Question about ganglia node name vs. PBS node name

[jsarlo@…]

We are using ROCKS 5.0 as our cluster framework. For ganglia, the node names are compute-X-Y.local, but for PBS, it uses compute-X-Y. Is there a way to have jobmonarch not think that these are two different nodes? Is there something simple I could add to the jobmond.py to just tack on a .local to the node names that the jobs are running on before it gets sent to gmetad?

Thanks. Jeff

[cwest@…]

Jeff,

While this may not be the best way to do it, here is a quick way to add the ".local" to the hostname. No promises this will work, and it requires that your nodes don't have any existing domain names when reported by the PBS system.

In jobmond.py locate "nodeslist.append ( host )" it is line 959 in the current trunk. This is where the hostname that gets sent to ganglia (by jobmond) is set. You are going to modify the "host" variable before it gets added to the nodeslist array.

Now locate "if not host in nodeslist:" (which should be the previous line with text - line 957). Add the follow line immediately before line 957, and at the same indent level as line 957.

host = host + ".local"

Craig.

[ramonb]

this problem seems to be reappearing in different places in different versions.

I will work on getting this fixed in the source for future release

[ramonb]

node hostname confusion is fixed in web 2.0 frontend version