Wednesday, December 17, 2008

Ganglia tricks

Ganglia is a powerful monitoring tool for clusters. Some tricks:

1) remove dead node
try to restart gmond and gmetad.
in /etc/gmond.cong, try to set host_dmax=3600 (sec) in globals.

2) two clusters in one subnetwork
for gmetad to collect data from corresponding cluster (not the other), the trick is to set the TCP/IP port for these two clusters to be different. For instance, the default port is 8649. You can set the second one as 8648. Then, in the server with web frontend, edit gmetad.conf, and put something
data_source "Cluster1" c1_headnode:8649
data_source "Cluster2" c2_headnode:8648
to show all info in one web interface.

No comments: