A compatibility layer for using the CDK for Terraform with AWS CDK constructs
The CDK for Terraform is a great tool for managing infrastructure, but it lacks the ecosystem of constructs that the AWS CDK has. This tool allows you to use the excellent aws-cdk-lib construct library, while also getting access to the benefits of terraform. Think of provisioning a CockroachDB cloud cluster and connecting it to a lambda using the high level nodejs function construct in the same stack.
Yes, but it's in technical preview and missing many essential features. It also doesn't seem to be actively maintained.
- Bidirectional references between AWS CDK and CDK for Terraform constructs
- Interstack referencing
- Dependency ordering compatibility
- Cloudformation intrinsics
- Assets
- References to AWS CDK constructs only work within the scope of adaptor classes
- AWS CDK aspects are not supported
- Custom resources are not supported
Essentially, just use the provided AwsTerraformAdaptorStack
class as your base stack class instead of TerraformStack
import {AwsTerraformAdaptorStack} from "./cdk-adaptor-stack";
import {Bucket} from "aws-cdk-lib";
import {S3Bucket} from "@cdktf/provider-aws/lib/s3-bucket/index.js";
import {S3BucketObject} from "@cdktf/provider-aws/lib/s3-bucket-object";
class MyStack extends AwsTerraformAdaptorStack {
constructor(scope: Construct, id: string, region: string) {
super(scope, id, region);
// You can instantiate AWS CDK constructs as normal
const bucket = new Bucket(this, 'MyBucket', {
bucket: 'my-bucket',
});
// You can also instantiate CDK for Terraform constructs
// Bidirectional references between the two are automatically created
const tfBucket = new S3BucketObject(this, 'MyTfBucket', {
bucket: bucket.bucketName,
key: 'my-tf-bucket',
source: 'my-tf-bucket',
});
}
}
Internally, the adaptor tries to use the CloudControl api for 1-1 compatibility with Cloudformation. In some cases this is enough, but in many others explicit resource mappings are required. The adaptor ships with a good variety of these mappings, but it is not exhaustive.
If you encounter an error like No resource mapping found for <resource type>
, you can add a mapping for it like so:
// Map between a L1 Cloudformation construct and a CDK for Terraform construct
registerMappingTyped(CfnBucketPolicy, S3BucketPolicy, {
resource(scope, id, props) {
return new S3BucketPolicy(
scope,
id,
deleteUndefinedKeys({
// ALL keys of the provided construct must be mapped
policy: Fn.jsonencode(props.PolicyDocument),
bucket: props.Bucket,
}),
);
},
attributes: {
// ALL attributes of the provided construct must be mapped
Ref: (resource) => resource.id,
},
});
The Cloudcontrol api is theoretically a stable crud API 1-1 with cloudformation, but it in practice it has some quirks. Update operations in particular seems to break with alarming regularity, so explicit mappings are preferred, especially for complex resources.