Monitoring Hadoop with Munin

This document is based on using Debian; some information could be different in a different distribution.

Features

Munin is a very flexible and powerful monitoring tool and framework written mainly in Perl for analyzing resources like CPU, memory, hard disks, networks, and services. Munin has two components:

1. Munin - the server (aka Grapher/Gatherer), which collects all the data and generates graphs.
2. Munin-Node - the client, which tracks the data from the machine and sends it to the server.
   Munin runs on every machine that supports Perl. The database of Munin is based on RRD (Round-Robin-Database).

Requirements

• Both
  • Operating system which supports and has Perl installed (must be POSIX compatible)
• Server
  • A configured Webserver/Webapp-Framework, where Munin stores the graphs and web interface.

Installation

Munin is on many Linux distributions installable through a package manager suck as apt, yum, yast, or zypper, This section explains how it works on a Debian 5 (Lenny) machine.

Munin (the server)

1. Install the server where the Datameer X distribution is located using the command:

% sudo apt-get install munin

or on Fedora:

% sudo yum install munin

This installs the server and the client. If you don't want the client on the monitoring machine, then remove munin-node from autostart configuration, such as inittab or rc.d (based on distribution). On some distributions there are no prepackages for Munin, you need to install them manually.

1. After installation, Munin created the graphs/web interface at the default location, htmldir which can be configured in /etc/munin/munin.conf. The default htmldir is /var/www/munin. If this folder hasn't been created, then wait a while, because the Munin-Server is not a listener which runs as process; it's a script that is executed through Cronjob. You can run the cronjob-run manually as the root user using the following command:

   % su -c /usr/bin/munin-cron munin

   If the Script does not return information usin stdout, then everything should be OK.

2. Now, you can access Munin through any browser. 

   http://SERVERADDRESS/munin

   You should see "Munin :: Overview" with a list of machines which will be monitored. In this example, you should see the machine you configured as the Munin-Server itself.

Munin-Node (the client)

1. Install the client where the Datameer X distribution is located, using the command:

   % sudo apt-get install munin-node

   or on Fedora:

   % sudo yum install munin-node

2. Next, run a process called munin-node. Munin-Node uses port 4949 by default. If you installed the server and the client on the same machine, you should see graphs in the web interface which are updated every five minutes.

Resources

Locations of the data and configuration, based on the default configuration.

Manage Service

% /etc/init.d/munin-node start

% /sbin/service munin-node start

% /etc/init.d/munin-node status

% /sbin/service munin-node status

% /etc/init.d/munin-node stop

% /sbin/service munin-node stop

% /etc/init.d/munin-node restart

% /sbin/service munin-node restart

% ntsysv

% ntsysv

Configuration

Munin (the server)

By default, you can find the configuration at /etc/munin/munin.conf

You can define the structure of monitored machines for the web interface. (The format looks similar to INI-Configuration-sections.)
For example:

[localhost.localdomain]
    address 127.0.0.1
    use_node_name yes

Use a group called localdomain and associate it with the machine localhost and use localhost as the name instead the address in the web interface.
To add a server, copy it and change localhost to the name of the machine and change the IP address to the correct IP address of the machine you want to monitor. For more detailed instructions, see Possible Configuration-Parameters for Munin (Server).

Munin-Node (the client)

By default you can find the configuration at /etc/munin/munin-node.conf. Additional information about this configuration can be found at Possible Configuration-Parameters for Munin-Node (Client).

If the cidr_* parameter won't work in the configuration, use allow instead; which command is supported depends on the version of Perl Net::Server.

Links

• Official homepage
• Possible configuration parameters for Munin (Server)
• Possible configuration parameters for Munin-Node (Client)
• CIDR-Notation