Skip to content

A Pytest plugin for speeding up failure detection via automated test prioritization

License

Notifications You must be signed in to change notification settings

softwareTestingResearch/pytest-ranking

Repository files navigation

pytest-ranking

CI PyPI pre-commit.ci status

A Pytest plugin that implements Regression Test Prioritization (RTP) for faster regression fault detection.

This pytest plugin allows you to find regression test failures faster and receive testing feedback sooner from CI build. It speeds up the failure detection of your test suite by executing earlier the tests that are faster, recently failed, and more code-change-related.

Installation

To install pytest-ranking via pip from PyPI:

pip install pytest-ranking

Usage

Pytest will automatically find the plugin and use it when you run pytest. You can run pytest-ranking with its default configuration, which runs faster tests first by passing the --rank option:

pytest --rank

Before the test run starts, if --rank is passed, the terminal header will report pytest-ranking's configuration of this run, for example:

Using --rank-weight=0-0-0
Using --rank-level=put
Using --rank-hist-len=50
Using --rank-seed=1744140050
Using --rank-replay=None

After the test run finishes, the terminal summary will show the overhead of pytest-ranking in this run, for example:

============================================= pytest-ranking summary info =============================================
Number of changed Python files: 0
Time to compute test-change similarity (s): 0.000865936279296875
Time to reorder tests (s): 0.0003600120544433594
Time to collect test features (s): 0.0004608631134033203

Optimizing test prioritization heuristics

You can set the weights of different test prioritization heuristics by passing the optional --rank-weight flag with formatted values:

pytest --rank --rank-weight=0-1-0
  • Weights are separated by -
    • The first weight is for running faster tests
    • The second weight is for running recently failed tests
    • The third weight is for running tests more similar to the changed *.py files since the last run
  • All weights must be integers or floats, and their sum will be normalized to 1
  • A higher weight means that a corresponding heuristic is favored.

The default value is 1-0-0, which only prioritizes faster tests.

Optimizing test prioritization levels

You can set at which level of your test suite will be reordered, by passing the optional --rank-level flag in one of these values: put, function, module, dir. For example:

pytest --rank --rank-level=function
  • The smallest test item that can be reordered in pytest test suite is parametrized unit test (PUT)
  • This option allows you to set at which level the reordering takes place:
    • put reorders the each PUT and re-arranges their order based on their assigned priority scores
    • function reorders each test function, parametrized values of a test function follow their default order
    • module reorders each test file, all tests in the test file follow their default order
    • dir reorders each test directory, all tests within each directory follow their default order

The default value is put.

Replaying specified test order

You can run/replay tests in a specific order by listing the to-be-run test IDs in a text file, where each line is a test ID, and pass the file path to the optional --rank-replay flag:

pytest --rank --rank-replay=replay_order.txt

Tracking data from historical runs

You can also set the maximum value of the number test runs since a test's last failure that could be recorded for each test, by passing the optional --rank-hist-len flag:

pytest --rank --rank-hist-len=30

The default value is 50. Note that pytest-ranking does not store any historical test run logs, it merely updated its cached data from the previous run with data from the latest run.

Running tests in random order

You can prompt pytest-ranking to run tests in random order, by setting the sum of --rank-weight option to 0, e.g., --rank-weight=0-0-0. You can also set the seed used when running tests in random order, via setting an integer to the option --rank-seed. For example, the command below runs tests randomly with seed 1234:

pytest --rank --rank-weight=0-0-0 --rank-seed=1234

By default, pytest-ranking uses 0 as the seed.

Setting configurable options via config file

You can always apply available options by adding them to the addopts setting in your pytest.ini.

For example, create pytest.ini in your codebase root folder as such:

[pytest]
addopts = --rank --rank-weight=0-1-0 --rank-hist-len=30

and run pytest on the command line.

Alternatively, you can also create pytest.ini in your codebase root folder as such:

[pytest]
rank_weight=0-1-0
rank_hist_len=30

and run pytest --rank on the command line.

Deployment

pytest-ranking is easy to deploy into CI workflow, please see deployment.

Compatibility

pytest-ranking works with test selection and parallelization. It also works with plugins for ordering tests, e.g., pytest-order, pytest-dependency by running ordered tests first in their declared order. Pytest options that order tests generally (e.g., --ff), or plugins that randomly order tests (e.g., pytest-randomly, pytest-random-order, pytest-reverse), can interfere with pytest-ranking as they use the same reordering hook.

Reference

Demo video

A 5-minute demo video with walkthrough of pytest-ranking: YouTube link

Bibtex citation

@inproceedings{cheng2025pytest,
  title={{pytest-ranking: A Regression Test Prioritization Tool for Python}},
  author={Cheng, Runxiang and Ke, Kaiyao and Marinov, Darko},
  booktitle={Companion Proceedings of the 33rd ACM International Conference on the Foundations of Software Engineering},
  year={2025},
}

Contributing

Contributions are very welcome. Tests can be run with tox.

License

Distributed under the terms of the MIT license, pytest-ranking is free and open-source software.

Issues

If you encounter any problems, please file an issue or pull request along with a detailed description.

About

A Pytest plugin for speeding up failure detection via automated test prioritization

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published