Flaky tests detection

Visualize tests whose state changes most often. During software development, it is often common that some tests start to randomly fail, but finding those tests is a tedious and time consuming problem. Flaky tests detection solves that problem by processing historical xunit test results and checks which tests changes state most often. Flaky tests detection is available as Github Action plugin and Python package. For usage, see example at the actions page.

Implementation is based on "Modeling and ranking flaky tests at Apple" by Kowalczyk, Emily & Nair, Karan & Gao, Zebao & Silberstein, Leo & Long, Teng & Memon, Atif.

Features

• Prints out top test names and their latest calculation window scores (normal fliprate and exponentially weighted moving average fliprate that take previous calculation windows into account).
• Calculation grouping options:
  • n days.
  • n runs.
• Heatmap visualization of the scores and history.

Parameters

Data options (choose one)

• --test-history-csv
  • Give a path to a test history csv file which includes three fields: timestamp, test_identifier and test_status.
• --junit-files
  • Give a path to a folder with JUnit test results.

Calculation options

• --grouping-option

  • days to use n days for fliprate calculation windows.
  • runs to use n runs for fliprate calculation windows.

• --window-size

  • Fliprate calculation window size n.

• --window-count

  • History size for exponentially weighted moving average calculations.

• --top-n

  • How many top highest scoring tests to print out.

Heatmap generation

• --heatmap
  • Turn heatmap generation on.
  • Two pictures generated: normal fliprate and exponentially weighted moving average fliprate score.
  • Same parameters used as with the printed statistics.

Full examples

• Precomputed test_history.csv with daily calulations. 1 day windows, 7 day history and 5 tests printed out.
  • --test-history-csv=example_history/test_history.csv --grouping-option=days --window-size=1 --window-count=7 --top-n=5
• JUnit files with calculations per 5 runs. 15 runs history and 5 tests printed out.
  • --junit-files=example_history/junit_files --grouping-option=runs --window-size=5 --window-count=3 --top-n=5
• Precomputed test_history.csv with daily calculations and heatmap generation. 1 day windows, 7 day history and 50 tests printed and generated to heatmaps.
  • --test-history-csv=example_history/test_history.csv --grouping-option=days --window-size=1 --window-count=7 --top-n=50 --heatmap

Install module

• make install

Install module and development packages

• make install_dev

Run pytest

• make run_test

Acknowledgement

The package was developed by F-Secure Corporation and University of Helsinki in the scope of IVVES project. This work was labelled by ITEA3 and funded by local authorities under grant agreement “ITEA-2019-18022-IVVES”