Thanks,
in fact, my problem was that I couldn't figure how a cluster group
(belonging nodes and associated name) was defined.
On Tue, 15 Oct 2002 13:28:51 -0700
Steven Wagner <[EMAIL PROTECTED]> wrote:
> Also, everything connected to the same multicast IP is, for all intents and
> purposes, on the same cluster... as far as the monitoring core's concerned.
That is what I was searching!
> gmetad apparently overrides these labels (although the perl one doesn't, I
> think?) with whatever you've named in gmetad.conf as the data source names.
In fact, gmetad doesn't set the name of the cluster: cluster names are defined
on the gmonds gmetad is listening to (the sources): in my gmetad I define "Clic
127.0.0.1 8649" meaning "eh gmetad listen to the gmond running locally", and in
my local gmond.conf file, I have "name "ClusterClic"". So the web front end
displays ClusterClic for the name of the cluster.
To sumarize: each node sees the cluster with a different name: running gstat
will show the same metrics but with diferent cluster name for each node.
>
> So in short, different clusters NEED to use different multicast IPs.
> Cluster names are NOT metrics - they are NOT shared amongst all member
> nodes (this could change, I guess ... ?), but are specified on each local
> host. So if you want the cluster name the same on all nodes, you'll have
> to make sure gmond.conf on all nodes has the right name in it.
I think that a new notion of cluster metrics should be used to solve this
inconsistency between nodes:
These should be defined in gmond, because gmond does not need gmetad to ensure
metrics retrieval.
As the number of processors, this metrics should not be updated each time a
gmond sends data, but where should it be set?
But I think that there is a problem in the definition of groups in gmond:
what happens now if a node wants to be part of 2 clusters (expl: 1 DNS for 2
clusters, DHCP,...)?
You have to run 2 gmonds on different multicast adresses, and on different
ports to ensure that if you want to gather the data with gmetad you won't have
2 answering gmond...
So I think that the name defined in gmond should be more significant: defining
groups, were a node can belong to more than one group (so more than one cluster
name), and were all gmond should have a cluster name.
Best regards,
Julien
>
> But it really doesn't matter if you're using gmetad. :)
>
> matt massie wrote:
> > julien-
> >
> > http://ganglia.sourceforge.net/ganglia_docs/configuration.html#GMOND-CONFIGURATION
> >
> > should answer your question. the cluster name is set by the Ganglia
> > Monitoring Daemon (gmond) in /etc/gmond.conf (by default) with the "name"
> > parameter. e.g.
> >
> > name "Cluster in Room 404b"
> >
> > ...
> > technically you don't need to set the cluster name attribute for all the
> > gmonds on your cluster.. only the gmonds which are used as a data_source
> > by gmetad. if you don't set the cluster name the default name is
> > "unspecified".
> >
> > -matt
> >
> >
> > Tomorrow, Julien Leduc wrote forth saying...
> >
> >
> >>Hi all,
> >>
> >> I have a problem: I can't set a cluster name to my nodes running gmond,
> >> because the gmetad script takes all the gmond stats on the network so that
> >> I have 4 more machines that are said to be part of my cluster, while these
> >> are just running gmond without any specified cluster name.
> >>
> >>Moreover, I have read the documentation, but can't find out were the
> >>cluster name is defined: Gmond or Gmetad, and what is the difference ?
> >>
> >> I have a 5 PCs cluster running under Mandrake 8.2 with the ganglia rpm
> >> (containing gmond 2.4.1).
> >>
> >> Thanks
> >>
> >> Julien Leduc
> >>
> >>
> >>-------------------------------------------------------
> >>This sf.net email is sponsored by:ThinkGeek
> >>Welcome to geek heaven.
> >>http://thinkgeek.com/sf
> >>_______________________________________________
> >>Ganglia-general mailing list
> >>[email protected]
> >>https://lists.sourceforge.net/lists/listinfo/ganglia-general
> >>
> >
> >
> >
> >
> > -------------------------------------------------------
> > This sf.net email is sponsored by:ThinkGeek
> > Welcome to geek heaven.
> > http://thinkgeek.com/sf
> > _______________________________________________
> > Ganglia-general mailing list
> > [email protected]
> > https://lists.sourceforge.net/lists/listinfo/ganglia-general
>
>
>
>
