如何在Monorepo中基于CDK v2正确构建多CodePipeline流水线？

当前做法属于不良实践 #

• 触发逻辑冗余：任意代码变更都会同时触发两条流水线，无论变更是否与流水线负责的模块相关，严重浪费计算资源和部署时间
• 扩展性差：后续新增服务时，流水线数量会同步增加，触发冗余问题会愈发严重，维护成本直线上升

优化方案 #

方案1：合并为单条自更新流水线，添加变更检测 #

将两条流水线合并为一条主流水线，在Synth步骤前新增变更检测逻辑，仅在对应模块代码变更时执行相关部署：

1. 用git diff检测变更路径，标记需要部署的模块
2. 流水线根据标记动态添加部署阶段

示例代码片段：

import { CodePipeline, ShellStep, CodePipelineSource } from "aws-cdk-lib/pipelines";
import { Stage, Construct } from "aws-cdk-lib";

// 自定义后端部署阶段
class BackendDeploymentStage extends Stage {
  constructor(scope: Construct, id: string) {
    super(scope, id);
    new ApiStack(this, "ApiStack");
    new AuthStack(this, "AuthStack");
    new DataStack(this, "DataStack");
    new LambdaStack(this, "LambdaStack");
  }
}

// 自定义前端部署阶段
class WebAppDeploymentStage extends Stage {
  constructor(scope: Construct, id: string) {
    super(scope, id);
    // 这里添加WebApp的部署逻辑
  }
}

// 变更检测步骤
const detectChanges = new ShellStep("DetectChanges", {
  input: CodePipelineSource.gitHub("your-org/my-monorepo", "main"),
  commands: [
    'git diff --name-only HEAD^ HEAD | grep -E "(packages/lambda|packages/lib|packages/infra/stacks/(ApiStack|AuthStack|DataStack|LambdaStack|BackendPipelineStack))" && echo "DEPLOY_BACKEND=true" >> $ENV_FILE || echo "DEPLOY_BACKEND=false" >> $ENV_FILE',
    'git diff --name-only HEAD^ HEAD | grep -E "(packages/web|packages/infra/stacks/WebAppPipelineStack)" && echo "DEPLOY_WEB=true" >> $ENV_FILE || echo "DEPLOY_WEB=false" >> $ENV_FILE'
  ],
  envFile: EnvironmentFile.fromPath("./deploy-vars")
});

// 主流水线
const pipeline = new CodePipeline(this, "MonorepoMainPipeline", {
  pipelineName: "MonorepoMainPipeline",
  publishAssetsInParallel: true,
  synth: new ShellStep("Synth", {
    input: detectChanges.outputs[0],
    primaryOutputDirectory: "packages/infra/cdk.out",
    commands: [
      "cd packages/infra",
      "npm ci",
      "npx cdk synth"
    ]
  })
});

// 根据检测结果添加部署阶段
const deployBackend = process.env.DEPLOY_BACKEND === "true";
const deployWeb = process.env.DEPLOY_WEB === "true";

if (deployBackend) {
  pipeline.addStage(new BackendDeploymentStage(this, "BackendDeploy"));
}

if (deployWeb) {
  pipeline.addStage(new WebAppDeploymentStage(this, "WebDeploy"));
}

方案2：为每条流水线配置精细化触发条件 #

保留两条独立流水线，但修改触发源的路径过滤规则，仅在对应模块代码变更时触发流水线：

• 后端流水线仅监听后端业务代码（packages/lambda、packages/lib）和后端基础设施代码（packages/infra/stacks/下的后端相关栈文件）的变更
• 前端流水线仅监听前端代码（packages/web）和前端基础设施代码（WebAppPipelineStack.ts）的变更

示例代码片段（GitHub触发器为例）：

// 后端流水线的源配置
const backendSource = CodePipelineSource.gitHub("your-org/my-monorepo", "main", {
  trigger: GitHubTrigger.WEBHOOK,
  filters: [
    FilterGroup.inclusion().filePath("packages/lambda/**/*"),
    FilterGroup.inclusion().filePath("packages/lib/**/*"),
    FilterGroup.inclusion().filePath("packages/infra/stacks/ApiStack.ts"),
    FilterGroup.inclusion().filePath("packages/infra/stacks/AuthStack.ts"),
    FilterGroup.inclusion().filePath("packages/infra/stacks/DataStack.ts"),
    FilterGroup.inclusion().filePath("packages/infra/stacks/LambdaStack.ts"),
    FilterGroup.inclusion().filePath("packages/infra/stacks/BackendPipelineStack.ts")
  ]
});

// 前端流水线的源配置
const webSource = CodePipelineSource.gitHub("your-org/my-monorepo", "main", {
  trigger: GitHubTrigger.WEBHOOK,
  filters: [
    FilterGroup.inclusion().filePath("packages/web/**/*"),
    FilterGroup.inclusion().filePath("packages/infra/stacks/WebAppPipelineStack.ts")
  ]
});

方案3：结合Monorepo工具链优化构建 #

如果使用Nx、TurboRepo等Monorepo工具，可利用它们的增量构建和变更影响分析能力进一步优化：

• 在CDK的Synth步骤中调用nx affected:deploy或turbo run deploy --filter=...，仅构建和部署受变更影响的模块
• 搭配方案1或2，最大化减少不必要的操作

内容的提问来源于stack exchange，提问作者Tsar Bomba