a bad idea to use just one
> heartbeat medium. I would recommend to use at least two heartbeat media
> (e.g. serial) and to monitor your network connectivity with something
> like mon. If mon detects a network failure it stops Heartbeat so that
> the other machine takes over. If the network is up again mon can restart
> Heartbeat. It is a little complicate to implement but worked for me very
Forgot one thing:
There is another advantage in using mon:
You can also monitor your application(s) and try to restart them if they
fail. Furthermore, if they can't be restarted you can shutdown heartbeat
so that the other node takes over and/or reboot the node with the failed
application in the hope that a restart cures it (Windoze approach ;-).
If each of us have one object, and we exchange them,
then each of us still has one object.
If each of us have one idea, and we exchange them,
then each of us now has two ideas.
More information about the Linux-HA