Testing in CI/CD — Layers & Strategy
Test Layers in the Pipeline
Each test layer has different speed, coverage, and placement in the pipeline.
┌──────────────────────────────────────────────────────┐
│ CI Pipeline │
│ │
│ [Lint] → [Unit] → [Integration] → [API] → [E2E] │
│ 30s 1min 3min 5min 5min │
│ │
└──────────────────────────────────────────────────────┘
Fail fast: cheapest checks run first.
Unit Tests
When: every commit, every PR. Speed: < 2 minutes for 1000+ tests. Rule: no I/O — no DB, no HTTP, no filesystem.
- name: Unit tests
run: uv run pytest -m unit -n auto --tb=short
Unit tests failing = code is broken. Stop the pipeline here.
Integration Tests
When: after unit tests pass, on PR and main branch. Speed: 3–5 minutes. Rule: real DB and services, no external third parties.
services:
postgres:
image: postgres:16
env:
POSTGRES_PASSWORD: test
options: >-
--health-cmd pg_isready
--health-interval 10s
--health-timeout 5s
--health-retries 5
steps:
- name: Integration tests
run: uv run pytest -m integration -n auto
env:
DB_HOST: localhost
DB_PORT: 5432
API Tests
When: after deployment to staging. Speed: 5–10 minutes. Rule: tests run against the deployed service, not mocks.
- name: API tests
run: uv run pytest -m api -n auto --timeout=60
env:
TEST_ENV: staging
TEST_BASE_URL: ${{ vars.STAGING_URL }}
TEST_API_TOKEN: ${{ secrets.STAGING_API_TOKEN }}
E2E Tests
When: after API tests pass on staging, before production deploy. Speed: 5–10 minutes (smoke only in pipeline). Rule: only critical user journeys. Minimum count, maximum value.
- name: E2E smoke tests
run: |
uv run playwright install chromium --with-deps
uv run pytest -m "smoke and e2e" --timeout=90
Contract Tests
When: on PR, before integration tests, in microservices architectures. Speed: < 1 minute per service pair. Rule: verify API contract between consumer and provider without running both services.
- name: Contract tests (Pact)
run: |
uv run pytest -m contract
uv run pact-verifier --provider-base-url=http://localhost:8000
Contract tests prevent the "provider changed the API, consumer broke in production" failure mode.
Parallel Execution
Parallelise both jobs and test runners:
test:
strategy:
matrix:
python-version: ["3.12", "3.13"]
steps:
- run: uv run pytest -n auto # parallel within each matrix job
For large suites, shard tests across workers:
test:
strategy:
matrix:
shard: [1, 2, 3, 4]
steps:
- run: |
uv run pytest -n auto \
--shard-id=${{ matrix.shard }} \
--num-shards=4
Selective Execution
Run different test subsets depending on trigger:
| Trigger | Tests |
|---|---|
| PR | lint + unit + integration |
| Merge to main | + API tests on staging |
| Before production deploy | + E2E smoke |
| Nightly schedule | Full regression including slow tests |
- name: Run tests
run: |
if [[ "${{ github.event_name }}" == "pull_request" ]]; then
uv run pytest -m "unit or integration" -n auto
else
uv run pytest -m "not slow" -n auto
fi