Have you checked out the munin scripts? If you want a starting point for something that pulls data out of jmx and prints it out (so you can catch and graph it). It is very easy to setup this to get any jmx value you want.
From: Aaron Morton [mailto:email@example.com]
Sent: Sunday, September 12, 2010 3:40 PM
Subject: Re: Monitoring with Cacti
This is my first encounter with cacti, and it's feels a lot like having a cactus violently inserted in me :) Hopefully this week I can get back to it with a clearer head, part of my annoyance was probably trying to rush it through on a Friday and it's somewhat taxing configuration.
Over the weekend I was thinking about going with some python (our in house favorite) in front of the jmxterm jar.
I'll also try to learn a bit more about cacti, it cannot be as hard as it seemed on Friday.
I'll email you out of the list this week if I make some progress.
On 11 Sep, 2010,at 03:31 PM, Edward Capriolo <firstname.lastname@example.org> wrote:
On Fri, Sep 10, 2010 at 7:29 PM, aaron morton <email@example.com> wrote:
> Am going through the rather painful process of trying to monitor cassandra using Cacti (it's what we use at work). At the moment it feels like a losing battle :)
> Does anyone know of some cacti resources for monitoring the JVM or Cassandra metrics other than...
> - provides templates and data sources that require ssh and can monitor JVM heap and a few things.
> Coded for version 0.6* , have made some changes to stop it looking for stats that no longer exist. Missing some metrics I think but it's probably the best bet so far. If I get it working I'll contribute it back to them Most of the problems were probably down the how much effort it takes to setup cacti.
> Allows for command line access to JMX. I started down the path of writing a cacti data source to use this just to see how it worked. Looks like a lot of work.
> Thanks for any advice.
Setting up cacti is easy, the second time, and third time :)
As for cassandra-cacti-m6 (i am the author). Unfortunately, I have
been fighting the jmx switcharo battle for about 3 years now
In a nutshell there is ALWAYS work involved. First, is because as you
noticed attributes change/remove/add/renamed. Second it takes a human
to logically group things together. For example, if you have two items
"cache hits" and "cache misses". You really do not want two separate
graphs that will scale independently. You want one slick stack graph,
with nice colors, and you want a CDEF to calculate the cache hit
percentage by dividing one into the other and show that at the bottom.
If you want to have a 70 branch to cassandra-cacti-m6 I would love
the help. We are not on 7.0 yet so I have not had the time just to go
out and make graphs for a version we are not using yet :) but if you
come up with patches they are happily accepted.