E-commerce UI + API Test Suite — Playwright + Python

Production-style automation for a web shop: 12 UI tests and 15 API tests covering accounts, authentication, search, cart and checkout. Five offline infrastructure tests check URL configuration and browser-free API fixtures. CI runs Ruff, Pyright and smoke checks on pull requests, with full regression on main and a weekday schedule.

Target application: Automation Exercise, a public training shop. The live suites create temporary accounts and delete them afterwards. They depend on the site's availability and its access policy for the runner's network.

Coverage

Suite	Tests	Main checks
tests/ui/test_smoke.py	1	Home page title and featured products
tests/ui/test_auth.py	4	Registration, login, wrong password, existing email
tests/ui/test_search.py	2	Matching products and no results
tests/ui/test_cart.py	3	Add, quantity × price totals, remove
tests/ui/test_checkout.py	2	Guest gate and complete purchase journey
tests/api/test_accounts.py	9	Create/read back account, duplicate email, valid/invalid login, missing credentials, unsupported method, delete and reject subsequent login
tests/api/test_products.py	6	Product/brand structure, unique IDs, unsupported method, known-product search, no results, missing query
tests/unit/test_fixtures.py	5	URL precedence in real serial/xdist sessions and API fixture without a browser

The full purchase test covers registration → product → cart → delivery address → payment → confirmation → account deletion. Card details are generated test data for this training target.

Local setup

Requires Python 3.12+. From the repository root:

# Creates a local environment and installs project dependencies into it.
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[dev]'

# Downloads Chromium into the user's Playwright cache.
playwright install chromium
pytest -m smoke --html=reports/smoke.html --self-contained-html

python -m pip install -r requirements.txt remains a compatibility entry point for the same installation. Direct dependency pins, including Playwright itself, live in pyproject.toml, alongside pytest, Ruff and Pyright configuration. Transitive dependencies are resolved by pip; this is not a complete lockfile.

For all three engines, playwright install chromium firefox webkit downloads the browser binaries. On Ubuntu, playwright install --with-deps chromium firefox webkit also installs system packages via apt and may require sudo. Use it when the browser reports missing shared libraries.

Select a suite

Run these commands inside the activated virtual environment:

pytest -m smoke                         # 8 checks: 4 UI + 4 API
pytest -m regression                    # all 27 live UI/API checks
pytest -m api                           # 15 API checks; no browser install needed
pytest -m 'ui and smoke'                 # 4 quick UI checks
pytest -m critical                      # account, cart and full purchase journeys
pytest tests/unit                       # 5 offline infrastructure checks
pytest tests/ui --browser firefox -n 2   # full UI suite in another engine
pytest                                  # all 32 checks, Chromium for UI

ruff check .
ruff format --check .
pyright

smoke and critical are subsets of regression. ui and api select a layer; unit selects offline checks. Unknown markers and configuration keys fail collection.

The same base URL is used for UI and API setup. Precedence is --base-url, then PYTEST_BASE_URL, then the default in pyproject.toml; this also works with xdist. Use an override only for a compatible, authorized instance of the target.

CI

Trigger	Quality	API	UI
Pull request	Ruff, formatting, Pyright, collection, unit tests	4 smoke tests	4 smoke tests, Chromium
Push to main	Same	All 15	All 12 per engine: Chromium, Firefox, WebKit
Weekdays at 05:23 UTC / manual run	Same	All 15	All 12 per engine

The workflow checks API availability before the API suite and browser access before each UI suite. It runs API tests once, then runs the browser jobs one at a time with two workers each. Superseded runs are cancelled. This limits traffic to the shared demo site. Persistent failures remain failures; there are no blanket reruns or skips that turn an inaccessible target green.

HTML reports, JUnit XML and preflight JSON are uploaded for 14 days. UI failures also include screenshots and Playwright traces. Artifacts are uploaded even when preflight fails; an HTML report exists only if pytest actually ran.

Design decisions

• Page Object Model: selectors and actions live in pages/; tests express scenarios. Shared registration steps live in pages/flows.py.
• Independent data: fixtures generate unique accounts and register cleanup before creation. Cleanup accepts an already deleted account and reports other errors.
• Browser-free API layer: tests/api/ uses Playwright request contexts. Ad blocking is attached only when a UI test requests a browser context.
• API contracts: tests check transport, application status, messages and relevant response data against the published API scenarios. This target returns HTTP 200 even for application errors; responseCode in JSON carries codes such as 400 or 404. Unexpected redirects are exposed at the first hop.
• Typed data: User and PaymentCard describe factory outputs. Account payloads explicitly map first_name/last_name to the API's firstname/lastname.
• Readiness and diagnostics: navigation waits for DOM readiness, checks HTTP errors and waits briefly for known access-interstitial titles to disappear. It then reports a target-access error if the challenge persists. Controls use Playwright auto-waiting.
• Dependency checks: Ruff and Pyright run in CI. Offline fixture checks protect the CLI/environment URL precedence and ensure API tests cannot silently start a browser.

Findings and bug reporting

The CI failure analysis links to the actual failing run. Its Chromium/Firefox jobs received an access challenge (One moment, please...), while account API requests entered HTTP 302 loops. Changing selectors cannot resolve a persistent site-side access restriction; the runner needs access to the training target.

A reproduced open-source bug report includes a minimal offline example of an ini-configured URL becoming None on xdist workers. This is a confirmation of an existing upstream issue, with the workaround covered by our fixture tests. It is not presented as a newly discovered or newly submitted defect.

Search field note: searching for dress has returned products whose displayed names do not contain that term. The public scenarios do not define name-only matching. The matching fields and expected relevance need owner confirmation, so this remains an observation rather than a confirmed product defect. The API suite additionally checks that a search using an actual catalog product name returns that exact product.

Possible next steps

• Visual regression for key pages.
• A locally hosted open-source shop for deterministic CI independent of the public demo.
• A generated lockfile for transitive dependencies.