乐于分享
好东西不私藏

OpenHands 源码-测试架构与 CI/CD

OpenHands 源码-测试架构与 CI/CD

OpenHands 源码解析系列

第 16 讲:测试架构与 CI/CD

基于 OpenHands 源码 · 2026-08-12

一、测试体系全景

OpenHands 的测试体系覆盖了后端 Python、前端 React/TypeScript、企业版模块、Docker 镜像构建以及发布流水线。整个仓库包含:

🔹 114 个 Python 后端单元测试 (tests/unit/)

🔹 178 个 企业版测试 (enterprise/tests/)

🔹 299 个 前端测试文件 (frontend/)

🔹 26 个 GitHub Actions 工作流

测试金字塔

OpenHands 遵循经典的三层测试策略:前端 Playwright E2E → 后端 pytest 单元 → 代码质量 Lint/类型检查,每层独立并行执行。

二、pytest 测试框架配置

OpenHands 的测试依赖通过 Poetry 的 test 依赖组管理。让我们看 pyproject.toml 中的配置:

📄 pyproject.toml (第 124-135 行)

test = [
  "gevent==25.9.1",
  "pandas",
  "pytest==9.0.3",
  "pytest-asyncio==1.3.0",
  "pytest-cov==7.1.0",
  "pytest-forked==1.6.0",
  "pytest-playwright==0.7.2",
  "pytest-timeout==2.4.0",
  "pytest-xdist==3.8.0",
  "reportlab==4.4.10",
]

关键插件解读:

插件作用为什么需要
pytest-asyncio异步测试支持OpenHands 大量使用 FastAPI + async/await
pytest-cov覆盖率收集PR 覆盖率评论 + 质量门禁
pytest-forked隔离测试进程避免全局状态污染(如 Docker、环境变量)
pytest-xdist并行执行-n auto 利用多核加速
pytest-playwright浏览器自动化E2E 测试前端 UI
pytest-timeout超时保护防止 CI 阻塞

三、CI 流水线架构

每个 PR 和 main 分支推送都会触发以下流水线:

1. Python 测试流水线 (py-tests.yml)

📄 .github/workflows/py-tests.yml (第 59-60 行)

      - name: Run Unit Tests
        run: PYTHONPATH=".:$PYTHONPATH" poetry run pytest \
              --forked -n auto -s ./tests/unit \
              --cov=openhands --cov-branch

执行策略:

🔹 --forked:每个测试在独立进程中运行,避免 Docker/K8s 测试污染全局状态

🔹 -n auto:自动使用所有 CPU 核心并行执行

🔹 --cov-branch:分支级别覆盖率(不仅是行覆盖率)

🔹 覆盖率文件通过 actions/upload-artifact 上传,后续合并

2. 企业版测试 (py-tests.yml test-enterprise)

📄 .github/workflows/py-tests.yml (第 90-92 行)

      - name: Run Unit Tests
        run: PYTHONPATH=".:$PYTHONPATH" poetry run --project=enterprise \
              pytest --forked -n auto -s -p no:ddtrace \
              -p no:ddtrace.pytest_bdd -p no:ddtrace.pytest_benchmark \
              ./enterprise/tests/unit --cov=enterprise --cov-branch

注意:企业版测试显式禁用了 ddtrace 插件(Datadog APM),避免 CI 环境中遥测干扰。

3. 覆盖率评论 (coverage-comment job)

📄 .github/workflows/py-tests.yml (第 102-125 行)

  coverage-comment:
    name: Coverage Comment
    if: github.event_name == 'pull_request'
    needs: [test-on-linux, test-enterprise]
    steps:
      - uses: actions/download-artifact@v8
        with:
          pattern: coverage-*
          merge-multiple: true
      - name: Coverage comment
        uses: py-cov-action/python-coverage-comment-action@v3
        with:
          GITHUB_TOKEN: ${{ github.token }}
          MERGE_COVERAGE_FILES: true

Python 后端和企业版覆盖率合并后自动发布到 PR 评论,Reviewer 可以直接看到变更的覆盖率影响。

四、Lint 与代码质量门禁

1. Pre-commit Hooks

📄 dev_config/python/.pre-commit-config.yaml (第 1-67 行)

repos:
  - repo: https://github.com/pre-commit/pre-commit-hooks
    rev: v5.0.0
    hooks:
      - id: trailing-whitespace
      - id: end-of-file-fixer
      - id: check-yaml
      - id: debug-statements

  - repo: local
    hooks:
      - id: warn-appmode-oss
        name: "Warn on AppMode.OSS in backend"
        entry: bash -lc 'if rg -n "\\bAppMode\\.OSS\\b" \
                  openhands tests/unit; then ...; fi'

  - repo: https://github.com/astral-sh/ruff-pre-commit
    rev: v0.12.5
    hooks:
      - id: ruff
        args: [--fix, --unsafe-fixes]
      - id: ruff-format

  - repo: https://github.com/pre-commit/mirrors-mypy
    rev: v1.15.0
    hooks:
      - id: mypy
        entry: mypy --config-file dev_config/python/mypy.ini openhands/

四层质量检查:

🔹 基础检查:trailing-whitespace、end-of-file-fixer、check-yaml、debug-statements

🔹 业务规则:warn-appmode-oss 禁止在 OSS 代码中使用 AppMode.OSS(统一用 AppMode.OPENHANDS)

🔹 Ruff lint + format:替代 flake8 + black + isort,单一工具完成 lint 和格式化

🔹 Mypy 类型检查:对 openhands/ 目录做严格的静态类型检查

2. Ruff 规则配置

📄 pyproject.toml (第 308-322 行)

[tool.ruff]
exclude = [ "enterprise/" ]
format.quote-style = "single"
lint.select = [
  "ASYNC",   # 异步代码最佳实践
  "B",       # flake8-bugbear (常见 bug 检测)
  "E",       # pycodestyle errors
  "F",       # pyflakes
  "I",       # isort (import 排序)
  "Q",       # 引号风格
  "UP006",   # 类型注解现代化
  "UP007",   # Union[X, Y] → X | Y
  "UP008",   # True/False 类型注解
  "W",       # pycodestyle warnings
]
lint.ignore = [
  "ASYNC109", "ASYNC110", "ASYNC220", "ASYNC221",
  "ASYNC230", "ASYNC251", "B003",
]

设计亮点:Ruff 排除了 enterprise/ 目录(企业版有独立的 lint 配置),启用了 ASYNC 规则族确保异步代码的正确性。

五、测试策略深度解析

1. Fixture 模式 — 企业版 conftest.py

📄 enterprise/tests/unit/conftest.py (第 47-57 行)

@pytest.fixture(autouse=True)
def allow_short_context_windows():
    old = os.environ.get('ALLOW_SHORT_CONTEXT_WINDOWS')
    os.environ['ALLOW_SHORT_CONTEXT_WINDOWS'] = 'true'
    try:
        yield
    finally:
        if old is None:
            os.environ.pop('ALLOW_SHORT_CONTEXT_WINDOWS', None)
        else:
            os.environ['ALLOW_SHORT_CONTEXT_WINDOWS'] = old

autouse fixture:每个企业版测试自动设置环境标志,测试结束后恢复原值。这是典型的"环境隔离"模式。

数据库 Fixture:

📄 enterprise/tests/unit/conftest.py (第 80-128 行)

@pytest.fixture(scope='function')
def db_path(tmp_path):
    """Create a unique temp file path for each test."""
    return str(tmp_path / 'test.db')

@pytest.fixture
def engine(db_path):
    engine = create_engine(
        f'sqlite:///{db_path}', connect_args={'check_same_thread': False}
    )
    Base.metadata.create_all(engine)
    return engine

@pytest.fixture
def async_engine(db_path):
    """Create an async engine using the SAME file-based database."""
    async_engine = create_async_engine(
        f'sqlite+aiosqlite:///{db_path}',
        connect_args={'check_same_thread': False},
    )
    # ... 自动建表
    return async_engine

每个测试使用 tmp_path 创建独立的 SQLite 文件,测试间完全隔离,无共享状态。

2. Mock 模式 — 中间件测试

📄 tests/unit/server/test_middleware.py (第 1-45 行)

@pytest.fixture
def app():
    """Create a test FastAPI application."""
    app = FastAPI()
    @app.get('/test')
    def test_endpoint():
        return {'message': 'Test endpoint'}
    return app

def test_localhost_cors_middleware_init_with_config():
    mock_config = MagicMock()
    mock_config.permitted_cors_origins = [
        'https://example.com', 'https://test.com',
    ]
    with patch(
        'openhands.app_server.middleware.get_global_config',
        return_value=mock_config
    ):
        app = FastAPI()
        middleware = LocalhostCORSMiddleware(app)
        assert 'https://example.com' in middleware.allow_origins
        assert len(middleware.allow_origins) == 2

测试模式:通过 patch 替换全局配置,用 FastAPI + TestClient 模拟 HTTP 请求,验证 CORS 中间件行为。

3. 参数化测试 — 环境变量验证

📄 tests/unit/app_server/utils/test_env_var_validation.py (第 12-31 行)

class TestIsValidEnvVarName:
    @pytest.mark.parametrize(
        'name',
        [
            'MY_VAR', 'my_var', 'MyVar', '_PRIVATE',
            '_', '__', 'A', 'a', 'VAR123', '_123',
            'API_KEY', 'DATABASE_URL', 'GITHUB_TOKEN',
        ],
    )
    def test_valid_names(self, name: str):
        assert is_valid_env_var_name(name) is True

    @pytest.mark.parametrize(
        'name',
        [
            'MY-VAR', 'MY VAR', 'MY.VAR', '123VAR',
            '-VAR', 'MY@VAR', 'MY$VAR', 'MY#VAR',
            # ... 30+ invalid cases
        ],
    )
    def test_invalid_names_special_chars(self, name: str):
        assert is_valid_env_var_name(name) is False

参数化测试:用 @pytest.mark.parametrize 将 13 个合法值和 30+ 非法值组合成两组测试,减少代码重复。

4. 异步测试 — GitHub Service

📄 tests/unit/integrations/github/test_github_service.py (第 18-88 行)

@pytest.mark.asyncio
async def test_github_service_token_handling():
    token = SecretStr('test-token')
    service = GitHubService(user_id=None, token=token)
    assert service.token == token
    headers = await service._get_headers()
    assert headers['Authorization'] == 'Bearer test-token'
    assert headers['Accept'] == 'application/vnd.github.v3+json'

@pytest.mark.asyncio
async def test_github_service_fetch_data():
    mock_response = AsyncMock()
    mock_response.status_code = 200
    mock_response.json.return_value = {'login': 'test-user'}
    mock_client = AsyncMock()
    mock_client.get.return_value = mock_response
    mock_client.__aenter__.return_value = mock_client
    with patch('httpx.AsyncClient', return_value=mock_client):
        service = GitHubService(user_id=None,
                               token=SecretStr('test-token'))
        _ = await service._make_request(
            'https://api.github.com/user')
        mock_client.get.assert_called_once()

异步 Mock 模式:用 AsyncMock 模拟 httpx.AsyncClient,用 __aenter__/__aexit__ 模拟异步上下文管理器,完全在内存中验证 API 调用逻辑。

六、前端测试体系

Vitest 单元 + Playwright E2E

📄 .github/workflows/fe-unit-tests.yml (第 44-46 行)

      - name: Run tests and collect coverage
        working-directory: ./frontend
        run: npm run test:coverage

📄 .github/workflows/fe-e2e-tests.yml (第 37-48 行)

      - name: Install Playwright browsers
        working-directory: ./frontend
        run: npx playwright install --with-deps chromium
      - name: Run Playwright tests
        working-directory: ./frontend
        run: npx playwright test --project=chromium
      - name: Upload Playwright report
        if: always()
        with:
          name: playwright-report
          path: frontend/playwright-report/

前端测试双轨制:

🔹 Vitest:299 个测试文件,覆盖组件、工具函数、翻译完整性

🔹 Playwright:真实 Chromium 浏览器 E2E 测试,报告自动上传为 CI 制品

🔹 TypeScript 编译检查:npm run build 作为 CI 门禁,确保类型安全

七、Docker 构建与发布流水线

多架构镜像构建

📄 .github/workflows/_build-image.yml (第 47-95 行)

  build:
    strategy:
      matrix:
        arch: [amd64, arm64]
    steps:
      - name: Build and push
        uses: docker/build-push-action@v7
        with:
          platforms: linux/${{ matrix.arch }}
          cache-from: |
            type=registry,ref=${{ inputs.image }}:buildcache-*
          cache-to: type=registry,...,mode=max
          provenance: ${{ inputs.provenance }}
          sbom: ${{ inputs.sbom }}

  merge:
    needs: build
    steps:
      - name: Merge manifests
        with:
          base-tags: ${{ steps.meta_base.outputs.tags }}
          archs: "amd64 arm64"

构建策略:

🔹 并行构建:amd64 和 arm64 同时构建,分别推送到 GHCR

🔹 Registry 缓存:buildcache 标签实现跨 CI 运行的构建缓存

🔹 Manifest 合并:构建完成后合并为多架构镜像

🔹 Enterprise 额外安全:provenance + SBOM 签名(仅企业版镜像)

Release-please 自动化版本管理

📄 .github/workflows/release.yml (第 15-31 行)

jobs:
  openhands-app:
    uses: OpenHands/release-actions/.github/workflows/
          release-please.yml@main
  cloud:
    uses: OpenHands/release-actions/.github/workflows/
          release-please.yml@main
    with:
      config-file: release-please-config.cloud.json
      manifest-file: .release-please-manifest.cloud.json

双发布线:OpenHands App (X.Y.Z) 和 OpenHands Cloud (cloud-X.Y.Z) 独立版本管理,基于 conventional commit 自动计算版本号。

八、测试目录结构总览

tests/
├── unit/                          # 114 个 Python 测试
│   ├── app_server/                # App Server 测试
│   │   ├── utils/                 #   工具函数
│   │   │   ├── logger/            #     日志系统
│   │   │   ├── test_jsonpatch_compat.py
│   │   │   └── test_env_var_validation.py
│   │   ├── analytics/             #   分析模块
│   │   └── file_store/            #   文件存储
│   ├── server/                    # 服务端测试
│   │   ├── routes/                #   API 路由
│   │   │   ├── test_skills_api.py
│   │   │   └── test_mcp_routes.py
│   │   ├── test_middleware.py
│   │   └── test_openapi_schema_generation.py
│   ├── integrations/              # 集成测试
│   │   ├── github/                #   GitHub 集成
│   │   ├── gitlab/                #   GitLab 集成
│   │   ├── bitbucket/             #   Bitbucket 集成
│   │   ├── bitbucket_data_center/ #   BBD C 集成
│   │   └── protocols/             #   协议层
│   ├── storage/                   # 存储测试
│   │   ├── settings/              #   设置存储
│   │   └── data_models/           #   数据模型
│   ├── utils/                     # 工具测试
│   │   ├── test_llm_utils.py
│   │   ├── test_git.py
│   │   └── ... (8 个文件)
│   ├── mcp/                       # MCP 集成测试
│   ├── frontend/                  # 前端翻译测试
│   └── enterprise/                # 企业迁移测试
└── (无 conftest.py — 测试完全自包含)

enterprise/tests/
└── unit/                          # 178 个企业版测试
    ├── conftest.py                # DB Fixture + autouse 环境
    ├── server/routes/conftest.py  # 路由专用 Fixture
    └── integrations/              # 企业集成 (Jira, Jira DC)

九、CI/CD 流水线全景图

PR 推送 ──────────────────────────────────────────────┐
                                                      │
    ├── Lint (lint.yml)                               │
    │   ├── lint-frontend  (ESLint + TSC + i18n)      │
    │   ├── lint-python    (pre-commit + ruff + mypy) │
    │   └── lint-enterprise (独立 pre-commit)          │
    │                                                │
    ├── Python Tests (py-tests.yml)                   │
    │   ├── test-on-linux  (--forked -n auto --cov)   │
    │   ├── test-enterprise (--forked -n auto --cov)  │
    │   └── coverage-comment (合并 + PR 评论)         │
    │                                                │
    ├── FE Unit Tests (fe-unit-tests.yml)             │
    │   └── vitest + coverage                         │
    │                                                │
    ├── FE E2E Tests (fe-e2e-tests.yml)               │
    │   └── Playwright Chromium                       │
    │                                                │
    ├── Docker Build (ghcr-build.yml)                  │
    │   ├── build_app  (amd64 + arm64)                │
    │   ├── build_enterprise (provenance + SBOM)      │
    │   └── update_pr_description                     │
    │                                                │
    └── PR Title Check (pr.yml)                       │
        └── Conventional Commits 校验                  │
                                                      │
Main 推送 ────────────────────────────────────────────┤
    ├── Release-please (release.yml)                   │
    │   ├── openhands-app  (X.Y.Z)                     │
    │   └── cloud        (cloud-X.Y.Z)                 │
    ├── Docker 自动构建 (同上)                          │
    └── PyPI Release (pypi-release.yml)                │
                                                      │
Fork PR 保护 ─────────────────────────────────────────┤
    ├── Docker 构建跳过 fork                           │
    ├── pr.yml 使用 pull_request_target (fork-safe)    │
    └── 无 secrets: inherit                            │

十、关键设计总结

设计决策实现方式效果
进程隔离--forked + tmp_path DBDocker/K8s 测试不污染全局
并行加速-n auto xdist114+178 个测试分钟级完成
覆盖率门禁cov-branch + PR 评论Reviewer 实时看到覆盖率变化
类型安全mypy + TypeScript前后端全栈类型检查
统一 LintRuff 替代 flake8+black+isort单工具快 10x,配置集中
构建缓存Registry buildcache跨 PR 复用 Docker 层
Fork 安全pull_request_target + 条件跳过Fork PR 不触发 Docker 构建
自动版本Release-please + poetry-dynamic-versioningConventional commit 驱动版本

📌 预告

下一讲深入 性能优化与缓存策略——OpenHands 如何在 Agent 对话、LLM 调用、文件存储等环节做性能优化,包括 LiteLLM 路由缓存、SQLite 查询优化、以及 Docker 构建缓存策略。

📚 系列导航

← 第 15 讲:错误处理与日志系统

→ 第 17 讲:性能优化与缓存策略

关注公众号「AI技术推荐官」获取更多源码解析内容

回首页,搜索你想要的文档

在首页搜索框输入关键词,快速定位资料

文库首页搜索指引 立即搜索

相关学习资料