CI CD

是什么

这是一份 CI/CD（持续集成与持续交付）流水线规范，覆盖 GitHub Actions 的触发、缓存、矩阵测试、质量门禁、跨环境晋升等关键节点，让团队从手工部署升级到一键发版，且每次发版都自带质量保障。

怎么用

1. 新项目立项时，按本规范拷贝标准 workflow 模板，5 分钟内就能跑通最小可用的 CI（持续集成）流程。
2. 给流水线加单元测试、覆盖率、安全扫描等质量门禁时，对照文档中的标准门禁清单逐项启用。
3. 配置云上部署 OIDC（开放身份连接）认证时，遵循免长期密钥章节的步骤，避免把 access key 写进 secrets。
4. 流水线变慢时，按缓存与并行小节先加 cache，再做 job 拆分，目标 P95 时长控制在 8 分钟内。
5. 从测试环境晋升到生产环境时，必须走文档约定的多环境晋升流程，加 manual approval（人工审批）兜底。

架构图

flowchart LR
    A[Push 或 PR] --> B[Lint 与单测]
    B --> C[构建产物]
    C --> D[质量门禁]
    D --> E[测试环境部署]
    E --> F[人工审批晋升]

CI/CD with GitHub Actions

A complete reference for building robust, secure, and efficient CI/CD pipelines using GitHub Actions, covering everything from workflow anatomy to multi-environment deployment promotion.

When to Activate

• Creating or modifying a GitHub Actions workflow
• Adding a quality gate (coverage, lint, security scan) to a pipeline
• Setting up CD with environment promotion from staging to production
• Optimizing a slow CI pipeline with caching or parallelism
• Publishing a Docker image, Python package, or npm package from CI
• Configuring secrets and OIDC-based cloud authentication

Workflow Anatomy

Every GitHub Actions workflow is defined in .github/workflows/*.yml. Understanding the core building blocks is essential before composing pipelines.

Triggers

Jobs

• Jobs run in parallel by default unless needs: is specified.
• Use needs: [job-a, job-b] to declare dependencies; a job waits for all listed jobs to succeed.
• Each job runs in a fresh virtual machine (or container).

Steps

• uses: — invokes a reusable action from the marketplace or a local path.
• run: — executes a shell command directly on the runner.
• Steps within a job share the same filesystem and runner environment.

Key Contexts

Expression Syntax

All dynamic values use ${{ expression }} syntax:

if: github.ref == 'refs/heads/main'
run: echo "SHA is ${{ github.sha }}"
env:
  BRANCH: ${{ github.ref_name }}

Minimal Workflow Skeleton

name: CI
on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Run tests
        run: make test

Standard Pipeline Structure

The canonical stage sequence moves from fast feedback (lint) to slower gates (security), then publishing.

lint → test → build → security-scan → publish

Job Dependency Graph

jobs:
  lint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Lint
        run: make lint

  test:
    needs: lint
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Test
        run: make test

  build:
    needs: test
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Build
        run: make build

  security-scan:
    needs: build
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Security scan
        run: make scan

  publish:
    needs: [build, security-scan]
    if: github.ref == 'refs/heads/main'
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Publish
        run: make publish

fail-fast Behavior

Use fail-fast: false in matrix strategies so you see failures across all OS/version combinations rather than stopping at the first one.

Dependency Caching

Caching package manager dependencies is the single highest-impact optimization for CI speed. The key strategy: hash the lockfile so the cache automatically invalidates when dependencies change.

Cache Key Strategy

cache-key = runner-os + lockfile-hash
restore-key = runner-os (fallback to most recent cache)

Python

- uses: actions/setup-python@v5
  with:
    python-version: '3.12'
    cache: 'pip'  # built-in caching, hashes requirements*.txt

For more control with pyproject.toml:

- uses: actions/cache@v4
  with:
    path: ~/.cache/pip
    key: ${{ runner.os }}-pip-${{ hashFiles('**/pyproject.toml', '**/requirements*.txt') }}
    restore-keys: |
      ${{ runner.os }}-pip-

Node.js

- uses: actions/setup-node@v4
  with:
    node-version: '20'
    cache: 'npm'  # hashes package-lock.json automatically

For pnpm or yarn, set cache: 'pnpm' or cache: 'yarn' respectively.

Go

- uses: actions/setup-go@v5
  with:
    go-version: '1.22'
    cache: true  # caches $GOPATH/pkg/mod and build cache

Cache Hit Rate Tips

• Always commit lockfiles (package-lock.json, poetry.lock, go.sum) to the repository.
• Use hashFiles('**/package-lock.json') to scope cache keys to the exact dependency set.
• Restore keys act as fallbacks: a partial hit is better than no cache.

Matrix Builds

Matrix builds let a single job definition run across multiple configurations (language versions, operating systems) in parallel.

Matrix Strategy

jobs:
  test:
    runs-on: ${{ matrix.os }}
    strategy:
      fail-fast: false
      matrix:
        python-version: ['3.11', '3.12', '3.13']
        os: [ubuntu-latest, windows-latest]
        include:
          - python-version: '3.12'
            os: ubuntu-latest
            publish: true      # custom flag for conditional steps
        exclude:
          - python-version: '3.11'
            os: windows-latest  # skip known unsupported combo
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: ${{ matrix.python-version }}
          cache: pip
      - run: pytest
      - name: Publish (only from designated matrix entry)
        if: matrix.publish == true
        run: make publish

Matrix Reference

Quality Gates

Coverage Threshold

Fail the build if coverage drops below the required threshold.

- name: Run tests with coverage
  run: pytest --cov=src --cov-report=xml --cov-fail-under=80

- name: Upload coverage report
  uses: codecov/codecov-action@v4
  with:
    file: ./coverage.xml
    fail_ci_if_error: true

For Node.js with Jest:

- name: Run tests with coverage
  run: jest --coverage --coverageThreshold='{"global":{"lines":80}}'

Inline Lint Annotations with reviewdog

reviewdog posts lint results as PR review comments, making issues visible without reading raw logs.

- uses: reviewdog/action-flake8@v3
  with:
    reporter: github-pr-review
    level: warning

- uses: reviewdog/action-eslint@v1
  with:
    reporter: github-pr-review
    eslint_flags: '--ext .js,.ts src/'

Branch Protection Rules

Configure these in GitHub Settings → Branches, not in workflow YAML:

• Required status checks: list every CI job name that must pass before merge.
• Required approvals: minimum 1 reviewer for main.
• Dismiss stale approvals: re-approval required after new commits are pushed.
• Require branches to be up to date: prevents merging outdated branches.

Security Scanning

- name: Python dependency audit
  run: pip-audit

- name: Node audit
  run: npm audit --audit-level=high

- name: Go vulnerability check
  run: govulncheck ./...

For container scanning, integrate Trivy:

- uses: aquasecurity/trivy-action@master
  with:
    image-ref: ghcr.io/${{ github.repository }}:${{ github.sha }}
    severity: 'CRITICAL,HIGH'
    exit-code: '1'

Secrets and Environment Variables

Contexts

env:
  APP_ENV: production                    # plain env var (visible in logs)
  DB_HOST: ${{ vars.DB_HOST }}           # repository variable (not secret, visible)
  API_KEY: ${{ secrets.API_KEY }}        # encrypted secret (masked in logs)

Variable Scoping

$GITHUB_TOKEN Permissions

Declare only what the job requires (principle of least privilege):

permissions:
  contents: read       # read repo files
  packages: write      # push to GHCR
  id-token: write      # request OIDC token for cloud auth
  pull-requests: write # post PR comments
  checks: write        # create check runs

Set permissions: {} at the workflow level, then grant specific permissions per job to minimize blast radius.

OIDC for Cloud Auth (No Long-Lived Credentials)

OIDC eliminates the need to store cloud credentials as GitHub secrets. The cloud provider verifies the workflow's identity token and grants temporary credentials.

AWS:

- uses: aws-actions/configure-aws-credentials@v4
  with:
    role-to-assume: arn:aws:iam::123456789:role/github-actions
    aws-region: us-east-1

Requires an IAM role with a trust policy that allows token.actions.githubusercontent.com as an OIDC provider.

Google Cloud:

- uses: google-github-actions/auth@v2
  with:
    workload_identity_provider: projects/123/locations/global/workloadIdentityPools/my-pool/providers/github
    service_account: deployer@my-project.iam.gserviceaccount.com

Requires a Workload Identity Pool configured in GCP with a GitHub OIDC provider binding.

Azure:

- uses: azure/login@v2
  with:
    client-id: ${{ secrets.AZURE_CLIENT_ID }}
    tenant-id: ${{ secrets.AZURE_TENANT_ID }}
    subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}

Uses federated credentials on the Azure App Registration — no client secret stored.

CD and Environment Promotion

The standard promotion pattern: build once, deploy to staging automatically, then gate production behind a manual approval.

Staging → Production Pipeline

jobs:
  build:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Build artifact
        run: make build
      - uses: actions/upload-artifact@v4
        with:
          name: app-artifact
          path: dist/

  deploy-staging:
    environment: staging
    runs-on: ubuntu-latest
    needs: build
    if: github.ref == 'refs/heads/main'
    steps:
      - uses: actions/download-artifact@v4
        with:
          name: app-artifact
          path: dist/
      - name: Deploy to staging
        run: ./scripts/deploy.sh staging

  smoke-test:
    runs-on: ubuntu-latest
    needs: deploy-staging
    steps:
      - name: Run smoke tests against staging
        run: ./scripts/smoke-test.sh https://staging.example.com

  deploy-production:
    environment: production   # required reviewers configured in GitHub Settings
    runs-on: ubuntu-latest
    needs: smoke-test
    steps:
      - uses: actions/download-artifact@v4
        with:
          name: app-artifact
          path: dist/
      - name: Deploy to production
        run: ./scripts/deploy.sh production
      - name: Notify on failure
        if: failure()
        run: ./scripts/notify-failure.sh "${{ job.status }}"

Environment Protection Configuration

Set these in GitHub Settings → Environments, not in YAML:

Deployment Status

Use job.status in post-deploy notifications:

- name: Notify Slack
  if: always()
  uses: slackapi/slack-github-action@v1
  with:
    payload: '{"text": "Deploy to production: ${{ job.status }}"}'
  env:
    SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK }}

Artifact and Package Publishing

Docker Image to GHCR

jobs:
  publish-image:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      packages: write
    steps:
      - uses: actions/checkout@v4

      - uses: docker/setup-buildx-action@v3

      - uses: docker/login-action@v3
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}

      - uses: docker/metadata-action@v5
        id: meta
        with:
          images: ghcr.io/${{ github.repository }}
          tags: |
            type=sha,prefix=,suffix=,format=short
            type=ref,event=branch
            type=semver,pattern={{version}}

      - uses: docker/build-push-action@v5
        with:
          context: .
          push: ${{ github.ref == 'refs/heads/main' }}
          tags: ${{ steps.meta.outputs.tags }}
          labels: ${{ steps.meta.outputs.labels }}
          cache-from: type=gha
          cache-to: type=gha,mode=max

Always tag images with the commit SHA. Never use latest as the only tag in production.

PyPI with Trusted Publishing (No Token Needed)

Configure a Trusted Publisher in PyPI project settings pointing to this repository and workflow, then:

jobs:
  publish-pypi:
    runs-on: ubuntu-latest
    permissions:
      id-token: write  # required for trusted publishing
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: '3.12'
      - run: pip install build
      - run: python -m build
      - uses: pypa/gh-action-pypi-publish@release/v1
        with:
          repository-url: https://upload.pypi.org/legacy/
          # For pre-release testing:
          # repository-url: https://test.pypi.org/legacy/

npm Publish

jobs:
  publish-npm:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
          registry-url: 'https://registry.npmjs.org'
      - run: npm ci
      - run: npm run build
      - run: npm publish --access public
        env:
          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}

Complete Workflow Examples

Python Service

name: Python CI
on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  ci:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: '3.12'
          cache: pip
      - run: pip install -e ".[dev]"
      - run: ruff check .
      - run: mypy src/
      - run: pytest --cov=src --cov-report=xml --cov-fail-under=80
      - run: pip-audit
      - name: Upload coverage
        uses: codecov/codecov-action@v4
        with:
          file: ./coverage.xml

TypeScript/Node Service

name: Node CI
on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  ci:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
          cache: npm
      - run: npm ci
      - run: npm run lint
      - run: npm run typecheck
      - run: npm test -- --coverage
      - run: npm audit --audit-level=high

Go Service

name: Go CI
on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  ci:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-go@v5
        with:
          go-version: '1.22'
          cache: true
      - run: go vet ./...
      - run: staticcheck ./...
      - run: go test -race -coverprofile=coverage.out ./...
      - run: go tool cover -func=coverage.out
      - run: govulncheck ./...

Decision Tables

Which Cache Approach to Use

When to Use OIDC vs Stored Secrets

Trigger Selection Guide

Red Flags

• Storing cloud credentials as GitHub Secrets — long-lived access keys can be exfiltrated via PR log injection; use OIDC federated credentials (aws-actions/configure-aws-credentials) instead
• Hardcoding runs-on: ubuntu-latest without version pinning — ubuntu-latest can shift to a new OS major version mid-project, silently changing available toolchains; pin to ubuntu-22.04 for stability
• Single monolithic job with 20+ steps — any step failure retries the entire job from scratch; split into separate jobs with needs: so fast gates (lint) don't block slow ones (security scan)
• Pushing to main without a required status check — branch protection "Required status checks" must be configured in GitHub Settings, not just in the YAML; YAML alone can be bypassed
• Cache key without a lockfile hash — using key: ${{ runner.os }}-pip without hashFiles(...) causes stale caches after dependency updates, producing false green builds
• Tagging Docker images with only latest — latest is overwritten on every build; a failed rollback cannot target a specific previous image; always add a commit SHA tag alongside latest
• Using pull_request_target without careful filtering — this trigger runs in the context of the base branch and has access to secrets, making it exploitable by a forked PR that modifies the workflow
• Skipping --cov-fail-under or equivalent threshold — coverage upload without a fail threshold lets coverage silently drop to zero; the gate must actively fail the build

Checklist

• Workflow triggers cover both push to main and pull_request
• Jobs run in correct dependency order using needs:
• Package manager caching configured (pip/npm/go modules)
• Lint, type-check, test, and security scan are all separate steps
• Coverage threshold enforced (--cov-fail-under / --coverage)
• $GITHUB_TOKEN permissions scoped to least privilege
• Secrets accessed via ${{ secrets.X }} — never hardcoded
• OIDC used for cloud credentials — no long-lived access keys stored as secrets
• CD deploys to staging first; production requires approval via environment protection
• Docker images tagged with commit SHA, not latest
• Failed jobs produce clear error output; no silent failures
• Pipeline completes in under 10 minutes on a typical PR