Welcome to Serverless Endpoints Microservice

The purpose of this microservice is to create backend endpoints without ever having to maintain a server (EC2). By connecting API Gateway to Lambda, we create an on-demand, scalable, and cheap (pay-as-you-go) backend.

Architecture

The architecture diagram looks more intense in the picture than in real life. What it actually entails is an API Gateway triggering a Lambda. The Lambda is a Docker image built with Python 3.9 and FastAPI (and Mangum). The microservice is deployed with CDK, which translates to a Cloudformation template.

For observability, you can inspect the Lambda's Cloudwatch logs: runtime duration, failures, and count of endpoint hits. If you are fancy, you can add metrics & alarms to the Lambda (and API Gateway). For the business/operations/SRE team, you can add New Relic to the Lambda such that there will be "single pane of glass" for 24/7 monitoring. You can also inspect the API Gateway's dashboard.

Miscellaneous details:

• In this branch BundlingOptions, I installed aws-lambda-powertools as part of requirements.txt. In the NoBundlingOptions branch, I decided to attached aws-lambda-powertools as a Lambda Layer instead because the core code would stay below 3 MB, which allows you to still have inline editing in the Lambda console.
• Another benefit is that the BundlingOptions specifies the Docker image for Python 3.9 and then installs the Lambda's requirements.txt and thus ensures Python compatibility. The NoBundlingOptions branch installs requirements.txt using the Python version on your machine that you use to cdk deploy, which might not be exactly Python 3.9 and thus might install dependencies that are not compatible with the Lambda's runtime Python version.
• cdk.json is basically the config file. I specified to deploy this microservice to us-west-1 (California). You can change this to your region of choice.
• The following is the AWS resources deployed by CDK and thus Cloudformation:

  

• Suppose you are worried about the "cold start" problem of having the Lambda start up for the first time (which my API response is 1.2 seconds). Suppose you only want "warm starts" (which my API response is 0.2 seconds). There are 2 ways around it: you can have Eventbridge scheduled events regularly trigger the Lambda to keep it warm. Or you can configure the Lambda to have a provisioned concurrency (so the Lambda is always on) but costs more money.

Deploying the Microservice Yourself

$ python -m venv .venv
$ source .venv/bin/activate
$ python -m pip install -r requirements.txt
$ cdk deploy  # Docker daemon must be running; also assumes AWS CLI is configured + npm installed with `aws-cdk`: detailed instructions at https://cdkworkshop.com/15-prerequisites.html