pytest-patterns（pytest 模式）

Pytest Patterns

When to Use

Activate this skill when:

• Writing unit tests for service or repository classes
• Writing integration tests for FastAPI endpoints with httpx.AsyncClient
• Creating or refactoring pytest fixtures and conftest files
• Setting up factory_boy factories for test data
• Testing async code with pytest-asyncio
• Mocking external services (HTTP APIs, email, queues)
• Adding parametrized tests for input variations
• Auditing or improving test coverage

Do NOT use this skill for:

• Frontend React component or hook tests (use react-testing-patterns)
• E2E browser tests with Playwright (use e2e-testing)
• TDD red-green-refactor workflow enforcement (use tdd-workflow)
• Writing application code (use python-backend-expert)

Instructions

Test Organization

tests/
├── conftest.py              # Root conftest: DB session, async client, auth helpers
├── unit/
│   ├── conftest.py          # Unit-specific fixtures (mocked repos, services)
│   ├── services/
│   │   ├── test_user_service.py
│   │   └── test_order_service.py
│   └── repositories/
│       └── test_user_repository.py
├── integration/
│   ├── conftest.py          # Integration-specific fixtures (test DB, seeding)
│   ├── test_users_api.py
│   └── test_orders_api.py
└── factories/
    ├── __init__.py
    ├── user_factory.py
    └── order_factory.py

Naming conventions:

• Test files: test_<module>.py
• Test classes: Test<Feature> (group related tests, no __init__)
• Test functions: test_<action>_<expected_outcome> or test_<scenario>
• Fixtures: descriptive noun (db_session, authenticated_client, sample_user)

Marker conventions:

# pyproject.toml
[tool.pytest.ini_options]
markers = [
    "unit: Unit tests (no DB, no network)",
    "integration: Integration tests (real DB, real HTTP)",
    "slow: Tests that take > 1 second",
]
asyncio_mode = "auto"

Run subsets: pytest -m unit, pytest -m integration, pytest -m "not slow".

Fixture Architecture

Conftest Hierarchy

Fixtures cascade: root conftest.py provides shared fixtures; subdirectory conftest files add layer-specific fixtures.

Root conftest (tests/conftest.py):

import pytest
from httpx import ASGITransport, AsyncClient
from sqlalchemy.ext.asyncio import AsyncSession, create_async_engine, async_sessionmaker
from app.main import app
from app.database import get_db

@pytest.fixture(scope="session")
def anyio_backend():
    return "asyncio"

@pytest.fixture(scope="session")
async def engine():
    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
    async with engine.begin() as conn:
        await conn.run_sync(Base.metadata.create_all)
    yield engine
    await engine.dispose()

@pytest.fixture
async def db_session(engine):
    async with async_sessionmaker(engine, class_=AsyncSession)() as session:
        yield session
        await session.rollback()

@pytest.fixture
async def client(db_session):
    async def override_get_db():
        yield db_session
    app.dependency_overrides[get_db] = override_get_db
    transport = ASGITransport(app=app)
    async with AsyncClient(transport=transport, base_url="http://test") as ac:
        yield ac
    app.dependency_overrides.clear()

Fixture Scopes

Rules:

• Default to function scope for data isolation
• Use session scope only for expensive, stateless resources (engine, event loop)
• Never use session scope for mutable data -- tests will interfere with each other
• Fixtures that yield must clean up (rollback, delete, close)

Auth Fixtures

@pytest.fixture
def auth_headers():
    """Return authorization headers for a standard test user."""
    token = create_test_token(user_id=1, role="member")
    return {"Authorization": f"Bearer {token}"}

@pytest.fixture
async def authenticated_client(client, auth_headers):
    """AsyncClient pre-configured with auth headers."""
    client.headers.update(auth_headers)
    return client

@pytest.fixture
def admin_headers():
    """Return authorization headers for an admin user."""
    token = create_test_token(user_id=99, role="admin")
    return {"Authorization": f"Bearer {token}"}

Factory Pattern

Use factory_boy for consistent, overridable test data.

import factory
from app.models import User, Order

class UserFactory(factory.Factory):
    class Meta:
        model = User

    id = factory.Sequence(lambda n: n + 1)
    email = factory.LazyAttribute(lambda o: f"user{o.id}@example.com")
    display_name = factory.Faker("name")
    role = "member"
    is_active = True

class OrderFactory(factory.Factory):
    class Meta:
        model = Order

    id = factory.Sequence(lambda n: n + 1)
    user_id = factory.LazyAttribute(lambda o: UserFactory().id)
    total_cents = factory.Faker("random_int", min=100, max=100000)
    status = "pending"

Usage in tests:

def test_user_defaults():
    user = UserFactory()
    assert user.is_active is True
    assert user.role == "member"

def test_user_override():
    admin = UserFactory(role="admin", display_name="Admin User")
    assert admin.role == "admin"

def test_user_batch():
    users = UserFactory.build_batch(5)
    assert len(users) == 5

SQLAlchemy integration (for integration tests that persist to DB):

class UserFactory(factory.alchemy.SQLAlchemyModelFactory):
    class Meta:
        model = User
        sqlalchemy_session = None  # Set per-test via conftest

    # ... fields same as above

Set session in conftest:

@pytest.fixture(autouse=True)
def set_factory_session(db_session):
    UserFactory._meta.sqlalchemy_session = db_session
    OrderFactory._meta.sqlalchemy_session = db_session

API Integration Tests

Test FastAPI endpoints with httpx.AsyncClient against the real app, but with a test database.

import pytest
from httpx import AsyncClient

class TestUsersAPI:
    """Integration tests for /api/v1/users endpoints."""

    async def test_create_user_success(self, authenticated_client: AsyncClient):
        response = await authenticated_client.post("/api/v1/users", json={
            "email": "new@example.com",
            "display_name": "New User",
        })
        assert response.status_code == 201
        data = response.json()
        assert data["email"] == "new@example.com"
        assert "id" in data

    async def test_create_user_duplicate_email(self, authenticated_client, sample_user):
        response = await authenticated_client.post("/api/v1/users", json={
            "email": sample_user.email,
            "display_name": "Duplicate",
        })
        assert response.status_code == 409
        assert "already exists" in response.json()["detail"]

    async def test_list_users_pagination(self, authenticated_client):
        response = await authenticated_client.get("/api/v1/users?limit=10&cursor=0")
        assert response.status_code == 200
        data = response.json()
        assert "items" in data
        assert "next_cursor" in data

    async def test_get_user_not_found(self, authenticated_client):
        response = await authenticated_client.get("/api/v1/users/99999")
        assert response.status_code == 404

    async def test_unauthenticated_request(self, client):
        response = await client.get("/api/v1/users")
        assert response.status_code == 401

Key patterns:

• Use authenticated_client for protected endpoints, plain client for auth testing
• Assert status code first, then response body
• Test error paths: 404, 409, 422, 401, 403
• Test pagination parameters
• Never assert on exact timestamps or auto-generated IDs (use "id" in data)

Async Tests

With asyncio_mode = "auto" in pyproject.toml, all async def test_* functions run automatically.

async def test_async_service_call(db_session):
    service = UserService(db_session)
    user = await service.create_user(email="test@example.com", display_name="Test")
    assert user.id is not None

async def test_concurrent_operations(db_session):
    service = UserService(db_session)
    import asyncio
    results = await asyncio.gather(
        service.get_user(1),
        service.get_user(2),
        service.get_user(3),
    )
    assert len(results) == 3

Common pitfalls:

• Do NOT mix sync and async fixtures carelessly -- an async fixture can only be used by async tests
• Always use pytest-asyncio (not anyio directly) for consistency
• If a test hangs, check for un-awaited coroutines or missing async keywords

Mocking Strategy

Mock external services -- YES:

from unittest.mock import AsyncMock, patch

async def test_send_notification(db_session):
    with patch("app.services.notification.EmailClient") as mock_email:
        mock_email.return_value.send = AsyncMock(return_value=True)
        service = NotificationService(db_session)
        result = await service.notify_user(user_id=1, message="Hello")
        assert result is True
        mock_email.return_value.send.assert_called_once()

Mock the database -- NO:

# BAD: Mocking the database hides real query issues
async def test_user_service(mock_db):
    mock_db.execute.return_value = MockResult([user_dict])  # Don't do this

# GOOD: Use a real test database (SQLite or PostgreSQL in Docker)
async def test_user_service(db_session):
    service = UserService(db_session)
    user = await service.create_user(email="test@example.com", display_name="Test")
    fetched = await service.get_user(user.id)
    assert fetched.email == "test@example.com"

What to mock:

respx for HTTP mocking:

import respx
from httpx import Response

@respx.mock
async def test_external_api_call():
    respx.get("https://api.example.com/data").mock(
        return_value=Response(200, json={"key": "value"})
    )
    service = ExternalDataService()
    result = await service.fetch_data()
    assert result["key"] == "value"

Parametrized Tests

Use @pytest.mark.parametrize for testing multiple input/output combinations:

@pytest.mark.parametrize("email,is_valid", [
    ("user@example.com", True),
    ("user@sub.domain.com", True),
    ("user+tag@example.com", True),
    ("", False),
    ("not-an-email", False),
    ("@missing-local.com", False),
    ("user@", False),
])
def test_email_validation(email, is_valid):
    if is_valid:
        assert validate_email(email) is True
    else:
        with pytest.raises(ValidationError):
            validate_email(email)

Parametrize with IDs for readable output:

@pytest.mark.parametrize("status,expected_code", [
    pytest.param("active", 200, id="active-user-ok"),
    pytest.param("suspended", 403, id="suspended-user-forbidden"),
    pytest.param("deleted", 404, id="deleted-user-not-found"),
])
async def test_user_access_by_status(authenticated_client, status, expected_code):
    ...

Parametrize multiple fixtures:

@pytest.mark.parametrize("role,can_delete", [
    ("admin", True),
    ("member", False),
    ("viewer", False),
])
async def test_delete_permission(client, role, can_delete):
    headers = {"Authorization": f"Bearer {create_test_token(role=role)}"}
    response = await client.delete("/api/v1/users/1", headers=headers)
    if can_delete:
        assert response.status_code == 204
    else:
        assert response.status_code == 403

Coverage Requirements

Minimum thresholds:

• Overall: 80% line coverage
• Service layer: 90% (critical business logic)
• Repository layer: 70% (straightforward CRUD)
• Routes: 80% (all success + primary error paths)

pyproject.toml configuration:

[tool.coverage.run]
source = ["app"]
omit = ["app/migrations/*", "app/main.py", "app/__init__.py"]

[tool.coverage.report]
fail_under = 80
show_missing = true
exclude_lines = [
    "pragma: no cover",
    "if TYPE_CHECKING:",
    "if __name__ ==",
    "@overload",
]

Running coverage:

pytest --cov=app --cov-report=term-missing --cov-report=html --cov-fail-under=80

Use scripts/check-test-coverage.sh to automate coverage checks with report output.

Test Anti-Patterns

1. Testing implementation details:

# BAD: Tests internal method calls
def test_service_calls_repo(mock_repo):
    service.create_user(data)
    mock_repo.insert.assert_called_once_with(...)

# GOOD: Tests observable behavior
async def test_create_user(db_session):
    service = UserService(db_session)
    user = await service.create_user(email="a@b.com", display_name="A")
    fetched = await service.get_user(user.id)
    assert fetched is not None

2. Shared mutable state between tests:

# BAD: Module-level mutable data
users = []

def test_add_user():
    users.append(User(id=1))  # Leaks to other tests

# GOOD: Use fixtures with function scope
@pytest.fixture
def users():
    return [UserFactory()]

3. Overly broad assertions:

# BAD
assert response.status_code == 200  # Only checks status

# GOOD
assert response.status_code == 200
data = response.json()
assert data["email"] == "test@example.com"
assert data["role"] == "member"

4. Missing error path tests: Every endpoint should have tests for at least: success, not found, validation error, and unauthorized.

Examples

See references/conftest-template.py for production conftest setup. See references/factory-template.py for factory_boy patterns. See references/api-test-template.py for API integration test patterns. See references/service-test-template.py for service unit test patterns. See references/integration-test-template.py for full integration test patterns.