Stack Trace Comparison and Clustering

Recently I have been working on a small library that parses java stack traces, compares them and makes it possible to cluster errors that are similar. When the same errors appears across different versions of software (the actual application and the libraries used) it is very useful to be able to identify that. The implementation is based on a masters thesis by John Lambert from 2002 (pdf).

I’ve decided to open source the library (Apache License v 2.0) and put it on github. No artifacts/releases available yet but it is a maven project with very few dependencies so it should be fairly easy to built on your own. You can get it from https://github.com/johanandren/Stack-Detective.

So far we have used it in an inhouse application at Databyrån that automatically receives stack traces from customers and clusters them, the result is perhaps not perfect but useful.