A construct library for painless Continuous Delivery of CDK applications.
This module contains two sets of APIs: an original and a modern version of CDK Pipelines. The modern API has been updated to be easier to work with and customize, and will be the preferred API going forward. The original version of the API is still available for backwards compatibility, but we recommend migrating to the new version if possible.
Compared to the original API, the modern API: has more sensible defaults; is more flexible; supports parallel deployments; supports multiple synth inputs; allows more control of CodeBuild project generation; supports deployment engines other than CodePipeline.
The README for the original API, as well as a migration guide, can be found in our GitHub repository.
Deploying your application continuously starts by defining a
MyApplicationStage
, a subclass of Stage
that contains the stacks that make
up a single copy of your application.
You then define a Pipeline
, instantiate as many instances of
MyApplicationStage
as you want for your test and production environments, with
different parameters for each, and calling pipeline.addStage()
for each of
them. You can deploy to the same account and Region, or to a different one,
with the same amount of code. The CDK Pipelines library takes care of the
details.
CDK Pipelines supports multiple deployment engines (see below), and comes with
a deployment engine that deployes CDK apps using AWS CodePipeline. To use the
CodePipeline engine, define a CodePipeline
construct. The following
example creates a CodePipeline that deploys an application from GitHub:
/** The stacks for our app are defined in my-stacks.ts. The internals of these
* stacks aren't important, except that DatabaseStack exposes an attribute
* "table" for a database table it defines, and ComputeStack accepts a reference
* to this table in its properties.
*/
import { DatabaseStack, ComputeStack } from '../lib/my-stacks';
import { Construct, Stage, Stack, StackProps, StageProps } from '@aws-cdk/core';
import { CodePipeline, CodePipelineSource, ShellStep } from '@aws-cdk/pipelines';
/**
* Stack to hold the pipeline
*/
class MyPipelineStack extends Stack {
constructor(scope: Construct, id: string, props?: StackProps) {
super(scope, id, props);
const pipeline = new CodePipeline(this, 'Pipeline', {
synth: new ShellStep('Synth', {
// Use a connection created using the AWS console to authenticate to GitHub
// Other sources are available.
input: CodePipelineSource.connection('my-org/my-app', 'main', {
connectionArn: 'arn:aws:codestar-connections:us-east-1:222222222222:connection/7d2469ff-514a-4e4f-9003-5ca4a43cdc41', // Created using the AWS console * });',
}),
commands: [
'npm ci',
'npm run build',
'npx cdk synth',
],
}),
});
// 'MyApplication' is defined below. Call `addStage` as many times as
// necessary with any account and region (may be different from the
// pipeline's).
pipeline.addStage(new MyApplication(this, 'Prod', {
env: {
account: '123456789012',
region: 'eu-west-1',
}
}));
}
}
/**
* Your application
*
* May consist of one or more Stacks (here, two)
*
* By declaring our DatabaseStack and our ComputeStack inside a Stage,
* we make sure they are deployed together, or not at all.
*/
class MyApplication extends Stage {
constructor(scope: Construct, id: string, props?: StageProps) {
super(scope, id, props);
const dbStack = new DatabaseStack(this, 'Database');
new ComputeStack(this, 'Compute', {
table: dbStack.table,
});
}
}
// In your main file
new MyPipelineStack(app, 'PipelineStack', {
env: {
account: '123456789012',
region: 'eu-west-1',
}
});
The pipeline is self-mutating, which means that if you add new
application stages in the source code, or new stacks to MyApplication
, the
pipeline will automatically reconfigure itself to deploy those new stages and
stacks.
(Note that have to bootstrap all environments before the above code will work, see the section CDK Environment Bootstrapping below).
This library uses prerelease features of the CDK framework, which can be enabled
by adding the following to cdk.json
:
{
// ...
"context": {
"@aws-cdk/core:newStyleStackSynthesis": true
}
}
To provision the pipeline you have defined, making sure the target environment
has been bootstrapped (see below), and then executing deploying the
PipelineStack
once. Afterwards, the pipeline will keep itself up-to-date.
Important: be sure to
git commit
andgit push
before deploying the Pipeline stack usingcdk deploy
!The reason is that the pipeline will start deploying and self-mutating right away based on the sources in the repository, so the sources it finds in there should be the ones you want it to find.
Run the following commands to get the pipeline going:
$ git commit -a
$ git push
$ cdk deploy PipelineStack
Administrative permissions to the account are only necessary up until this point. We recommend you shed access to these credentials after doing this.
The self-mutation feature of the Pipeline might at times get in the way
of the pipeline development workflow. Each change to the pipeline must be pushed
to git, otherwise, after the pipeline was updated using cdk deploy
, it will
automatically revert to the state found in git.
To make the development more convenient, the self-mutation feature can be turned
off temporarily, by passing selfMutation: false
property, example:
// Modern API
const pipeline = new CodePipeline(this, 'Pipeline', {
selfMutation: false,
...
});
// Original API
const pipeline = new CdkPipeline(this, 'Pipeline', {
selfMutating: false,
...
});
This section of the documentation describes the AWS CodePipeline engine, which comes with this library. If you want to use a different deployment engine, read the section Using a different deployment engine below.
To define a pipeline, instantiate a CodePipeline
construct from the
@aws-cdk/pipelines
module. It takes one argument, a synth
step, which is
expected to produce the CDK Cloud Assembly as its single output (the contents of
the cdk.out
directory after running cdk synth
). "Steps" are arbitrary
actions in the pipeline, typically used to run scripts or commands.
For the synth, use a ShellStep
and specify the commands necessary to install
dependencies, the CDK CLI, build your project and run cdk synth
; the specific
commands required will depend on the programming language you are using. For a
typical NPM-based project, the synth will look like this:
const source = /* the repository source */;
const pipeline = new CodePipeline(this, 'Pipeline', {
synth: new ShellStep('Synth', {
input: source,
commands: [
'npm ci',
'npm run build',
'npx cdk synth',
],
}),
});
The pipeline assumes that your ShellStep
will produce a cdk.out
directory in the root, containing the CDK cloud assembly. If your
CDK project lives in a subdirectory, be sure to adjust the
primaryOutputDirectory
to match:
const pipeline = new CodePipeline(this, 'Pipeline', {
synth: new ShellStep('Synth', {
input: source,
commands: [
'cd mysubdir',
'npm ci',
'npm run build',
'npx cdk synth',
],
primaryOutputDirectory: 'mysubdir/cdk.out',
}),
});
The underlying @aws-cdk/aws-codepipeline.Pipeline
construct will be produced
when app.synth()
is called. You can also force it to be produced
earlier by calling pipeline.buildPipeline()
. After you've called
that method, you can inspect the constructs that were produced by
accessing the properties of the pipeline
object.
The commands you pass to new ShellStep
will be very similar to the commands
you run on your own workstation to install dependencies and synth your CDK
project. Here are some (non-exhaustive) examples for what those commands might
look like in a number of different situations.
For Yarn, the install commands are different:
const pipeline = new CodePipeline(this, 'Pipeline', {
synth: new ShellStep('Synth', {
input: source,
commands: [
'yarn install --frozen-lockfile',
'yarn build',
'npx cdk synth',
],
})
});
For Python projects, remember to install the CDK CLI globally (as
there is no package.json
to automatically install it for you):
const pipeline = new CodePipeline(this, 'Pipeline', {
synth: new ShellStep('Synth', {
input: source,
commands: [
'pip install -r requirements.txt',
'npm install -g aws-cdk',
'cdk synth',
],
})
});
For Java projects, remember to install the CDK CLI globally (as
there is no package.json
to automatically install it for you),
and the Maven compilation step is automatically executed for you
as you run cdk synth
:
const pipeline = new CodePipeline(this, 'Pipeline', {
synth: new ShellStep('Synth', {
input: source,
commands: [
'npm install -g aws-cdk',
'cdk synth',
],
})
});
You can adapt these examples to your own situation.
In CodePipeline, Sources define where the source of your application lives.
When a change to the source is detected, the pipeline will start executing.
Source objects can be created by factory methods on the CodePipelineSource
class:
The recommended way of connecting to GitHub or BitBucket is by using a connection. You will first use the AWS Console to authenticate to the source control provider, and then use the connection ARN in your pipeline definition:
CodePipelineSource.connection('org/repo', 'branch', {
connectionArn: 'arn:aws:codestar-connections:us-east-1:222222222222:connection/7d2469ff-514a-4e4f-9003-5ca4a43cdc41',
});
You can also authenticate to GitHub using a personal access token. This expects that you've created a personal access token and stored it in Secrets Manager. By default, the source object will look for a secret named github-token, but you can change the name. The token should have the repo and admin:repo_hook scopes.
CodePipelineSource.gitHub('org/repo', 'branch', {
// This is optional
authentication: SecretValue.secretsManager('my-token'),
});
You can use a CodeCommit repository as the source. Either create or import
that the CodeCommit repository and then use CodePipelineSource.codeCommit
to reference it:
const repository = codecommit.fromRepositoryName(this, 'Repository', 'my-repository');
CodePipelineSource.codeCommit(repository);
You can use a zip file in S3 as the source of the pipeline. The pipeline will be triggered every time the file in S3 is changed:
const bucket = s3.Bucket.fromBucketName(this, 'Bucket', 'my-bucket');
CodePipelineSource.s3(bucket, 'my/source.zip');
You can use a Docker image in ECR as the source of the pipeline. The pipeline will be triggered every time an image is pushed to ECR:
const repository = ecr.Repository(this, 'Repository');
CodePipelineSource.ecr(repository);
ShellStep
allows passing in more than one input: additional
inputs will be placed in the directories you specify. Any step that produces an
output file set can be used as an input, such as a CodePipelineSource
, but
also other ShellStep
:
const prebuild = new ShellStep('Prebuild', {
input: CodePipelineSource.gitHub('myorg/repo1'),
primaryOutputDirectory: './build',
commands: ['./build.sh'],
});
const pipeline = new CodePipeline(this, 'Pipeline', {
synth: new ShellStep('Synth', {
input: CodePipelineSource.gitHub('myorg/repo2'),
additionalInputs: {
'subdir': CodePipelineSource.gitHub('myorg/repo3'),
'../siblingdir': prebuild,
},
commands: ['./build.sh'],
})
});
After you have defined the pipeline and the synth
step, you can add one or
more CDK Stages
which will be deployed to their target environments. To do
so, call pipeline.addStage()
on the Stage object:
// Do this as many times as necessary with any account and region
// Account and region may different from the pipeline's.
pipeline.addStage(new MyApplicationStage(this, 'Prod', {
env: {
account: '123456789012',
region: 'eu-west-1',
}
}));
CDK Pipelines will automatically discover all Stacks
in the given Stage
object, determine their dependency order, and add appropriate actions to the
pipeline to publish the assets referenced in those stacks and deploy the stacks
in the right order.
If the Stacks
are targeted at an environment in a different AWS account or
Region and that environment has been
bootstrapped
, CDK Pipelines will transparently make sure the IAM roles are set up
correctly and any requisite replication Buckets are created.
By default, all applications added to CDK Pipelines by calling addStage()
will
be deployed in sequence, one after the other. If you have a lot of stages, you can
speed up the pipeline by choosing to deploy some stages in parallel. You do this
by calling addWave()
instead of addStage()
: a wave is a set of stages that
are all deployed in parallel instead of sequentially. Waves themselves are still
deployed in sequence. For example, the following will deploy two copies of your
application to eu-west-1
and eu-central-1
in parallel:
const europeWave = pipeline.addWave('Europe');
europeWave.addStage(new MyApplicationStage(this, 'Ireland', {
env: { region: 'eu-west-1' }
}));
europeWave.addStage(new MyApplicationStage(this, 'Germany', {
env: { region: 'eu-central-1' }
}));
CDK Pipelines can transparently deploy to other Regions and other accounts
(provided those target environments have been
bootstrapped).
However, deploying to another account requires one additional piece of
configuration: you need to enable crossAccountKeys: true
when creating the
pipeline.
This will encrypt the artifact bucket(s), but incurs a cost for maintaining the KMS key.
Example:
const pipeline = new CodePipeline(this, 'Pipeline', {
// Encrypt artifacts, required for cross-account deployments
crossAccountKeys: true,
});
Every addStage()
and addWave()
command takes additional options. As part of these options,
you can specify pre
and post
steps, which are arbitrary steps that run before or after
the contents of the stage or wave, respectively. You can use these to add validations like
manual or automated gates to your pipeline. We recommend putting manual approval gates in the set of pre
steps, and automated approval gates in
the set of post
steps.
The following example shows both an automated approval in the form of a ShellStep
, and
a manual approvel in the form of a ManualApprovalStep
added to the pipeline. Both must
pass in order to promote from the PreProd
to the Prod
environment:
const preprod = new MyApplicationStage(this, 'PreProd', { ... });
const prod = new MyApplicationStage(this, 'Prod', { ... });
pipeline.addStage(preprod, {
post: [
new ShellStep('Validate Endpoint', {
commands: ['curl -Ssf https://my.webservice.com/'],
}),
],
});
pipeline.addStage(prod, {
pre: [
new ManualApprovalStep('PromoteToProd'),
],
});
Because many CloudFormation deployments result in the generation of resources with unpredictable names, validations have support for reading back CloudFormation Outputs after a deployment. This makes it possible to pass (for example) the generated URL of a load balancer to the test set.
To use Stack Outputs, expose the CfnOutput
object you're interested in, and
pass it to envFromCfnOutputs
of the ShellStep
:
class MyApplicationStage extends Stage {
public readonly loadBalancerAddress: CfnOutput;
// ...
}
const lbApp = new MyApplicationStage(this, 'MyApp', { /* ... */ });
pipeline.addStage(lbApp, {
post: [
new ShellStep('HitEndpoint', {
envFromCfnOutputs: {
// Make the load balancer address available as $URL inside the commands
URL: lbApp.loadBalancerAddress,
},
commands: ['curl -Ssf $URL'],
});
],
});
As part of a validation, you probably want to run a test suite that's more
elaborate than what can be expressed in a couple of lines of shell script.
You can bring additional files into the shell script validation by supplying
the input
or additionalInputs
property of ShellStep
. The input can
be produced by the Synth
step, or come from a source or any other build
step.
Here's an example that captures an additional output directory in the synth step and runs tests from there:
const synth = new ShellStep('Synth', { /* ... */ });
const pipeline = new CodePipeline(this, 'Pipeline', { synth });
pipeline.addStage(/* ... */, {
post: [
new ShellStep('Approve', {
// Use the contents of the 'integ' directory from the synth step as the input
input: synth.addOutputDirectory('integ'),
commands: ['cd integ && ./run.sh'],
}),
],
});
CDK pipelines will generate CodeBuild projects for each ShellStep
you use, and it
will also generate CodeBuild projects to publish assets and perform the self-mutation
of the pipeline. To control the various aspects of the CodeBuild projects that get
generated, use a CodeBuildStep
instead of a ShellStep
. This class has a number
of properties that allow you to customize various aspects of the projects:
new CodeBuildStep('Synth', {
// ...standard ShellStep props...
commands: [/* ... */],
env: { /* ... */ },
// If you are using a CodeBuildStep explicitly, set the 'cdk.out' directory
// to be the synth step's output.
primaryOutputDirectory: 'cdk.out',
// Control the name of the project
projectName: 'MyProject',
// Control parts of the BuildSpec other than the regular 'build' and 'install' commands
partialBuildSpec: codebuild.BuildSpec.fromObject({
version: '0.2',
// ...
}),
// Control the build environment
buildEnvironment: {
computeType: codebuild.ComputeType.LARGE,
},
// Control Elastic Network Interface creation
vpc: vpc,
subnetSelection: { subnetType: ec2.SubnetType.PRIVATE },
securityGroups: [mySecurityGroup],
// Additional policy statements for the execution role
rolePolicyStatements: [
new iam.PolicyStatement({ /* ... */ }),
],
});
You can also configure defaults for all CodeBuild projects by passing codeBuildDefaults
,
or just for the synth, asset publishing, and self-mutation projects by passing synthCodeBuildDefaults
,
assetPublishingCodeBuildDefaults
, or selfMutationCodeBuildDefaults
:
new CodePipeline(this, 'Pipeline', {
// ...
// Defaults for all CodeBuild projects
codeBuildDefaults: {
// Prepend commands and configuration to all projects
partialBuildSpec: codebuild.BuildSpec.fromObject({
version: '0.2',
// ...
}),
// Control the build environment
buildEnvironment: {
computeType: codebuild.ComputeType.LARGE,
},
// Control Elastic Network Interface creation
vpc: vpc,
subnetSelection: { subnetType: ec2.SubnetType.PRIVATE },
securityGroups: [mySecurityGroup],
// Additional policy statements for the execution role
rolePolicy: [
new iam.PolicyStatement({ /* ... */ }),
],
},
synthCodeBuildDefaults: { /* ... */ },
assetPublishingCodeBuildDefaults: { /* ... */ },
selfMutationCodeBuildDefaults: { /* ... */ },
});
If you want to add a type of CodePipeline action to the CDK Pipeline that
doesn't have a matching class yet, you can define your own step class that extends
Step
and implements ICodePipelineActionFactory
.
Here's an example that adds a Jenkins step:
class MyJenkinsStep extends Step implements ICodePipelineActionFactory {
constructor(private readonly provider: codepipeline_actions.JenkinsProvider, private readonly input: FileSet) {
}
public produceAction(stage: codepipeline.IStage, options: ProduceActionOptions): CodePipelineActionFactoryResult {
// This is where you control what type of Action gets added to the
// CodePipeline
stage.addAction(new codepipeline_actions.JenkinsAction({
// Copy 'actionName' and 'runOrder' from the options
actionName: options.actionName,
runOrder: options.runOrder,
// Jenkins-specific configuration
type: cpactions.JenkinsActionType.TEST,
jenkinsProvider: this.provider,
projectName: 'MyJenkinsProject',
// Translate the FileSet into a codepipeline.Artifact
inputs: [options.artifacts.toCodePipeline(this.input)],
}));
return { runOrdersConsumed: 1 };
}
}
Docker can be used in 3 different places in the pipeline:
- If you are using Docker image assets in your application stages: Docker will run in the asset publishing projects.
- If you are using Docker image assets in your stack (for example as images for your CodeBuild projects): Docker will run in the self-mutate project.
- If you are using Docker to bundle file assets anywhere in your project (for
example, if you are using such construct libraries as
@aws-cdk/aws-lambda-nodejs
): Docker will run in the synth project.
For the first case, you don't need to do anything special. For the other two cases, you need to make sure that privileged mode is enabled on the correct CodeBuild projects, so that Docker can run correctly. The follow sections describe how to do that.
You may also need to authenticate to Docker registries to avoid being throttled. See the section Authenticating to Docker registries below for information on how to do that.
If your PipelineStack
is using Docker image assets (as opposed to the application
stacks the pipeline is deploying), for example by the use of LinuxBuildImage.fromAsset()
,
you need to pass dockerEnabledForSelfMutation: true
to the pipeline. For example:
const pipeline = new CodePipeline(this, 'Pipeline', {
// ...
// Turn this on because the pipeline uses Docker image assets
dockerEnabledForSelfMutation: true,
});
pipeline.addWave('MyWave', {
post: [
new CodeBuildStep('RunApproval', {
commands: ['command-from-image'],
buildEnvironment: {
// The user of a Docker image asset in the pipeline requires turning on
// 'dockerEnabledForSelfMutation'.
buildImage: LinuxBuildImage.fromAsset(this, 'Image', {
directory: './docker-image',
})
},
})
],
});
Important: You must turn on the
dockerEnabledForSelfMutation
flag, commit and allow the pipeline to self-update before adding the actual Docker asset.
If you are using asset bundling anywhere (such as automatically done for you
if you add a construct like @aws-cdk/aws-lambda-nodejs
), you need to pass
dockerEnabledForSynth: true
to the pipeline. For example:
const pipeline = new CodePipeline(this, 'Pipeline', {
// ...
// Turn this on because the application uses bundled file assets
dockerEnabledForSynth: true,
});
Important: You must turn on the
dockerEnabledForSynth
flag, commit and allow the pipeline to self-update before adding the actual Docker asset.
You can specify credentials to use for authenticating to Docker registries as part of the pipeline definition. This can be useful if any Docker image assets — in the pipeline or any of the application stages — require authentication, either due to being in a different environment (e.g., ECR repo) or to avoid throttling (e.g., DockerHub).
const dockerHubSecret = secretsmanager.Secret.fromSecretCompleteArn(this, 'DHSecret', 'arn:aws:...');
const customRegSecret = secretsmanager.Secret.fromSecretCompleteArn(this, 'CRSecret', 'arn:aws:...');
const repo1 = ecr.Repository.fromRepositoryArn(stack, 'Repo', 'arn:aws:ecr:eu-west-1:0123456789012:repository/Repo1');
const repo2 = ecr.Repository.fromRepositoryArn(stack, 'Repo', 'arn:aws:ecr:eu-west-1:0123456789012:repository/Repo2');
const pipeline = new CodePipeline(this, 'Pipeline', {
dockerCredentials: [
DockerCredential.dockerHub(dockerHubSecret),
DockerCredential.customRegistry('dockerregistry.example.com', customRegSecret),
DockerCredential.ecr([repo1, repo2]);
],
// ...
});
For authenticating to Docker registries that require a username and password combination
(like DockerHub), create a Secrets Manager Secret with fields named username
and secret
, and import it (the field names change be customized).
Authentication to ECR repostories is done using the execution role of the relevant CodeBuild job. Both types of credentials can be provided with an optional role to assume before requesting the credentials.
By default, the Docker credentials provided to the pipeline will be available to
the Synth, Self-Update, and Asset Publishing actions within the
*pipeline. The scope of the credentials can be limited via the DockerCredentialUsage
option.
const dockerHubSecret = secretsmanager.Secret.fromSecretCompleteArn(this, 'DHSecret', 'arn:aws:...');
// Only the image asset publishing actions will be granted read access to the secret.
const creds = DockerCredential.dockerHub(dockerHubSecret, { usages: [DockerCredentialUsage.ASSET_PUBLISHING] });
An environment is an (account, region) pair where you want to deploy a CDK stack (see Environments in the CDK Developer Guide). In a Continuous Deployment pipeline, there are at least two environments involved: the environment where the pipeline is provisioned, and the environment where you want to deploy the application (or different stages of the application). These can be the same, though best practices recommend you isolate your different application stages from each other in different AWS accounts or regions.
Before you can provision the pipeline, you have to bootstrap the environment you want to create it in. If you are deploying your application to different environments, you also have to bootstrap those and be sure to add a trust relationship.
After you have bootstrapped an environment and created a pipeline that deploys to it, it's important that you don't delete the stack or change its Qualifier, or future deployments to this environment will fail. If you want to upgrade the bootstrap stack to a newer version, do that by updating it in-place.
This library requires the modern bootstrapping stack which has been updated specifically to support cross-account continuous delivery. Starting, in CDK v2 this new bootstrapping stack will become the default, but for now it is still opt-in.
The commands below assume you are running
cdk bootstrap
in a directory wherecdk.json
contains the"@aws-cdk/core:newStyleStackSynthesis": true
setting in its context, which will switch to the new bootstrapping stack automatically.If run from another directory, be sure to run the bootstrap command with the environment variable
CDK_NEW_BOOTSTRAP=1
set.
To bootstrap an environment for provisioning the pipeline:
$ env CDK_NEW_BOOTSTRAP=1 npx cdk bootstrap \
[--profile admin-profile-1] \
--cloudformation-execution-policies arn:aws:iam::aws:policy/AdministratorAccess \
aws://111111111111/us-east-1
To bootstrap a different environment for deploying CDK applications into using
a pipeline in account 111111111111
:
$ env CDK_NEW_BOOTSTRAP=1 npx cdk bootstrap \
[--profile admin-profile-2] \
--cloudformation-execution-policies arn:aws:iam::aws:policy/AdministratorAccess \
--trust 11111111111 \
aws://222222222222/us-east-2
If you only want to trust an account to do lookups (e.g, when your CDK application has a
Vpc.fromLookup()
call), use the option --trust-for-lookup
:
$ env CDK_NEW_BOOTSTRAP=1 npx cdk bootstrap \
[--profile admin-profile-2] \
--cloudformation-execution-policies arn:aws:iam::aws:policy/AdministratorAccess \
--trust-for-lookup 11111111111 \
aws://222222222222/us-east-2
These command lines explained:
npx
: means to use the CDK CLI from the current NPM install. If you are using a global install of the CDK CLI, leave this out.--profile
: should indicate a profile with administrator privileges that has permissions to provision a pipeline in the indicated account. You can leave this flag out if either the AWS default credentials or theAWS_*
environment variables confer these permissions.--cloudformation-execution-policies
: ARN of the managed policy that future CDK deployments should execute with. By default this isAdministratorAccess
, but if you also specify the--trust
flag to give another Account permissions to deploy into the current account, you must specify a value here.--trust
: indicates which other account(s) should have permissions to deploy CDK applications into this account. In this case we indicate the Pipeline's account, but you could also use this for developer accounts (don't do that for production application accounts though!).--trust-for-lookup
: gives a more limited set of permissions to the trusted account, only allowing it to look up values such as availability zones, EC2 images and VPCs.--trust-for-lookup
does not give permissions to modify anything in the account. Note that--trust
implies--trust-for-lookup
, so you don't need to specify the same acocunt twice.aws://222222222222/us-east-2
: the account and region we're bootstrapping.
Be aware that anyone who has access to the trusted Accounts effectively has all permissions conferred by the configured CloudFormation execution policies, allowing them to do things like read arbitrary S3 buckets and create arbitrary infrastructure in the bootstrapped account. Restrict the list of
--trust
ed Accounts, or restrict the policies configured by--cloudformation-execution-policies
.
Security tip: we recommend that you use administrative credentials to an account only to bootstrap it and provision the initial pipeline. Otherwise, access to administrative credentials should be dropped as soon as possible.
On the use of AdministratorAccess: The use of the
AdministratorAccess
policy ensures that your pipeline can deploy every type of AWS resource to your account. Make sure you trust all the code and dependencies that make up your CDK app. Check with the appropriate department within your organization to decide on the proper policy to use.If your policy includes permissions to create on attach permission to a role, developers can escalate their privilege with more permissive permission. Thus, we recommend implementing permissions boundary in the CDK Execution role. To do this, you can bootstrap with the
--template
option with a customized template that contains a permission boundary.
The bootstrap stack is a CloudFormation stack in your account named CDKToolkit that provisions a set of resources required for the CDK to deploy into that environment.
The "new" bootstrap stack (obtained by running cdk bootstrap
with
CDK_NEW_BOOTSTRAP=1
) is slightly more elaborate than the "old" stack. It
contains:
- An S3 bucket and ECR repository with predictable names, so that we can reference assets in these storage locations without the use of CloudFormation template parameters.
- A set of roles with permissions to access these asset locations and to execute
CloudFormation, assumable from whatever accounts you specify under
--trust
.
It is possible and safe to migrate from the old bootstrap stack to the new bootstrap stack. This will create a new S3 file asset bucket in your account and orphan the old bucket. You should manually delete the orphaned bucket after you are sure you have redeployed all CDK applications and there are no more references to the old asset bucket.
You might be using CDK constructs that need to look up runtime
context,
which is information from the target AWS Account and Region the CDK needs to
synthesize CloudFormation templates appropriate for that environment. Examples
of this kind of context lookups are the number of Availability Zones available
to you, a Route53 Hosted Zone ID, or the ID of an AMI in a given region. This
information is automatically looked up when you run cdk synth
.
By default, a cdk synth
performed in a pipeline will not have permissions
to perform these lookups, and the lookups will fail. This is by design.
Our recommended way of using lookups is by running cdk synth
on the
developer workstation and checking in the cdk.context.json
file, which
contains the results of the context lookups. This will make sure your
synthesized infrastructure is consistent and repeatable. If you do not commit
cdk.context.json
, the results of the lookups may suddenly be different in
unexpected ways, and even produce results that cannot be deployed or will cause
data loss. To give an account permissions to perform lookups against an
environment, without being able to deploy to it and make changes, run
cdk bootstrap --trust-for-lookup=<account>
.
If you want to use lookups directly from the pipeline, you either need to accept
the risk of nondeterminism, or make sure you save and load the
cdk.context.json
file somewhere between synth runs. Finally, you should
give the synth CodeBuild execution role permissions to assume the bootstrapped
lookup roles. As an example, doing so would look like this:
new CodePipeline(this, 'Pipeline', {
synth: new CodeBuildStep('Synth', {
input: // ...input...
commands: [
// Commands to load cdk.context.json from somewhere here
'...',
'npm ci',
'npm run build',
'npx cdk synth',
// Commands to store cdk.context.json back here
'...',
],
rolePolicyStatements: [
new iam.PolicyStatement({
actions: ['sts:AssumeRole'],
resources: ['*'],
conditions: {
StringEquals: {
'iam:ResourceTag/aws-cdk:bootstrap-role': 'lookup',
},
},
}),
],
}),
});
The above example requires that the target environments have all
been bootstrapped with bootstrap stack version 8
, released with
CDK CLI 1.114.0
.
It's important to stay safe while employing Continuous Delivery. The CDK Pipelines library comes with secure defaults to the best of our ability, but by its very nature the library cannot take care of everything.
We therefore expect you to mind the following:
-
Maintain dependency hygiene and vet 3rd-party software you use. Any software you run on your build machine has the ability to change the infrastructure that gets deployed. Be careful with the software you depend on.
-
Use dependency locking to prevent accidental upgrades! The default
CdkSynths
that come with CDK Pipelines will expectpackage-lock.json
andyarn.lock
to ensure your dependencies are the ones you expect. -
Credentials to production environments should be short-lived. After bootstrapping and the initial pipeline provisioning, there is no more need for developers to have access to any of the account credentials; all further changes can be deployed through git. Avoid the chances of credentials leaking by not having them in the first place!
To keep tabs on the security impact of changes going out through your pipeline, you can insert a security check before any stage deployment. This security check will check if the upcoming deployment would add any new IAM permissions or security group rules, and if so pause the pipeline and require you to confirm the changes.
The security check will appear as two distinct actions in your pipeline: first
a CodeBuild project that runs cdk diff
on the stage that's about to be deployed,
followed by a Manual Approval action that pauses the pipeline. If it so happens
that there no new IAM permissions or security group rules will be added by the deployment,
the manual approval step is automatically satisfied. The pipeline will look like this:
Pipeline
├── ...
├── MyApplicationStage
│ ├── MyApplicationSecurityCheck // Security Diff Action
│ ├── MyApplicationManualApproval // Manual Approval Action
│ ├── Stack.Prepare
│ └── Stack.Deploy
└── ...
You can insert the security check by using a ConfirmPermissionsBroadening
step:
const stage = new MyApplicationStage(this, 'MyApplication');
pipeline.addStage(stage, {
pre: [
new ConfirmPermissionsBroadening('Check', { stage }),
],
});
To get notified when there is a change that needs your manual approval,
create an SNS Topic, subscribe your own email address, and pass it in as
as the notificationTopic
property:
import * as sns from '@aws-cdk/aws-sns';
import * as subscriptions from '@aws-cdk/aws-sns-subscriptions';
import * as pipelines from '@aws-cdk/pipelines';
const topic = new sns.Topic(this, 'SecurityChangesTopic');
topic.addSubscription(new subscriptions.EmailSubscription('test@email.com'));
const stage = new MyApplicationStage(this, 'MyApplication');
pipeline.addStage(stage, {
pre: [
new ConfirmPermissionsBroadening('Check', {
stage,
notificationTopic: topic,
}),
],
});
Note: Manual Approvals notifications only apply when an application has security check enabled.
Here are some common errors you may encounter while using this library.
If you see the following error during deployment of your pipeline:
CREATE_FAILED | AWS::CodePipeline::Pipeline | Pipeline/Pipeline
Internal Failure
There's something wrong with your GitHub access token. It might be missing, or not have the right permissions to access the repository you're trying to access.
If you see the following error during deployment of your pipeline:
CREATE_FAILED | AWS::KMS::Key | Pipeline/Pipeline/ArtifactsBucketEncryptionKey
Policy contains a statement with one or more invalid principals.
One of the target (account, region) environments has not been bootstrapped with the new bootstrap stack. Check your target environments and make sure they are all bootstrapped.
If you see this error during the Synth step, it means that CodeBuild
is expecting to find a cdk.out
directory in the root of your CodeBuild project,
but the directory wasn't there. There are two common causes for this:
cdk synth
is not being executed:cdk synth
used to be run implicitly for you, but you now have to explicitly include the command. For NPM-based projects, addnpx cdk synth
to the end of thecommands
property, for other languages addnpm install -g aws-cdk
andcdk synth
.- Your CDK project lives in a subdirectory: you added a
cd <somedirectory>
command to the list of commands; don't forget to tell theScriptStep
about the different location ofcdk.out
, by passingprimaryOutputDirectory: '<somedirectory>/cdk.out'
.
If you see the following error during execution of your pipeline:
Stack ... is in ROLLBACK_COMPLETE state and can not be updated. (Service:
AmazonCloudFormation; Status Code: 400; Error Code: ValidationError; Request
ID: ...)
The stack failed its previous deployment, and is in a non-retryable state. Go into the CloudFormation console, delete the stack, and retry the deployment.
You may see this if you are using TypeScript or other NPM-based languages,
when using NPM 7 on your workstation (where you generate package-lock.json
)
and NPM 6 on the CodeBuild image used for synthesizing.
It looks like NPM 7 has started writing less information to package-lock.json
,
leading NPM 6 reading that same file to not install all required packages anymore.
Make sure you are using the same NPM version everywhere, either downgrade your workstation's version or upgrade the CodeBuild version.
The above error may be produced by npx
when executing the CDK CLI, or any
project that uses the AWS SDK for JavaScript, without the target application
having been installed yet. For example, it can be triggered by npx cdk synth
if aws-cdk
is not in your package.json
.
Work around this by either installing the target application using NPM before
running npx
, or set the environment variable NPM_CONFIG_UNSAFE_PERM=true
.
If, in the 'Synth' action (inside the 'Build' stage) of your pipeline, you get an error like this:
stderr: docker: Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?.
See 'docker run --help'.
It means that the AWS CodeBuild project for 'Synth' is not configured to run in privileged mode,
which prevents Docker builds from happening. This typically happens if you use a CDK construct
that bundles asset using tools run via Docker, like aws-lambda-nodejs
, aws-lambda-python
,
aws-lambda-go
and others.
Make sure you set the privileged
environment variable to true
in the synth definition:
const pipeline = new CdkPipeline(this, 'MyPipeline', {
...
synthAction: SimpleSynthAction.standardNpmSynth({
sourceArtifact: ...,
cloudAssemblyArtifact: ...,
environment: {
privileged: true,
},
}),
});
After turning on privilegedMode: true
, you will need to do a one-time manual cdk deploy of your
pipeline to get it going again (as with a broken 'synth' the pipeline will not be able to self
update to the right state).
An "S3 Access Denied" error can have two causes:
- Asset hashes have changed, but self-mutation has been disabled in the pipeline.
- You have deleted and recreated the bootstrap stack, or changed its qualifier.
Some constructs, such as EKS clusters, generate nested stacks. When CloudFormation tries to deploy those stacks, it may fail with this error:
S3 error: Access Denied For more information check http://docs.aws.amazon.com/AmazonS3/latest/API/ErrorResponses.html
This happens because the pipeline is not self-mutating and, as a consequence, the FileAssetX
build projects get out-of-sync with the generated templates. To fix this, make sure the
selfMutating
property is set to true
:
const pipeline = new CdkPipeline(this, 'MyPipeline', {
selfMutating: true,
...
});
While attempting to deploy an application stage, the "Prepare" or "Deploy" stage may fail with a cryptic error like:
Action execution failed Access Denied (Service: Amazon S3; Status Code: 403; Error Code: AccessDenied; Request ID: 0123456ABCDEFGH; S3 Extended Request ID: 3hWcrVkhFGxfiMb/rTJO0Bk7Qn95x5ll4gyHiFsX6Pmk/NT+uX9+Z1moEcfkL7H3cjH7sWZfeD0=; Proxy: null)
This generally indicates that the roles necessary to deploy have been deleted (or deleted and re-created);
for example, if the bootstrap stack has been deleted and re-created, this scenario will happen. Under the hood,
the resources that rely on these roles (e.g., cdk-$qualifier-deploy-role-$account-$region
) point to different
canonical IDs than the recreated versions of these roles, which causes the errors. There are no simple solutions
to this issue, and for that reason we strongly recommend that bootstrap stacks not be deleted and re-created
once created.
The most automated way to solve the issue is to introduce a secondary bootstrap stack. By changing the qualifier that the pipeline stack looks for, a change will be detected and the impacted policies and resources will be updated. A hypothetical recovery workflow would look something like this:
- First, for all impacted environments, create a secondary bootstrap stack:
$ env CDK_NEW_BOOTSTRAP=1 npx cdk bootstrap \
--qualifier randchars1234
--toolkit-stack-name CDKToolkitTemp
aws://111111111111/us-east-1
- Update all impacted stacks in the pipeline to use this new qualifier. See https://docs.aws.amazon.com/cdk/latest/guide/bootstrapping.html for more info.
new MyStack(this, 'MyStack', {
// Update this qualifier to match the one used above.
synthesizer: new DefaultStackSynthesizer({
qualifier: 'randchars1234',
}),
});
- Deploy the updated stacks. This will update the stacks to use the roles created in the new bootstrap stack.
- (Optional) Restore back to the original state:
- Revert the change made in step #2 above
- Re-deploy the pipeline to use the original qualifier.
- Delete the temporary bootstrap stack(s)
Alternatively, the errors can be resolved by finding each impacted resource and policy, and correcting the policies
by replacing the canonical IDs (e.g., AROAYBRETNYCYV6ZF2R93
) with the appropriate ARNs. As an example, the KMS
encryption key policy for the artifacts bucket may have a statement that looks like the following:
{
"Effect" : "Allow",
"Principal" : {
// "AWS" : "AROAYBRETNYCYV6ZF2R93" // Indicates this issue; replace this value
"AWS": "arn:aws:iam::0123456789012:role/cdk-hnb659fds-deploy-role-0123456789012-eu-west-1", // Correct value
},
"Action" : [ "kms:Decrypt", "kms:DescribeKey" ],
"Resource" : "*"
}
Any resource or policy that references the qualifier (hnb659fds
by default) will need to be updated.
There are some usability issues that are caused by underlying technology, and cannot be remedied by CDK at this point. They are reproduced here for completeness.
- Console links to other accounts will not work: the AWS CodePipeline console will assume all links are relative to the current account. You will not be able to use the pipeline console to click through to a CloudFormation stack in a different account.
- If a change set failed to apply the pipeline must restarted: if a change set failed to apply, it cannot be retried. The pipeline must be restarted from the top by clicking Release Change.
- A stack that failed to create must be deleted manually: if a stack failed to create on the first attempt, you must delete it using the CloudFormation console before starting the pipeline again by clicking Release Change.