unittest

To use Captain with unittest, you need to configure your test suite to output test results to a file and then tell Captain where to find those test results.

Getting started

unittest does not natively produce a test results artifact that Captain can parse. To integrate, you'll need to install and configure the unittest-xml-reporting test runner. For details on its usage, refer to their README where they have details on Django support and command line usage.

Once you have unittest-xml-reporting installed and configured, configure Captain by creating a .captain/config.yaml file in the root directory of your repository:

test-suites:
  your-project-unittest:
    command: python -m xmlrunner -o tmp
    results:
      language: Python
      framework: unittest
      path: tmp/*.xml

You can change your-project-unittest to any name you like, but we typically recommend using the name of your project followed by a dash followed by unittest. The command is the command you already use to run your test suite. Captain will invoke this command to run your tests. The example above shows what you might use if you use python -m xmlrunner and want to store test results in tmp.

Once Captain is configured, you can run captain run your-project-unittest --print-summary. If you see your typical test output following by a captain block like this:

--------------------------------------------------------------------------------
----------------------------------- Captain ------------------------------------
--------------------------------------------------------------------------------

then you've configured everything correctly! You can now supercharge your test framework's capabilities. See below for configuring each of Captain's features.

Quarantining Tests

Traditionally, you might mark a test as pending or skipped to triage flaky or failing tests. With Captain, you can quarantine them instead. When only quarantined tests fail, Captain will still report your build as successful and exit with a 0 exit code. Unlike skipped tests, quarantined tests will continue to run, so you can still view their failure messages and see how frequently they are failing.

You can quarantine tests in OSS mode with captain add quarantine like so:

captain add quarantine your-project-unittest \
  --file test.py \
  --description "test.TestFunctions.test_truth"

See the identifying tests section of this page for more information on finding the file and description, and see the OSS quarantining guide for more information on managing quarantined tests in OSS mode.

Retrying Tests

You can configure Captain to automatically retry failed tests to help you determine if failing tests are flaky or are genuinely failing. To configure retries, update your .captain/config.yaml file like so:

test-suites:
  your-project-unittest:
    command: python -m xmlrunner -o tmp
    results:
      language: Python
      framework: unittest
      path: tmp/*.xml
    output:
      print-summary: true
    retries:
      attempts: 2
      command: python -m xmlrunner -o tmp {{ tests }}

Once configured, Captain will invoke your original test command, check for any failures, and retry your tests however many times you've specified (in this example, two additional times) by templating the failures into the command specified by retries.command. The output.print-summary option is not required, but we've added it for convenience in understanding the overall results after the retries have been factored in.

Partitioning

Captain can optimally partition your test suite's files into multiple groups for execution on multiple CI nodes. Captain tracks your test file runtime so that it can balance each partition.

You can configure Captain to partition your tests by updating your .captain/config.yaml file like so:

test-suites:
  your-project-unittest:
    command: python -m xmlrunner -o tmp
    results:
      language: Python
      framework: unittest
      path: tmp/*.xml
    output:
      print-summary: true
    partition:
      command: python -m xmlrunner -o tmp {{ testFiles }}
      globs:
        - test/**/test_*.py

Captain will fill in the testFiles placeholder of your partition.command with the files resulting from expanding your configured partition.globs.

Running with Partitioning

Partitioning the files requires specification of the total number of partitions (--partition-total) and the specific, 0-based partition (--partition-index) being run. These values are used alongside the configured partition command and glob patterns.

captain run your-project-unittest --partition-index 0 --partition-total 8

This command partitions your suite into 8 groups and fills in the testFiles at the specified index -- in this case, partition 0.

When initially configuring partitioning, we recommend comparing the test count with and without partitioning to ensure all of your tests are being run (i.e. all of your test files are covered by the configured globs).

GitHub Actions Example

Partitioning works on any CI provider with parallelized or matrix jobs. Partition your tests by passing the appropriate values through to captain run.

Here is a full example using a GitHub Actions workflow:

run_tests:
  runs-on: ubuntu-latest
  strategy:
    fail-fast: false
    matrix:
      partition_index: [0, 1, 2, 3, 4, 5, 6, 7]
      partition_total: [8]
  steps:
    - uses: actions/checkout@v3
    - uses: actions/setup-python@v4
    - run: pip install -r requirements.txt
    - uses: rwx-research/setup-captain@v1
    - name: run tests
      run: |
        captain run your-project-unittest \
        --partition-index ${{ matrix.partition_index }} \
        --partition-total ${{ matrix.partition_total }}

Identifying Tests

Captain uses framework specific "identity recipes" to identify the tests in your suite. These recipes are order dependent components extracted from native test framework output.

We use this identity to track the executions of a test over the course of their lifetime in your suite. This enables us to do things like flake detection, quarantining, and retries.

For unittest, Captain constructs the identity by parsing out the file and description attributes.