Joe Griffin wrote:
Steve,
I only know "how" to get things working. People like
Matt know "why". My belief is:
I was trying to give credit where credit is due.
I was trying to make a joke! :D
gmond communicates with everyone on the same port. and
gmetad gets data from specific nodes.
It also communicates with everyone on the same multicast IP. This is
basically just how multicast works, insofar as an application has to
send a packet to a specific address and has to listen on a specific
address in order to get a response.
Although I got my system working, and was able to help Yohann,
what I did not understand was that I have several clusters using
different multcast ports 8649, 8650, 8651, but the
/etc/gmetad.conf file doesn't use those numbers on
the data_source line:
# data_source "my box" localhost my.machine.edu:8655 1.2.3.5:8655
^^^^ ^^^^
I thought the data_source line would use the same port numbers, but alas,
8649 was needed. From my master node which communicates with the
head node from each cluster, I would get data from "telnet headnode
8649" not
"telnet headnode 8650"
There's the multicast port, and then there's the unicast port.
The monitoring core and gmetric ship out the XDR packets on the
multicast port (and, in the case of the monitoring core, listen on it as
well). The monitoring core listens for connections on the unicast port
and dumps the metric tree in XML format to anyone who connects and has
their IP allowed in gmond.conf.
I thought the port number would change, but
empirical evedence shows to use 8649.
It can change, if you like... :)
