Cluster monitoring (proxy setup) 🔗 link

The more I read through the archives, the more confused I am on how to
implement this. Basically, I want to monitor 64 compute nodes that are on a
private network. I do not want to have to install the client on all of the
nodes as all I care about is the ping test. I have a main hobbit server that
I want to merge the info from the cluster and from various other systems
that I monitor. 

To work as a bridge, I installed hobbit on the head node of the cluster as
it is the only system with an interface on both networks. On the head node I
disabled all of the hobbitd modules and setup bbproxy to forward messages to
my main hobbit server. This didn¹t work. After reading up on the bbproxy man
page it looks like bbproxy is really meant to forward client messages on to
a bbdisplay. Since I don¹t have the hobbit client installed on the compute
nodes it doesn¹t sound like bbproxy will work for this. I am a bit confused
on what actually does the network tests. Do I have to have hobbitd running
on the head node in order for the network tests to run? If that is the case,
how do I get hobbitd to forward information on to a different hobbit server?

I am pretty sure that I have simply failed to understand one of the man
pages or something, but I cant seem to figure out how to accomplish what I
am looking to do.

Hi.

If you wish to forward messages using multiple displays you should
change a couple of things, revert the configuration files
(hobbitlaunch.cfg) to make hobbit act as a display server then you
should change hobbitserver.cfg as follows:
BBDISP="0.0.0.0"
BBDISPLAYS="$BBSERVERIP 12.34.56.78"

The BS ip should be changed to whatever the ip is for your main hobbit
server.

This configuration *should* work. Just reply if it doesn't.

With regards,
  Sigurdur Gudbrandsson

On Tue, 2007-06-26 at 13:21 -0400, Ben Fenstermaker wrote:

bbproxy to forward messages to my main hobbit server. This didn’t

don’t have the hobbit client installed on the compute nodes it doesn’t

On 6/27/07 4:50 AM, "Sigur›ur Gu›brandsson" <user-f6f2e0690a1f@xymon.invalid> wrote:

This almost works... On my secondary hobbit server I now have it displaying
correctly again and there doesn't seem to be any errors on that side. On the
main hobbit server I don't see the status for my 64 compute nodes, however
they all show up in the ghost clients list. I figured this wasn't a problem
as I can just add them to the bb-hosts file. If I do this they always change
to red since my main hobbit server tries to query them and cant reach the
protected network.

-- 

Ben Fenstermaker

On Wed, Jun 27, 2007 at 04:03:53PM -0400, Ben Fenstermaker wrote:

Define them with a "noconn" setting, or use the "NET:foo" tag to setup
different sets of network tests.


Regards,
Henrik

Ah, finally, something I can comment on competently.  

 
The way we do it is we have two pollers and one display.  The display
does no network testing (we disabled it in hobbitlaunch.cfg) but that's
not necessarily required.  For your situation, you have to choices.

 
1)       You use the Net:hostname moniker in your bb-hosts file, and
then you can keep the bb-hosts files synced however you wish (we just
scp when we make a change).  Any hosts you don't have the Net:hostname
tag in will be tested by any servers.

2)       Maintain two separate bb-hosts, with all hosts on the display
servers' copy, but then you have to manually put a noconn tag in the
hosts to not ping in the bb-hosts file on the display, otherwise, your
display server will be testing those as well, and that status will be
over-riding the status from your other server.

 
Obviously, choice 1 is much easier to maintain, but you will have to put
a Net:thing tag in every line of your current script (easy to do with a
quick perl script or whatnot.  I have one somewhere if you need such).
I'd really recommend going that route.

 
Tod Hansmann

Network Engineer

 
  <http://www.directpointe.com/>;  

From: Ben Fenstermaker [mailto:user-1ecb3d80b308@xymon.invalid] 
Sent: Wednesday, June 27, 2007 2:04 PM
To: user-ae9b8668bcde@xymon.invalid
Subject: Re: [hobbit] Cluster monitoring (proxy setup)

 
On 6/27/07 4:50 AM, "Sigur>ur Gu>brandsson" <user-f6f2e0690a1f@xymon.invalid>
wrote:

This almost works... On my secondary hobbit server I now have it
displaying correctly again and there doesn't seem to be any errors on
that side. On the main hobbit server I don't see the status for my 64
compute nodes, however they all show up in the ghost clients list. I
figured this wasn't a problem as I can just add them to the bb-hosts
file. If I do this they always change to red since my main hobbit server
tries to query them and cant reach the protected network.

-- 
Ben Fenstermaker