The Ganglia Interface


Monitoring data collected with the Ganglia Monitoring Tool can be published into the MonALISA Monitoring Service. MonALISA provides two separate ways of collecting Ganglia info: One simply connects to port 8649 of a host that runs the gmond daemon and parses the output XML file (monIGangliaTCP). The other way is to configure the host on which the MonALISA service runs to be in the Multicast range of the cluster that is monitored with Ganglia (monMcastGanglia). The MonALISA developers provide Monitoring Modules for both of the above scenarios.

Using the monIGangliaTCP monitoring module

In Monitoring the STAR Linux Clusters at BNL I use the first way described above. As of version v1.0.1 the module monIGangliaTCP that connects to the Ganglia Monitoring Daemon (gmond) and pulls the XML file containing the monitored data has been included in the monitoring service distribution jar files and thus there is no need to be configured as a custom module. To use this module enter the following in your <farm_name>.conf config file:
               *PN{monIGangliaTCP, host_name, port}
where host_name is the host where Ganglia monitoring daemon (gmond) is running and port is the port that is listening to (typically 8649). PN (Processing Nodes) above is the Cluster Unit name. If a Cluster Unit name starts with PN then the farm filters are used to report global views of the farm.
You have to make sure that the host where the MonALISA Service runs is a trusted_host for Ganglia. To test this, try from your MonALISA host to telnet to port 8649 of the gmond host. You should get the expected XML file.

Using the IGangliamod monitoring module

In prevous versions of MonALISA I used the GangliaMod Module which does the same thing that monIGangliaTCP does. Only this time, since this module is not included in the distribution jar files, you have to use it as a Custom Monitoring Module. The lia.Monitor.CLASSURLs property (set in the file should be set to the directory where the Module is located:
If you use other customs module, the directories should be separated by comma.
In your <farm_name>.conf file add the following:
               *PN_BGC {IGanglia,, 8649}%40
where IGanglia is the class name of the Module.

Current Issues with the Ganglia Interface

Using the above Ganglia Interface Modules I got a few TimeOut and OutOfMemory errors. However the errors were not reproducable and thus difficult to debug. The source of these errors may not even be the MonALISA Service.

Stratos Efstathiadis - page was last modified