I’ve done some testing of the Ganglia monitoring tool on three small Banana Pi clusters.
I used a cluster of four database servers, another cluster of four nodes running GlusterFS, and a cluster of two master nodes.
The Ganglia meta demon has to be installed on the master node (the node where Ganglia stores data and the web front-end). Some configuration changes need to be made in /etc/ganglia/gmetad.conf:
- set the ‘gridname’ directive
- add data_source lines for each cluster
- add a list of IP addresses to the trustedhosts directive
The data_source directive specifies the name and host/port details of a cluster. There is one data_source line for each cluster.
The Ganglia monitor needs to be installed on each node. It’s configuration file (/etc/ganglia/gmond.conf) also needs to be edited:
- edit the cluster name
- change the port number that the monitor uses to communicate with the master node.
Ganglia uses different ports to distinguish between different clusters, so the ports in gmond.conf should be the same for every node in a cluster, and each cluster needs to use a different port.
You can read about the details here: http://banoffeepiserver.com/server-monitoring/ganglia/set-up-ganglia-on-multiple-clusters.html.