Open-source QA automation framework
A complete QA engineering framework that validates web applications at three layers — UI, API, and database — built to be maintainable, CI/CD-native, and fully observable.
CI pipeline
Stack
Validated at every layer
Playwright specs drive real browsers
Page objects · video & traces · visual baselines
Contract and error-path assertions
Typed client · JSON Schema · auth flows
Integrity checked with direct SQL
SQLAlchemy · SQLite ↔ PostgreSQL · polling
Many teams accumulate test scripts that validate one layer in isolation: UI checks that fail noisily on data problems, API scripts disconnected from real user flows, and no verification of what actually landed in the database. Without shared architecture or pipeline integration, regression testing stays manual, slow, and easy to skip.
Nexus treats test code as production code. One Python framework owns all three validation layers behind clean abstractions, marker-based suites map directly to pipeline stages, and Docker Compose orchestrates the database, the application under test, and the runner with healthcheck gating. Every failure ships with its evidence attached.
Playwright drives cross-browser journeys through Page Object Model structure. Contexts record video; failures capture screenshots and traces. A visual validator compares screenshots against pixel baselines, and a scanner performs baseline accessibility checks such as missing alt text.
A reusable REST client returns typed response models. Responses are validated against JSON Schema contracts before assertions run, so contract drift is caught at the API level — before any browser is involved.
SQL assertions verify what the application actually persisted. A single SQLAlchemy abstraction serves SQLite for zero-dependency local runs and a PostgreSQL 16 service container for Dockerized and CI runs, with connection polling so suites never race the database.
smoke, regression, e2e, api, database, visual, and accessibility markers let pipelines run exactly the right depth of testing at the right time.
docker-compose brings up PostgreSQL (healthcheck-gated), a FastAPI mock application under test, and the automation runner — tests start only when every dependency is genuinely ready.
GitHub Actions runs the smoke suite on every push and pull request, full regression nightly, and enforces ruff lint plus mypy type checks as a pull-request quality gate. Allure results upload even on failure.
Allure reports collect screenshots, logs, video, traces, and environment metadata — including which database backend served each run. Failures arrive with complete context for debugging.
Pydantic-typed configuration per environment, a secrets manager, retry engine, structured logging, execution timing metrics, and Faker-backed test data factories keep the framework maintainable as it grows.
pytest-xdist parallelizes suites; pytest-rerunfailures retries flaky steps before they page anyone — with reruns visible in reports rather than hidden.
- name: Build Docker Images
run: docker compose build
- name: Execute Smoke Suite
run: docker compose run --rm -e TEST_ENV=docker automation pytest -m smoke
- name: Upload Allure Results
if: always()
uses: actions/upload-artifact@v4