[CentOS] Ganglia

Thu Jun 17 22:20:03 UTC 2010
Whit Blauvelt <whit at transpect.com>

On Tue, Jun 15, 2010 at 04:07:57PM +0100, Simon Billis wrote:

> Take a look at ganglia - http://ganglia.sourceforge.net/
> 
> This may do what you need. 

It's what I've ended up going with. (Munin also looked promising - if I
could get the syntax right to modify its CPU test for individual cores,
which looks quite possible, I just didn't achieve it yet).

A few notes on Ganglia 3.1.7 build/install:

- best complied from source, there are big dependency problems with the
  available rpms 

- dependencies to satisfy before compilation include (among others):
  apr-devel libconfuse libconfuse-devel expat-devel pcre-devel - for the
  libconfuse I went to dag/rpmforge

- the "make install" stage doesn't fully install, despite a required
  --sysconfdir flag being used. In particular, "gmond -t > gmond.conf" will
  provide the missing file to add to your config dir for that. The
  ganglia-3.1.7/gmond/modules/conf.d contents should be copied to
  /etc/ganglia/conf.d. Then a line with "include
  ('/etc/ganglia/conf.d/*.conf')" should be added to gmond.conf. And the man
  pages (in "mans" and one for gmond.conf in "gmond") may be copied
  /usr/share/man/man1 and man5 as appropriate. Also, the init files for
  gmond and gmetad need to be copied to init.d - but at least this, unlike
  the other hand-installation requirements, is documented.

- Beyond that, it's good to change the cluster "name =" in gmond.conf to
  something appropriate before you start to run. You only need gmetad
  compiled on the system to run the web reporting front end (and it takes an
  configure flag to do that). On other systems just rsycing over the
  /etc/ganglia contents will handle configuration just fine (assuming this
  is a single cluster). The web pages merely require copying to someplace in
  your PHP-capable server's space.

- The multi-core CPU graphing module - the main functionality I was after -
  requires some uncommenting in its conf file to get it going. The PCRE
  section is enough to uncomment, with pcre installed on your system.

It's pretty simple once the dependencies are installed, and the "make
install" deficiencies are worked around. It gives a _lot_ of graphs
(probably too many, but studying them over time will tell).

Whit