[EMAIL PROTECTED] writes: > I'm not sure I agree with your conclusion. Why couldn't > the nagios plugin connect to a gmond on one of the nodes > in each cluster and parse the XML. It should also have > a timeout to go to another node, if the "primary" happens > to be down.
A Nagios plugin can only report the status for a single service on a single host. So, if you want to monitor five metrics on each node in a 200 node cluster, you would have to do the connect'n'parse thing 1000 times (each time throwing 99.5% of the data away) for each monitoring cycle, if you do normal, active, service checks. That's going to get expensive real fast. So, the only real option is to have a separate process that parses the XML, checks the metrics against your threshold values, and feeds the results to Nagios as passive service checks. But even if you do this, you can only do *service* checks. You can't tell Nagios that a node has gone down, since Nagios doesn't support passive host checks. (Unless thing have changed since I looked last). I guess you could combine these approaches and do the passive-checks-via-external-process for monitoring metrics, and let Nagios do active host checks through a plugin that connects to a gmond or gmetad and checks the REPORTED attribute for the host you want to check. I don't know how expensive that would turn out to be, but in my opinion it's getting complicated and ugly. -- Leif Nixon Systems expert ------------------------------------------------------------ National Supercomputer Centre Linkoping University ------------------------------------------------------------
