跳到主要内容

测试策略实战

问题​

Python 项目的测试策略应该如何设计？pytest 有哪些高级用法？

答案​

测试金字塔​

单元测试​

tests/test_service.py
import pytest
from unittest.mock import AsyncMock, patch

class TestOrderService:
    @pytest.fixture
    def service(self):
        return OrderService(db=MockDB(), cache=MockCache())

    def test_create_order_success(self, service):
        order = service.create_order(user_id=1, items=[{"id": 1, "qty": 2}])
        assert order.status == "pending"
        assert order.total > 0

    def test_create_order_empty_items(self, service):
        with pytest.raises(ValueError, match="items cannot be empty"):
            service.create_order(user_id=1, items=[])

    @pytest.mark.parametrize("discount,expected", [
        (0, 100),
        (0.1, 90),
        (0.5, 50),
        (1.0, 0),
    ])
    def test_apply_discount(self, service, discount, expected):
        result = service.apply_discount(100, discount)
        assert result == expected

Mock 外部依赖​

tests/test_api.py
from unittest.mock import patch, AsyncMock

@pytest.mark.asyncio
async def test_fetch_user_from_api():
    mock_response = AsyncMock()
    mock_response.json.return_value = {"id": 1, "name": "Alice"}
    mock_response.status_code = 200

    with patch("httpx.AsyncClient.get", return_value=mock_response):
        user = await user_service.get_user(1)
        assert user.name == "Alice"


# 使用 pytest-mock（更简洁）
async def test_send_email(mocker):
    mock_send = mocker.patch("services.email.send_email", return_value=True)
    result = await notify_user(user_id=1, message="Hello")
    mock_send.assert_called_once_with("user@example.com", "Hello")
    assert result is True

Fixture 与工厂​

tests/conftest.py
import pytest
from sqlalchemy import create_engine
from sqlalchemy.orm import Session

@pytest.fixture(scope="session")
def engine():
    """整个测试会话共享一个数据库"""
    engine = create_engine("sqlite:///:memory:")
    Base.metadata.create_all(engine)
    yield engine
    engine.dispose()

@pytest.fixture
def db_session(engine):
    """每个测试用例独立事务"""
    conn = engine.connect()
    trans = conn.begin()
    session = Session(bind=conn)
    yield session
    session.close()
    trans.rollback()  # 回滚，保持数据库干净
    conn.close()

@pytest.fixture
def user_factory(db_session):
    """工厂 Fixture"""
    def create_user(**kwargs):
        defaults = {"name": "Test User", "email": "test@example.com"}
        defaults.update(kwargs)
        user = User(**defaults)
        db_session.add(user)
        db_session.commit()
        return user
    return create_user

FastAPI 集成测试​

tests/test_api_integration.py
import pytest
from httpx import AsyncClient, ASGITransport
from app.main import app

@pytest.fixture
async def client():
    transport = ASGITransport(app=app)
    async with AsyncClient(transport=transport, base_url="http://test") as ac:
        yield ac

@pytest.mark.asyncio
async def test_create_user(client):
    response = await client.post("/users", json={"name": "Alice", "email": "a@b.com"})
    assert response.status_code == 201
    data = response.json()
    assert data["name"] == "Alice"

@pytest.mark.asyncio
async def test_get_user_not_found(client):
    response = await client.get("/users/999")
    assert response.status_code == 404

常见面试问题​

Q1: pytest vs unittest？​

答案：

特性pytestunittest
断言assert 原生self.assertEqual()
Fixture灵活、可组合setUp/tearDown
参数化@pytest.mark.parametrize第三方库
插件丰富（200+）较少
推荐新项目首选兼容旧代码

Q2: 什么时候用 Mock，什么时候不用？​

答案：

  • 用 Mock：外部 API、邮件服务、支付接口、文件系统
  • 不用 Mock：纯逻辑函数、数据库（用内存 SQLite 替代）
  • 原则：Mock 边界，不 Mock 内部逻辑

Q3: 覆盖率目标多少合适？​

答案：

  • 整体 80% 以上
  • 核心业务逻辑 90%+
  • 工具类/配置类 60%+ 即可
  • 不要追求 100%，关注有意义的测试

相关链接​