Great idea, I missed to create an issue earlier since the older hbase EC2 scripts also enabled Ganglia, but I wasn't sure where to put this. Makes totally sense!

The Python scripts have some Ganglia code you might be able to resurrect.

Basic ganglia support

Commit notes for the patch:

• Only a 1-deep hierarchy is supported (n monitors all report to one meta)
• The metad is also part of the monitored cluster (i.e. reports to itself)
• The webfrontend is installed on the metad node.
• There can be only one instance with the ganglia-metad node. Things will probably (certainly) break if there are more than 1 metas.
• No real service tests yet (should do HTTP get on the webfrontend?)

Nice work Karel!

> No real service tests yet (should do HTTP get on the webfrontend?)

Sounds like a good idea to make a simple request just to check the webfrontend is running. For checking Ganglia you could do some simple asserts on the XML output generated by gmetad & gmond (e.g. http://code.google.com/p/embeddedgmetric/wiki/ParsingGmond )

Thanks. I wasn't going to check the XML because the service is bound to localhost (or was it the internal ip), but it's easy enough to do it over ssh.

Updated patch - includes testcase for ganglia webfrontend and ganglia XML
I didn't know how to SSH to the hosts for getting the ganglia XML, so I just opened up the port in via the firewall rules. Perhaps we could expose the 'getCompute' in the TestClusterController class?

Another patch. I'm now testing all gmond services - so both ganglia-metad and ganglia-monitors service are tested.

This patch adds ganglia-metad and ganglia-monitor roles.
You can have only 1 ganglia-metad instance.

+1

Looks like there are some spurious changes in recipes/nodes-byon.yaml which should probably be removed.

Just to be sure: Should I clean up the patch or will you drop the changes to recipes/nodes-byon.yaml

Updated patch just for the record.

I've just committed this. Thanks Karel! Integration tests pass on ec2 & cloudservers-us.