translator-openpredict/TESTING.md

Testing plan for the OpenPredict API published at https://openpredict.semanticscience.org

Manual tests

Use the docs/openpredict-examples.ipynb Jupyter notebook to manually try queries against the OpenPredict API.

Automated testing plan

Testing of the Translator OpenPredict API is separated in 3 parts:

• Integration: the API is tested using integration tests, on a local API started for the tests, at every push to the master branch. This allows us to prevent deploying the OpenPredict API if the changes added broke some of its features
• Production: the API hosted in production is tested by a workflow everyday at 1:00 GMT+1, so that we are quickly notified if the production API is having an issue
• Deployment: a workflow tests and publish the OpenPredict API Docker image build process to insure the API can be redeployed easily

When one of those 3 workflows fails we take action to fix the source of the problem.

Requirements to run the tests: Docker

To run the test locally, you will need to first start the OpenPredict API with docker:

docker-compose up

Production tests

Integration tests are run automatically by the GitHub Action workflow .github/workflows/run-tests-prod.yml everyday at 01:00am GMT+1 on the OpenPredict production API

We test for an expected number of results and a few specific results.

• POST /query TRAPI operation by requesting:
  • Predicted drugs for a given disease
  • Predicted diseases for a given drug
• GET /predict BioThings API operation by requesting:
  • Predicted drugs for a given disease
  • Predicted diseases for a given drug

Run the tests of the OpenPredict production API locally:

docker-compose exec api pytest tests/production

Integration tests

Integration tests on a local API are run automatically by the GitHub Action workflow .github/workflows/run-tests.yml at each push to the master branch.

We test the embeddings computation with a Spark local context (setup with a GitHub Action), and without Spark context (using NumPy and pandas)

You can run all the local integration tests with docker-compose:

docker-compose run tests

To run a specific test in a specific file, and display print() lines in the output:

docker-compose run tests --entrypoint pytest tests/integration/test_openpredict_api.py::test_post_trapi -s

Docker tests

At each new release we run the GitHub Action workflow .github/workflows/publish-docker.yml to test the deployment of the OpenPredict API in a Docker container, and we publish a new image for each new version of the OpenPredict API.