This project is a minimum working example of deploying an AWS Lambda function using Docker and AWS CDK.
Deploying Lambda functions using Docker has a number of benefits
- Package all necessary libraries into a single Docker image
- Bypass AWS Lambda's size constraint of 512 mb. Docker images stored on AWS ECR have a maximum size of 10 gb.
I will show you how to deploy
Tensorflow
models in another post! - It’s easy!
CDK init & deploy
I won’t cover setting up CDK and bootstrapping the environment. You can find that information here.
Once you have set up CDK, we need to set up the project:
mkdir cdk_docker_lambda && cd cdk_docker_lambda
cdk init --language python
source .venv/bin/activate
-
pip install -r requirements.txt && pip install -r requirements-dev.txt
Now deploy empty stack to AWS:
cdk deploy
Stack design
Our stack will deploy only a lambda function. The lambda function will be built using Docker
, so be sure to have Docker
installed and the Docker
daemon running.
# cdk_docker_lambda/cdk_docker_lambda_stack.py
from aws_cdk import Stack
from aws_cdk import aws_lambda as _lambda
from constructs import Construct
class CdkDockerLambdaStack(Stack):
def __init__(self, scope: Construct, construct_id: str, **kwargs) -> None:
super().__init__(scope, construct_id, **kwargs)
self.build_lambda_func()
def build_lambda_func(self):
self.prediction_lambda = _lambda.DockerImageFunction(
scope=self,
id="ExampleDockerLambda",
# Function name on AWS
function_name="ExampleDockerLambda",
# Use aws_cdk.aws_lambda.DockerImageCode.from_image_asset to build
# a docker image on deployment
code=_lambda.DockerImageCode.from_image_asset(
# Directory relative to where you execute cdk deploy
# contains a Dockerfile with build instructions
directory="cdk_docker_lambda/ExampleDockerLambda"
),
)
Lambda function
Create a new directory in cdk_docker_lambda
called ExampleDockerLambda
. Here we are going to put a Dockerfile
, requirements.txt
which holds our function’s dependencies, and the lambda function itself, example_docker_lambda.py
cdk_docker_lambda/ExampleDockerLambda/Dockerfile
FROM amazon/aws-lambda-python:latest
LABEL maintainer="Wesley Cheek"
# Installs python, removes cache file to make things smaller
RUN yum update -y && \
yum install -y python3 python3-dev python3-pip gcc && \
rm -Rf /var/cache/yum
# Copies requirements.txt file into the container
COPY requirements.txt ./
# Installs dependencies found in your requirements.txt file
RUN pip install -r requirements.txt
# Be sure to copy over the function itself!
# Goes last to take advantage of Docker caching.
COPY example_docker_lambda.py ./
# Points to the handler function of your lambda function
CMD ["example_docker_lambda.handler"]
cdk_docker_lambda/ExampleDockerLambda/requirements.txt
requests
cdk_docker_lambda/ExampleDockerLambda/example_docker_lambda.py
# Very simple
import requests
def handler(event, context):
return "Hello Lambda!"
Now cdk deploy
. AWS CDK
will build your new Lambda function using Docker
and then push it for you to the ECR
repository that was originally created for you by running cdk bootstrap
during the CDK setup. How convenient.
After the image is built and pushed, CDK will deploy the necessary infrastructure. You can navigate to the AWS CloudFormation
console to view the deployment. It should only take a couple minutes.
Once finished, you will find your beautiful Docker
deployed Lambda function on the Lambda console
Test your Lambda function
We can use any kind of event since the function always just returns a string.
Have fun easily deploying any sized Lambda you’d like using AWS CDK
and Docker
! Make sure to cdk destroy
when you're done to avoid any charges.
Top comments (3)
Thanks for this I found it really helpful. One thing to note, in the Docker file the .py is copied on before the requirements and pip install. The result will be that every time the python file is changed and you debug docker will rebuild the image. Docker has some very clever caching, so copy the python file last!
Ah that's a good tip! Thanks - I've added your suggestion to the example
Dockerfile
.Very straightforward and simple, thanks for this!