Munin—the Raven Reports

Long-term monitoring with Munin is not restricted to system parameters. Why not monitor data of personal interest or data of interest to your colleagues?

The long-term monitoring tool Munin was developed in Norway, and its name is derived from Norse mythology. One of the two ravens who report the news of the world to the god Odin is called Munin, and the other is named Hugin. Munin is “memory”, and Hugin is “thought”.

The cool thing about the Munin long-term monitoring suite is that it's not restricted to supervising only typical system and network parameters. For example, you easily could do your colleagues from the marketing department a favor and monitor product sales for them. All you need to do is to write a script or program in the language of your choice that returns the current value of the monitored parameter in the form <parameter>.value <value> (ending with a newline character) on the standard output.

It's a little less trivial to understand how Munin actually works. This is because older parts of the official documentation (including the source code written in Perl) claim that Munin was a client-server application, which has caused much confusion. More recently, the Munin developers have referred to it as having a master-node architecture, which is far more appropriate.

A Master to Collect, Store and Present Data

To use the monitoring software, you need a machine that hosts a Web server and has Tobias Oetiker's RRDtool installed. On this computer, you run your Munin central, the Munin master. Debian and Ubuntu users will find it in the munin package. It consists of a set of Perl scripts to be run by the cron dæmon in five-minute intervals. These scripts collect data from several Munin nodes, archive them in round-robin databases (RRDs), generate diagrams and update the Web pages that present them. One of them, dubbed munin-limits, also warns about values that violate limits, if configured accordingly.

RRDs have the advantage of never changing size. As time goes by, older data is squashed and stored in less detail until it finally phases out completely. Munin uses RRDs that store the data from the past 48 hours in five-minute resolution. Average values for the past ten days are stored with a resolution of half an hour; average values for the past 46 days are stored with a resolution of two hours, and average values for the past 449 days are stored with a resolution of one day.

The Munin master reads the munin.conf configuration file usually placed in /etc/ (under Debian/Ubuntu, /etc/munin/munin.conf) to find out where to ask for data and under which (host)name it should appear in the Web interface. A typical configuration entry looks like this:

    use_node_name yes

In brackets, you put the name of the machine to be monitored (here, localhost.localdomain). If you use a Fully Qualified Domain Name (FQDN), Munin automatically will present this machine as a member of a group named after the given domain part. If you prefer to use your own groups, add the relevant group name in front of the machine name in brackets (here, Airport; the result is shown in Figure 1).

Make sure you use a semicolon as the delimiter, without any whitespaces before and after, and decide on the group name before you start monitoring. This will make life easier for you, as Munin uses the group name as the name of the directory that contains the round-robin databases and the diagrams (in our example, on an Ubuntu system, /var/www/munin/Airport/).

Figure 1. Munin's main page presents all plugins ordered by group and, within a group, by hostname. Categories available for each host are presented in brackets.

The filenames of the databases and diagrams contain the hostname given in brackets. If you change the content of the brackets afterward, make sure to change file and directory names accordingly (and before the next five-minute interval is over); otherwise, Munin will use new empty RRDs, and you may lose data.

From Where to Obtain Data?

Use the address parameter to specify the Internet address from which the Munin master obtains the relevant data. This allows you to use an intermediate slave machine to gather data from the actual target machines without giving misleading information in the Munin Web interface.

What seems to complicate matters at first glance is actually a very useful feature, as it allows you to restrict the plain-text communication of the Munin protocol to trusted machines. In addition, you do not always have the opportunity to install the Munin node software on the actual target machine. This is, for example, the case if a Munin node gathers data via SNMP. In this case, you must set the use_node_name parameter to no.

On each machine given as the value for an address parameter, you need to install the Munin node software; otherwise, the Web pages generated by the Munin master will remain empty. On Debian/Ubuntu systems, the relevant package is called munin-node.

The Munin node consists of a dæmon that, on request from the Munin master, starts the plugins responsible for collecting their specific type of data. Its configuration file, munin-node.conf, is stored in the same directory as the munin.conf; don't confuse the two if your Munin master runs on a machine that also acts as a node. munin-node.conf defines, among other things, the log file and log level, the port to be used (usually 4949), and most important, the machines that are allowed to connect to the dæmon:

allow ^127\.0\.0\.1$

In this case, the regular expression ^127\.0\.0\.1$ restricts access to the Munin master running on the same machine, localhost.