Send collectd statistics from your Ruby script
Pull request Compare This branch is 8 commits behind revett:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.



ruby-collectd lets you send statistics to collectd from Ruby.

This can be useful if you want to instrument data from within daemons or servers written in Ruby.

ruby-collectd works by talking the collectd network protocol, and sending stats periodicially to a network-aware instance of collectd.


You need to have collectd load the network plugin, and listen on UDP port 25826 (so it's acting as a server):

# /etc/collectd.conf

LoadPlugin network

<Plugin "network">
  Listen "ff18::efc0:4a42"

To install the gem, make sure the Gemcutter RubyGems server is known to your local RubyGems, then install it:

gem sources -a
gem install collectd

Add sudo in front of the gem install command if you want it to be installed system wide.


require 'rubygems'
require 'collectd'

First of all, specify a server to send data to:

Collectd.add_server(interval=10, addr='ff18::efc0:4a42', port=25826)

Each server you add will receive all the data you push later. An interval of 10 is quite reasonable. Because of UDP and collectd's network buffering, you can set the interval to less than 10, but you won't see much benefit.

ruby-collectd gives you a free data collector out of the box, and it's a nice gentle introduction to instrumenting your app.

To collect memory and CPU statistics of your Ruby process, do:

Stats = Collectd.my_process(:woo_data)

In the first line, we set up a new plugin. my_process is the plugin name (magically handled by method_missing), and :woo_data is the plugin instance.

A plugin name is generally an application's name, and a plugin instance is a unique identifier of an instance of an application (i.e. you have multiple daemons or scripts running at the same time).

In the second line, with_full_proc_stats is a method provided by ruby-collectd that collects stats about the current running process. It makes use of polled gauges, which we talk about later.

Behind the scenes, with_full_proc_stats is using a simple interface you can use to instrument your own data.

Back in the first line we set up a plugin which we wanted to record some data on. with_full_proc_stats sets up types, which are a kind of data you are measuring (in this case CPU and memory usage).

You can do this yourself like this:

Stats = Collectd.my_daemon(:backend)

# Set counter absolutely
Stats.my_counter(:my_sleep).counter = 0
Stats.my_gauge(:my_gauge).gauge = 23

loop do
  # Increment counter relatively
  Stats.my_counter(:my_sleep).count! 5
  # Set gauge absolutely
  Stats.my_gauge(:my_stack).gauge = rand(40)
  sleep 5

(Don't worry if this doesn't make sense - gauges and counters are explained below)

You can also poll for your data, if you feel comfortable with that:

Stats.counter(:seconds_elapsed).polled_counter do

Glossary / collectd's data model

collectd groups data by six categories:

  • hostname is grabbed from hostname -f
  • plugin is the application's name
  • plugin-instance is passed from the programs' side with the programs instance identifier, useful if you're running the same script twice (PIDs are quite too random)
  • type is the kind of data you are measuring and must be defined in types.db for collectd to understand
  • type-instance provides further distinction and have no relation to other type-instances. Multiple type-instances are only rendered into one graph by collection3 if defined with module GenericStacked.
  • values are one or more field names and types belonging together. The exact amount of fields and their corresponding names (useful to collection3) are specified in collectd's types.db.

A value can be either of two types:

  • COUNTER is for increasing counters where you want to plot the delta. Network interface traffic counters are a good example.
  • GAUGE is values that go up and down to be plotted as-is, like a temperature graph.