AWS-Lambda + Python + Terraform example

Deploy your Python app in minutes!

This is a template for rapid deploy of Python web services (FastAPI, Flask, etc) on AWS-Lambda, automated by Terraform.

Requirements

• Python +3.7 and python3-venv

sudo apt-get install python3-venv

• AWS-cli, (check AWS cli intall docs), and IAM credentials

  curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip"
  unzip awscliv2.zip
  sudo ./aws/install

• Terraform

Hello, world!

1. Clone this repository, open a terminal in the repository root

git clone [email protected]:ieferrari/aws-lambda-python-terraform-example.git
cd  aws-lambda-python-terraform-example

2. Crete the env files (make sure to have every other environment deactivated):

chmod +x ./create_env.sh
./create_env.sh

3. upload to AWS-lambda (may apply charges)

cd ./infra
terraform init
chmod +x ./upload.sh
./upload.sh

4. you will see a prompt to confirm the Terraform plan, if everything looks good, type "yes". If everything went right, after a couple of minutes you will see this:

Testing API endpoint
...
{"Hello":"World!!!"}

5. When you finish, your testing, remember to clean up and delete the infrastructure running:

terraform destroy

---

Troubleshooting

Depending on your Terraform, you may need to run terraform init multiple timer until you get the Terraform has been successfully initialized! message, for the first time. Then you can run upload.sh normally.

---

Quick Deploy for your project

1. Replace /app with the code of your Python App, or change the value of APP_DIR in upload.sh to make it point to the path of your app.

• this example uses FastAPI, is almost the same for Flask an other micro-frameworks.
• Remember to import Mangum and create a handler for the main function. this is going to be used by Lambda.

2. In /env create a folder for the virtual environment with the requirements of your Python app, or change the value of PACKAGES_DIR in upload.sh.

• you can replace the requirements.txt in the root folder and run create_env.sh to create the content of ./env

4. setup your system for automatic deploy:

• install AWS-cli
• create and save IAM credentials
• install terraform

5. Deploy:

• edit the variable.tf file for custom names
• run upload.sh

Each step is explained with more details in the following sections.

.
├── app
|   └── ...
├── env
|   └── ...
├── temp
|   └── ...
├── temp
|   └── ...
├── main.tf
├── upload.sh
└── variables.tf

---

Lambda Functions on AWS

Basic setup:

• Python code:
  • simple code on aws-console or a zip file uploaded to S3 for bigger apps
  • every library outside the standard library must be included in the zip file
  • you need a handler for Lambda, in this case we use Mangum
• API gateway to trigger the lambda function
  • gateway URL used as a proxy to pass requests to the app router
  • permissions for gateway: in this case will be public.

Server-less architecture drawbacks

Generally speaking there are three main disadvantages of using lambda functions:

1. cold startups, with something like 1 second of latency when the function hasn't been used in more than 30 min (rough estimation)
2. it can became more expensive than EC2 beyond certain usage level
3. vendor lock, specially with very fragmented design patterns that uses a specific api gateway for each function.

Solutions :

1. To avoid cold startup, there are many ways to warm up your function, like ping it every half hour, (if it is actually necessary )
2. If your app grows a lot, at some point you may want to migrate to EC2. In this example we are going to set everything up to be prepared for that moment (check this article).
3. To avoid vendor lock and to be ready to migrate, a good option is to mount a group of function inside one app, using FastAPI, Flask or other similar framework, with only one API gateway working as a proxy, and your app managing the routes for different calls. This way you can migrate your whole app in one single step.

Other option is to create a container for your app, and run the container inside lambda. AFAIK is a bit faster to avoid the container and use it only if you want to migrate, anyway all the compatibility issues are solved by virtual environment and the right Python version inside Lambda

---

Python apps for lambda

create virtual env and install requirements for your app, go to the root folder of the repository and run (make sure to deactivate every other repository first):

python3 -m venv ./env
source ./env/bin/activate
pip install -r ./app/requirements.txt

# The same is done by running **_create_env.sh_**

Create a zip file with all the content of /env/lib/Python3.7/sit-packages/ copied at the same level of the main file of the project, this way Python can call every library without any special path.

In this case we copy the content of /app and /env/lib/Python3.7/sit-packages/ to /temp/, then you can create a zip file to upload to S3

APP_DIR=./app/*
PACKAGES_DIR=./env/lib/python3.7/site-packages/*
TEMPORARY_DIR=./temp
# change the DIR variables to address the custom path of your project.

mkdir $TEMPORARY_DIR
cp -r $PACKAGES_DIR $TEMPORARY_DIR
cp -r $APP_DIR $TEMPORARY_DIR

# The same is done when you run **_upload.sh_**

Remember to add a handler for Lambda, in this case, we use Mangum. There are other options, like chalice, the aws-Python-micro-web-framework, AFAIK there is no advantage over FastAPI+Mangum, and if you use chalice you may want to consider the vendor lock risk.

from mangum import Mangum
from fastapi import FastAPI
[...]

app = FastAPI()
[...]

# create a handler for Lambda, this is the entry point to call the app
handler = Mangum(app)

check the usage example of Mangum in /app/hello_lambda.py

---

Terraform

see

check the comments inside main.tf for more details about the terraform implementation

Basic commands:

terraform init

terraform apply

terraform destroy

Terraform output

You can use the output for other automation tasks or scripts:

$ terraform output -raw deployment_invoke_url
https://example.url/for/yor/api/endoint

or get a JSON outpu:

$ terraform output -ra

$ terraform output -json
{
  "deployment_invoke_url": {
    "sensitive": false,
    "type": "string",
    "value": "https://4yr90px2ua.execute-api.sa-east-1.amazonaws.com/test"
  },
  "lambda": {
    "sensitive": true,
    "type": "string",
    "value": "arn:aws:lambda:sa-east-1:189743987374:function:my_lambda_NAME:$LATEST"
  }
}

For example, to test your API, run:

curl $(terraform output -raw deployment_invoke_url)

#  the same is done at the end of **upload.sh** to test the API

---

Next steps

This is a fast example to deploy your python app. It is a good start, but depending on the architecture of your project you may want to organize the infrastructure in some specific way.

One improvement, is to separe concerns in terraform to deploy only some parts of the infractructure. For example when you modify the code but you want to preserve the same api gateway. Check this article