Prashant Bhamidipati wrote:
b) I did think of your second suggestion. but for that I would have had
to make changes on the gmond.conf of every single machine in the cluster
that I wanted to monitor ( right .. ?? )
Not all of them, just the ones that you want to query.
Picking one or two from each rack / switch should be enough to satisfy
the department of redundancy department... IIRC, they are tried in the
order they appear in the config file, every time. In other words,
gmetad doesn't "remember" which one it last polled successfully - it
goes through the list every time.
Depending on the amount of XML data and the frequency of polling, not to
mention the size and nature of your cluster iron, you may see a CPU hit
on the gmond that gets polled.
But we are probably talking many hundreds of nodes polled every ten or
fifteen seconds...
that is why I wanted to check if t here was an easier way to do this ..
( laziness on a rainy day )
See, Matt? We are spoiling cluster admins with our easy-to-use
software. In this spirit, I suggest we code-name the first g3 release
"Deep Thought."
Or perhaps we should save that for the gexec/gmetad/gmond coalesced
cluster scheduler/monitor version...
