akhvorov / TraceSim

TraceSim algorithm for measuring stacktrace similarity

TraceSim

TraceSim is a method for measuring stacktrace similarity.

We describe details of algorithm in our MaLTeSQuE 2020 paper "TraceSim: A Method for Calculating Stack Trace Similarity", which can be found here: https://arxiv.org/abs/2009.12590.

Also, we implement other well-known methods for comparison on our data.

Usage

python main.py
    --stacktrace_dir [path_to_dir_with_stacktraces]
    --labels_path [path_to_labels] 
    --method tracesim

We assume that

• stacktrace_dir is a directory with json files. See example of report file with stacktrace below. Expected that report files have name like id.json, where id is stacktrace id.
• labels_path is a csv file with header rid1,rid2,label.
• method mean name of the method for measure similarity. One of the lerch, moroo, rebucket, cosine, levenshtein, brodie, prefix. Also, you can pass all for all methods comparison.

Report example

{
  "id": 123,
  "timestamp": 1.942130912798E12,
  "class": [
    "java.lang.ClassNotFoundException",
    "java.lang.RuntimeException"
  ],
  "frames": [
    [
      "java.lang.ClassLoader.loadClass",
      "java.lang.Class.forName0",
      "java.lang.Class.forName",
      "java.util.concurrent.FutureTask.run",
      "java.util.concurrent.ThreadPoolExecutor.runWorker",
      "java.util.concurrent.ThreadPoolExecutor$Worker.run",
      "java.lang.Thread.run"
    ]
  ]
}

• id - unique id of report
• timestamp - timestamp of error occurring
• class - classes of thrown exceptions
• frames - one element list of list of method calls

This file should located in stacktrace_dir and have name 123.json

Of course, this report is not real :)

Labels example

The file with labels contains information about whether two reports are is similar or not. We consider reports similar if they belong to the same group, and different otherwise.

Thus, for reports 123 and 124 from the same group, 123,124,1 will be written in the file. For report 125 from another group, 123,125,0 will be recorded. Also, by definition, it is correct to write 123,123,1, but you should not do this.

In this case, the label file may look like this

rid1,rid2,label
123,124,1
123,125,0
124,125,0