Break the rules of virtualization, build Lambdas container images for any platform, from any platform, with CDK & Docker buildx

How often are you deploying a Lambda container image, basically a Lambda running on a Docker image, for a platform that doesn't match your localhost platform? Often I deploy functions running on ARM rather than on X86_64, this is a personal preference and it doesn't come with any massive advantage (there are online some comparison), and although my laptop is ARM-based, the CI/CD server is not :/

This is true even for GitHub Workflows which run on X86_64, rather than ARM if you don't create your custom runner.

So if you have a piece of code like the following, and you try to deploy it via a CI/CD server which is X86_64 it will fail with the error

The error

Warning: The requested image's platform (linux/arm64/v8) does not match the detected host platform (linux/amd64) and no specific platform was requested

exec /bin/sh: exec format error
The command '/bin/sh -c ...' returned a non-zero code: 1

The code

from aws_cdk import (    
    Stack,
    aws_lambda,
    aws_ecr_assets as ecr
)
from constructs import Construct

class LambdaMultiplatDemStack(Stack):

    def __init__(self, scope: Construct, construct_id: str, **kwargs) -> None:
        super().__init__(scope, construct_id, **kwargs)

        my_function = aws_lambda.DockerImageFunction(
            self, 
            "MyDifferentPlatformFn",
            code=aws_lambda.DockerImageCode.from_image_asset(
              "."
            ),
            architecture=aws_lambda.Architecture.ARM_64
        )

The Dockerfile

FROM public.ecr.aws/lambda/python:3.8

COPY requirements.txt  .

RUN  pip3 install -r requirements.txt --target "${LAMBDA_TASK_ROOT}"
COPY my_py_app.py ${LAMBDA_TASK_ROOT}

CMD [ "my_py_app.handler" ]

This is true even for the simplest Docker image because Docker is pulling executable dependencies for your it that are not compatible with your execution platform. How to solve this issue? Thanks to Docker and CDK!

Let's start with Docker

Is not news that Docker offers the possibility to build images for a platform different than yours using buildx, to find out more just visit the official documentation link.

So first thing first, we need to check if buildx is installed on our machine; you can follow the official doc here, but if you want to simplify the process

1. Run docker buildx ls to list all the builders on your machine, if the output presents something similar to what is below you should be good (see the various platform available)

[panilo@fedora ~]$ docker buildx ls 
NAME/NODE DRIVER/ENDPOINT STATUS  BUILDKIT PLATFORMS
default * docker                           
  default default         running 23.0.1   linux/arm64, linux/amd64, linux/amd64/v2, linux/riscv64, linux/ppc64le, linux/s390x, linux/386, linux/mips64le, linux/mips64

1. If the output doesn't present the platform you're interested in, you need to
   • run the following command docker run --privileged --rm tonistiigi/binfmt --install all
   • run again the docker buildx ls command, this time the platforms available should be more

Now we can build images for other platforms, great!

Step 2: CDK

Just modify the stack to add a special platform property in the aws_lambda.DockerImageCode.from_image_asset definition

from aws_cdk import (    
    Stack,
    aws_lambda,
    aws_ecr_assets as ecr
)
from constructs import Construct

class LambdaMultiplatDemStack(Stack):

    def __init__(self, scope: Construct, construct_id: str, **kwargs) -> None:
        super().__init__(scope, construct_id, **kwargs)

        my_function = aws_lambda.DockerImageFunction(
            self, 
            "MyDifferentPlatformFn",
            code=aws_lambda.DockerImageCode.from_image_asset(
              ".",
              platform=ecr.Platform.LINUX_ARM64 # Magic Switch!
            ),
            architecture=aws_lambda.Architecture.ARM_64
        )

By adding this little parameter CDK will know to use Docker's buildx build command and prepare the image for the specified platform.

We can now deploy anywhere, building anywhere :)

Please find a complete example here.

I hope this helped you,
as usual, any feedback is more than appreciated!

Cheers
Danilo

Post published also on Virtuability's website here.