[@aws-cdk/pipelines] Source (branch) mapping to account/stage

❓ General Issue

The Question

Is it possible, using the new CDKPipeline (preview) library @aws-cdk/pipelines to have a specify a different source action for each stage/account? This could be a usage failing, but my scenario looks like this:

We are using GitFlow- so we have develop, release and master branches. Each one of these branches should deploys to a different AWS account. Let’s assume that develop is a develop account, release is a release account and master is a master account.

In this scenario, if I use a GitHub action for my source, I don’t appear to have a way to filter the GitHub action’s branch to which stage (account) this should deploy to?

Our pipeline process is essentially the same in each of these branches - albeit with a few bits turned on or off at each stage (I know that the Context lookup isn’t yet available, but we should be able to map those settings appropriately when it is).

We have a build, deploy and test action in our pipeline, and that pipeline is currently created 3 times (3 pipeline stacks deployed separately) for each branch we want to map to an AWS Account… so it feels like this should be something I use CdkPipelines for.

Any recommended usage patterns, questions, or documentation would be helpful if I’m going about this the wrong way - equally I thought as this is in Preview and this (may) be a valuable feature I should raise it.

Environment

• CDK CLI Version: 1.56.0
• Module Version: 1.56.0
• Node.js Version: 12.4.0
• OS: Ubuntu (WSL)
• Language (Version): TypeScript (3.9.7)

Other information

If this is within the scope of CdkPipelines, An example of what I might expect this to look like:

    const devStage = pipeline.addApplicationStage(new MyApplication(this, 'Dev', {
      env: {
        account: '123456789012',
        region: 'eu-west-1',
      },
      // sourceAction is a new paramter we can pass to the applicationStage? 
      sourceAction: new GitHubSourceAction({ 
         repo: "somerepo",
         actionName: "source",
         oauthToken: new SecretValue(""),
         owner: "someowner",
         branch: "develop",
         output: devArtifact
       });
    }));

[MarcoAlessandroRiggio]

A single pipeline is just not able to express that logic, so you will have to resort to multiple pipelines. Fortunately, since this is code you should be able to make 3 similar pipelines easily enough with a for loop.

Hi @rix0rrr, we’ve tried to create two CdkPipeline in a for loop like you suggested but in UpdatePipeline Stage one of the two pipelines always fails with the following error: Call failed: describeChangeSet({"StackName":"<pipeline-stack-id>","ChangeSetName":"<change-set-name>"}) => ChangeSet [<change-set-name>] does not exist

 ["development", "production"].forEach(environment => {
            process.env.NODE_ENV = environment;
            delete require.cache[require.resolve("config")];
            const config = require('config');  // reload module for change configuration file based on environment 

            new CdkPipeline(this, `${environment}-cloud-pipeline`, {
                pipelineName: `${environment}-cloud-pipeline`,
                cloudAssemblyArtifact,
                sourceAction: new codepipeline_actions.CodeCommitSourceAction({
                    actionName: 'CodeCommit_Source',
                    repository: Repository.fromRepositoryName(this, `${environment}-repository`, config.repositoryName),
                    output: sourceArtifact,
                    branch: config.branch
                }),
                synthAction: SimpleSynthAction.standardNpmSynth({
                    sourceArtifact,
                    cloudAssemblyArtifact,
                    buildCommand: 'npm run build'
                })
            }).addApplicationStage(new CdkInfrastructureStage(this, environment, environment, {
                env: {account: config.account, region: config.region}
            }));
        });

How should it work? Do you have any suggestions?

Thanks Best Regards

[BrianFarnhill]

@AlanChauchet Yep, the codes on an private repo but here is the class that sets up the two pipelines:

const sourceArtifact = new codepipeline.Artifact();
const cloudAssemblyArtifact = new codepipeline.Artifact();

const synthParams = {
    buildCommand: 'npm run build --prefix ../../ && npm run test --prefix ../../',
    installCommand: 'npm run bootstrap --prefix ../../',
    subdirectory: 'packages/infrastructure',
}

function CreatePipeline(stack: cdk.Stack, sourceRepo: codecommit.IRepository, branchName: string) {

    const pipeline = new pipelines.CdkPipeline(stack, `${branchName}Pipeline`, {
        cloudAssemblyArtifact,
        sourceAction: new actions.CodeCommitSourceAction({
            actionName: 'CodeCommit',
            output: sourceArtifact,
            repository: sourceRepo,
            branch: branchName
        }),
        synthAction: pipelines.SimpleSynthAction.standardNpmSynth({
            ...synthParams, sourceArtifact, cloudAssemblyArtifact
        })
    })

    const stageName = (branchName === 'master' ? 'prod' : 'dev')
    const accountId = (branchName === 'master' ? 'ACCOUNTID-PROD' : 'ACCOUNTID-DEV')

    pipeline.addApplicationStage(new CloudshotStage(stack, stageName, {
        env: { account: accountId, region: 'ap-southeast-2' }
    }))

    return pipeline
}

export class MyPipelines extends cdk.Stack {

    constructor(scope: cdk.Construct, id: string, props?: cdk.StackProps) {
        super(scope, id, props)
        
        const sourceRepo = codecommit.Repository.fromRepositoryName(this, "SourceRepo", 'MyRepoName')

        CreatePipeline(this, sourceRepo, 'dev')
        CreatePipeline(this, sourceRepo, 'master')
    }

}

Tweak the approach to putting in your branch names or account IDs to suit however you want to do that, but that’s the basic premise to how it works.