Munin-node-awk

munin-node-awk is a simple implementation of munin-node, which in turn is a daemon used on servers (nodes) to query the system for statistics every time a "munin master" comes asking for it. Typically every 5 minutes. It is written in awk, with two goals:

1. Play around
2. Prototype the idea of a super-light non-forking munin-node

Usage

If you want it to run in the background, you'll have to figure it out yourself. (Ie: use normal UNIX-tricks).

::

./munin-node-awk [-v hostname=yourhostname] [-v domain=yourdomain] 
    [-v debug=integer] [-v port=tcpport]

All options are ... optional. The defaults are:

hostname uname -n

domain ".awk" (Since I still haven't added enough plugins to replace the real munin-node on my system)

debug 0

port 4919 (NB: This is NOT the default munin port. This is for the same reason as above: I run munin-node and munin-node-awk in parallel)

cpus calculated from /proc/stats and cached. Don't see the point in over-riding it, but you're welcome.

Deviations from munin-node

Because the "plugins" are all contained inside the node, they do not need to support suggest or autoconf or any magic markers. Those are aides for munin-node, and we don't need them.

munin-node-awk still doesn't support "cap" properly, mostly because there isn't a need until a plugin needs it.

I also had some issues with hidden protocol details (like the fact that munin-update expects a banner on connect, even if it has a # at the start. It simply froze until it got it). The reason for the dot (".") is also undocumented as far as the protocol goes.

Ideas

Some ideas I am testing:

• Caching and sharing of relevant variables (ie: number of cpus)
• Virtually fork-free execution. There is some need for forking, but the goal is to avoid forking for every query, which is what munin-node does.
• "Wildcard" plugins can probably be expanded
• Parallel connection to stuff like varnish to fetch statistics real time without opening a file (not implemented, just and idea right now).

This is thus far a prototype, and the one big missing thing I have is the lack of "dynamic" plugins. I have to manually add all plugins to the switch() statements. That might be acceptable, though.

Use case

Two primary use-cases:

1. Low-end systems, like dd-wrt, where you don't want a big node and a large amount of plugins.
2. Overloaded systems, like my Varnish servers when I am doing stress testing. Having munin-node fork a quadzillion processes is immensely slow if your system already has a load of 300. This could also be expanded to do real-time stats.

Credits

Munin: http://www.munin-monitoring.org

Author of munin-node-awk: Kristian Lyngstøl [email protected] (See my blog, http://kristianlyng.wordpress.com )

That's about it, I borrowed the cpu stuff from the cpu graph, but as that is uncredited too, it doesn't really help :)