I have been playing lately with collectd again, which is a very nice piece of software to monitor out things, especially when you want more than just the network statistics that other tools like MRTG. In a coffee break someone commented that it would be interesting to process and display some if the data sent by collectd daemons installed in machines in real-time, and it sounded like a funny idea, so the first thing I did was implementing the network protocol. At the moment I have a working pure-Python implementation of it working: collectd.py, capable of receiving data (sending data is stil work in progress).

As a quick introduction, you could use the module as follows:

from collectd import Reader

rdr = Reader()
while True:
    print "-" * 80
    for item in rdr.interpret():
        print item

This will join the default IPv4 multicast group (239.19274.66) used by collectd (sorry, IPv6 support is incomplete) and start listening for incoming UDP packets in the default port (25826), will interpret each incoming data packet, and then print each component of the data packet to standard output in a rather readable format:

[1237749608] localhost.localdomain/swap/swap/used [(1, 48082944.0)]
[1237749608] localhost.localdomain/swap/swap/free [(1, 1007009792.0)]
[1237749608] localhost.localdomain/swap/swap/cached [(1, 18640896.0)]
[1237749608] localhost.localdomain/users/users [(1, 5.0)]

Each data item is an instance of collectd.Values (which behaves mostly as a list, and can contain multiple values) and collectd.Notification. Both of them share some fields inherited from collectd.Data. For example, you could filter messages regarding swap space usage from machines containing foobar in their host names:

if "foobar" in item.host and item.plugin == "swap":
    print item

If you want to bind to a particular address/port or join any other multicast group than de default (sorry: joining multiple groups is not implemented right now) you can pass parameters to the constructor:

rdr = collectd.Reader(host="192.168.0.32", port=9000)

The protocol was pretty straightforward to interpret, the hardest part was guessing how to properly handle multicast, as I never had to deal with it before. Now it is time to spend some time with PyCairo, or even PyCha which looks pretty convenient, and make a small application which dumps real-time graphs to a window :-)