Joe Kaiser wrote:
Hi,
I work for a High Energy Physics lab, and we are evaluating ganglia for
some of our monitoring needs. I have cluster on the same subnet that I
want to separate into serveral logical clusters. I have been able to do
so thus far by putting different logical clusters on a different
multicast route. I have divided them into a Main Cluster and an RD
Cluster.
On the web interface, I see the 8 machines in the RD Cluster under the
RD cluster. However, the Main Cluster shows all of the rest of the
machines in the subnet plus the 8 in the RD Cluster. Those 8 are grayed
out on the Main Cluster web view.
Additionally, the Metacluster page (the startup page) shows headings for
Metacluster, RD Cluster, Main Cluster, and the graphic of the machine
shows an additional 8 machines. So that the RD Cluster seems to be
counted twice for the Metacluster.
I have done this by having two gmetad collectors. One for each
sub-cluster. There is a one-way trust relation to the machine that
servers the webpages from the second gmetad collector machine. The Main
cluster listens on the default multicast route for ganglia and the RD
Cluster listens on a different multicast route.
Has anyone seen this before? Is there a simple way to designate
subclusters on the same subnet such that each subcluster is not counted
in any other cluster?
We are using the most current versions of ganglia-core, gmetad, and the
webfrontend.
Hi Joe,
That all sounds a little wacky to me. I currently have two test clusters
up, and here's the setup:
** front-end unit running apache+php, gmetad and front-end.
*--> gmetad is configured to poll the "fs" cluster and the "ds" cluster.
* "ds" cluster nodes are multicasting on 239.11.2.79, all_trusted is on.
* "fs" cluster nodes are multicasting on 239.11.2.81, all_trusted is on.
* I have done no special monkeying with routing tables on any systems
or network appliances. Not that it matters since they're all on the
same network segment. :)
And I am getting exactly the setup you want - on my gmetad-web front page,
I see the metacluster data, the fileserver cluster, and the other cluster.
No machine overlap.
Although the solaris fileservers appear to be "dropping" metrics
intermittently. I'm looking into that... :)
Hope this answers your question...
