2.5.1 should support the concept of DMAX for individual metrics. I think
that extends to hosts, as well. Basically it's metric aging - if a metric
hasn't been transmitted for X seconds, take it off the list. It's designed
exactly for this type of thing - getting rid of hosts that have been dead a
long time. It doesn't delete RRDs, though (I think ... Fed can overrule me
on that one).
Funny thing, in my experience you only need to kill the monitoring cores
that your metadaemon polls.
Hope that helps...
Phil Radden wrote:
I wondered if anybody could help with an annoying situation I keep running
in to! Due to a couple of teething problems setting this all up, I keep
getting boxes coming up into the wrong multicast group. Unfortunately,
once all the other boxes in that cluster have noticed the imposter, it
seems remarkably difficult to persuade them all to forget about him again.
The only fix I've found to work is to stop _all_ the gmonds in the
affected cluster simultaneously, delete the appropriate rrds, and then
start all the gmonds up again. I'm really hoping there's a quicker way!
Thanks in advance
Phil
PS Many thanks to all of you who've sent comments on my disk-throughput
problems - storing the rrdbs on a loopback mount is appearing to do
the job for me nicely.
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Ganglia-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ganglia-general
