korg 0.0.8

Why a logstash / grok port to Python?

I like the logstash grok approach to logfile parsing. So I want to use this in Python.

One solution would be to use the C version of logstash / grok (https://github.com/jordansissel/grok) and to write a wrapper:

• https://github.com/kiwi0530/python-grok

• https://github.com/emgee/libgrok-py

Basically grok assembles regular expressions. I already know that in Python file processing with regular expressions is blazingly fast so I choose to port it to Python.

Unfortunately a grok package already existed in Python for something completely different - consequently I had to “reverse-engineer” it. Thus the name korg.

The pattern files are updated from the logstash grok project: https://github.com/logstash-plugins/logstash-patterns-core

A big thank you belongs to the logstash community for an awesome job maintaining the regex pattern files!

Examples using korg

• extracting metrics from logfiles: https://github.com/finklabs/loganalyser

Status

• Base functionality is implemented including tests

• Logstash patterns are included

• Some grok features are still missing (not sure which ones are necessary)

I made some first benchmarks to verify whether my performance requirements can be realized with this approach. Please do not use this results in any blog posts or articles since this is not a complete benchmark (from a statistical view point the sample size is way too small).

Processing a 1.7MB apache access log with korg

..
200
200
404
200
200

real    0m0.248s
user    0m0.172s
sys     0m0.040s

Processing the same logfile with logstash

..
200
200
404
200
200
^CSIGINT received, shutting down. {:level=>:warn}

real    0m11.752s
user    0m23.948s
sys     0m0.528s

Note: both implementations read in all available patterns from the ‘patterns’ folder.

Resources

• http://jpmens.net/2012/08/06/my-logstash-and-graylog2-notes/

• http://grokdebug.herokuapp.com/