I had this same problem myself when I was trying to sub-divide clusters by
role ... something I agree is not entirely obvious in the documentation. My
solution was to have my /etc/init.d/gmond script spawn off multiple gmond
instances. Mine looks something like :
NUM_INSTANCES=3
for ((N=1; N<=NUM_INSTANCES; N++))
do
prog="gmond$N"
ops="--conf=/etc/ganglia/gmond$N.conf"
PORT=$((BASE_PORT+N))
[ ! -L /usr/sbin/$prog ] && ln -sf /usr/sbin/gmond /usr/sbin/$prog
You'll need separate gmond.conf files ( I have gmond1, gmond2, gmondN...)
and just start at a port number like 8600 and let port++ per instance. I'm
sure this isn't the right way to do it, but it was easier than sifting
through the sparse documentation for how to spawn multiple uniquely named
clusters within a grid. I ran into a problem where if I didn't specify a
cluster = <name> then it simply wouldn't generate any graphs. This way,
each gmond.conf represents a specific cluster. Scales very nicely ...
> From: Lozgachev Ivan <[EMAIL PROTECTED]>
> Subject: Re: [Ganglia-general] How to divide the entire list of all
> nodes into 3 parts?
> To: Ofer Inbar <[EMAIL PROTECTED]>
> Cc: [email protected]
> Message-ID: <[EMAIL PROTECTED]>
> Content-Type: text/plain; charset=us-ascii
>
> Hello All,
>
> Sunday, November 23, 2008, 11:04:00 PM, you wrote:
>
> > Hello Ofer Inbar,
>
> > Wednesday, November 12, 2008, 2:30:33 AM, you wrote:
>
> >> Lozgachev Ivan <[EMAIL PROTECTED]> wrote:
> >>> I have 3 clusters. On each node of clusters started daemon
> >>> service gmond. But
> >>> I have 1 cluster in my web-ganglia interface. All clusters
> >>> belong to the same LAN. My configuration files:
> >>>
> >>> gmetad.conf
> >>> data_source "UNN cluster" localhost
> >>>
> >>> gmond.conf
> >>> cluster {
> >>> name = "UNN cluster"
> >>> }
> >>> udp_send_channel {
> >>> host = 127.0.0.1
> >>> port = 8649
> >>> }
> >>>
> >>> gmetad service started on 1 main node in first cluster.
> >>>
> >>> Nodes belonging to different clusters are contained in one list.
> >>> How to divide the entire list of all nodes into 3 parts named (for
> >>> example) "cluster1", "cluster2", "cluster3"?
>
> >> This is another one of those counterintuitive things about Ganglia
> >> you
> >> probably would miss from reading the docs: the cluster name= in
> >> gmond.conf does *not* actually define clusters. Rather, it puts a
> >> string in the right field for gmond to get a cluster name from when
> >> it
> >> uses that node as a data source. Which hosts are members of the
> >> cluster is determined by which hosts that node knows about. If all
> >> your gmond nodes read each others' multicasts and have all of them
> >> in
> >> the XML, then they're one cluster.
>
> >> To have separate clusters, you need to separate out the traffic so
> >> that gmond nodes in each cluster only see multicast traffic from
> >> other
> >> nodes in the same cluster. You can do this in two ways:
> >> - use a different port for each cluster
> >> - use a different multicast IP address for each cluster
>
> >> Or, you can use unicast, and explicitly decide which gmond talks to
> >> which collector gmond. Just make sure each collector gmond only
> >> gets
> >> updates from other nodes in the same cluster, and not from others.
>
> >> gmetad will get the cluster's name from the name= in the gmond it
> >> uses
> >> as a date source, but the cluster *membership* doesn't depend on
> >> name.
> >> -- Cos
>
> > I try to use the different port. But all nodes contains in one list.
> > What is the problem? Must I run the gmetad daemon on main nodes of
> > clusters? Please send to me a example of configuration files.
>
> > P. S.
> > I used the next configurations:
>
> > cluster {
> > name = "cluster1"
> > }
> > udp_send_channel {
> > host = 127.0.0.1
> > port = 8649
> > }
>
>
> > cluster {
> > name = "cluster2"
> > }
> > udp_send_channel {
> > host = 127.0.0.1
> > port = 8648
> > }
>
>
> > cluster {
> > name = "cluster3"
> > }
> > udp_send_channel {
> > host = 127.0.0.1
> > port = 8647
> > }
>
> I am sorry for spam. It's working fine!
> I edited configuration files incorrectly. I use HPC2008 server and log
> on under non-administrator user. When I try to edit file and save it,
> the system saved file under my user account and main file under
> administrator account was not edited and saved :-) But gmond uses it
> file.
>
> --
> Best regards,
> Lozgachev Ivan. mailto:[EMAIL PROTECTED]
>
>
>
--
Brad E. Fino
[EMAIL PROTECTED]
858-245-9099
-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Ganglia-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ganglia-general
