Serverless redirect with CloudFront Functions

In a project I started working on some time ago I needed an easy way to do redirects from one domain, example.com to a different domain, example-2.com. Normally I would try and use DNS as far as possible, but when I needed to handle example.com/abc and redirect that to example-2.com/abc there was a problem that needed to be solved. The original example.com (from here on referred to as site A) was hosted as a static website, classic S3 and CloudFront. Example-2.com was on a completely different setup (from here on referred to as site B). I could of course add folders in S3 for /abc with a single index.html file with some redirect javascript, but that presents other problems as CloudFront is an CDN and would not automatically fetch index.html from the /abc path. I also needed an easy way to add and remove mappings between site A and B, redirecting siteA/abc to siteB/xyz.

CloudFront functions and the new KeyValueStore to the rescue!!

Architecture overview

There are two parts to this overall architecture, the actual redirect part in CloudFront and a simple API to add and remove mapping keys, powered by ApiGateway and StepFunctions.

To get into some more details, the call flow when a user is redirected would be.

Let's start to build this solution, but before we start....

KeyValueStore limitations

When starting to work with CloudFront KeyValueStores there are some things you need to understand.

The store has the following limits:

• Total size – 5 MB
• Key size – 512 characters
• Value size – 1024 characters

That means that you need to keep your KeyValueStored trimmed, you can't create a massive amount of key value pairs and you can't store massive amount of data, you can't use it as a database. In a normal use-case these limits should not be a problem.

When working with a KeyValueStore you not only need the ARN of the store, you also need the ETag representing the last version of the store. This is important to remember when adding and removing keys.

It's also important to understand that updates are eventual consistent, it does take a couple of seconds for a change to replicate across edge locations.

Create the store

Before we even start doing anything we need to create a KeyValueStore. Navigate to CloudFront section of the Console. The KeyValueStore is well hidden under Functions.

Just click on the Create KeyValueStore and fill in the form.

Or deploy this CloudFormation Template to do it in an automatic way.

AWSTemplateFormatVersion: "2010-09-09"
Transform: "AWS::Serverless-2016-10-31"
Description: Creates a serverless url redirect
Parameters:
  Environment:
    Type: String
    Description: Environment type, dev, test, prod
    AllowedValues:
      - dev
      - test
      - prod
  ApplicationName:
    Type: String
    Description: The application that owns this setup.

Resources:
  RedirectKeyValueStore:
    Type: AWS::CloudFront::KeyValueStore
    Properties:
      Comment: !Sub Key Value Store for the ${ApplicationName} ${Environment}
      Name: redirect-urls

We will add more and more resources to this template as we go.

Management API

As said, I need a easy way to create, update, and delete redirect mappings. So we create a API Gateway HTTP Api and integrate that with two StepFunctions to create and remove mappings.

This StepFunction will only have two states, and I use the recently released SDK integration. The first thing we must do is to call DescribeKeyValueStore, we need to get the Etag so we can modify the KeyValueStore.

In the second state, here PutKey we use the ETag returned by describe and the Key and Value that will be sent as data from our API call.

We can create the StepFunctions from the visual editor in the Console, we need one for delete and one to put. Or we add additional resources to our CloudFormation template and deploy that.

PutKeyStateMachineLogGroup:
    Type: AWS::Logs::LogGroup
    Properties:
      LogGroupName: !Sub "${ApplicationName}/putstatemachine"
      RetentionInDays: 1

  PutKeyStateMachine:
    Type: AWS::Serverless::StateMachine
    Properties:
      DefinitionUri: statemachine/put-key.asl.yaml
      Tracing:
        Enabled: true
      Logging:
        Level: ALL
        IncludeExecutionData: True
        Destinations:
          - CloudWatchLogsLogGroup:
              LogGroupArn: !GetAtt PutKeyStateMachineLogGroup.Arn
      DefinitionSubstitutions:
        KvsArn: !GetAtt RedirectKeyValueStore.Arn
      Policies:
        - Statement:
            - Effect: Allow
              Action:
                - logs:*
              Resource: !GetAtt PutKeyStateMachineLogGroup.Arn
        - Statement:
            - Effect: Allow
              Action:
                - cloudfront-keyvaluestore:DescribeKeyValueStore
                - cloudfront-keyvaluestore:PutKey
              Resource: !GetAtt RedirectKeyValueStore.Arn
      Type: EXPRESS

  DeleteKeyStateMachineLogGroup:
    Type: AWS::Logs::LogGroup
    Properties:
      LogGroupName: !Sub "${ApplicationName}/deletestatemachine"
      RetentionInDays: 1

  DeleteKeyStateMachine:
    Type: AWS::Serverless::StateMachine
    Properties:
      DefinitionUri: statemachine/delete-key.asl.yaml
      Tracing:
        Enabled: true
      Logging:
        Level: ALL
        IncludeExecutionData: True
        Destinations:
          - CloudWatchLogsLogGroup:
              LogGroupArn: !GetAtt DeleteKeyStateMachineLogGroup.Arn
      DefinitionSubstitutions:
        KvsArn: !GetAtt RedirectKeyValueStore.Arn
      Policies:
        - Statement:
            - Effect: Allow
              Action:
                - logs:*
              Resource: !GetAtt DeleteKeyStateMachineLogGroup.Arn
        - Statement:
            - Effect: Allow
              Action:
                - cloudfront-keyvaluestore:DescribeKeyValueStore
                - cloudfront-keyvaluestore:DeleteKey
              Resource: !GetAtt RedirectKeyValueStore.Arn
      Type: EXPRESS

This is the two ASL definitions used to create the actual StepFuntions.

Comment: Put a new or update an existing key in a KeyValueStore
StartAt: DescribeKeyValueStore
States:
    DescribeKeyValueStore:
        Type: Task
        Parameters:
            KvsARN: ${KvsArn}
        Resource: arn:aws:states:::aws-sdk:cloudfrontkeyvaluestore:describeKeyValueStore
        ResultPath: $.DescribeResult
        Next: PutKey
    PutKey:
        Type: Task
        Parameters:
            IfMatch.$: $.DescribeResult.ETag
            KvsARN.$: $.DescribeResult.KvsARN
            Key.$: $.Key
            Value.$: $.Value
        Resource: arn:aws:states:::aws-sdk:cloudfrontkeyvaluestore:putKey
        End: true

Comment: Delete an existing key in a KeyValueStore
StartAt: DescribeKeyValueStore
States:
    DescribeKeyValueStore:
        Type: Task
        Parameters:
            KvsARN: ${KvsArn}
        Resource: arn:aws:states:::aws-sdk:cloudfrontkeyvaluestore:describeKeyValueStore
        ResultPath: $.DescribeResult
        Next: DeleteKey
    DeleteKey:
        Type: Task
        End: true
        Parameters:
            IfMatch.$: $.DescribeResult.ETag
            KvsARN.$: $.DescribeResult.KvsARN
            Key.$: $.Key
        Resource: arn:aws:states:::aws-sdk:cloudfrontkeyvaluestore:deleteKey

But, we need a way to invoke the StepFunctions, so for that we create an API Gateway with a Post method and a Delete method and integrate that with our two StepFunctions.

Keep add resources to the template and deploy it, or create it manually from the console.

  HttpApi:
    Type: AWS::Serverless::HttpApi
    Properties:
      DefinitionBody:
        Fn::Transform:
          Name: AWS::Include
          Parameters:
            Location: api/api.yaml

  HttpApiRole:
    Type: AWS::IAM::Role
    Properties:
      AssumeRolePolicyDocument:
        Version: 2012-10-17
        Statement:
          - Effect: Allow
            Principal:
              Service: apigateway.amazonaws.com
            Action:
              - sts:AssumeRole
      Policies:
        - PolicyName: ApiDirectInvokeStepFunctions
          PolicyDocument:
            Version: 2012-10-17
            Statement:
              Action:
                - states:StartSyncExecution
              Effect: Allow
              Resource:
                - !Ref PutKeyStateMachine
                - !Ref DeleteKeyStateMachine

The actual API definition is in an Open API specification.

openapi: "3.0.1"
info:
  title: "update-cloudfront-key-value-store"
paths:
  /keyvalue:
    post:
      responses:
        default:
          description: "Default response"
      x-amazon-apigateway-integration:
        integrationSubtype: "StepFunctions-StartSyncExecution"
        credentials:
          Fn::GetAtt: [HttpApiRole, Arn]
        requestParameters:
          Input: "$request.body"
          StateMachineArn:
            Fn::GetAtt: [PutKeyStateMachine, Arn]
        payloadFormatVersion: "1.0"
        type: "aws_proxy"
        connectionType: "INTERNET"
    delete:
      responses:
        default:
          description: "Default response"
      x-amazon-apigateway-integration:
        integrationSubtype: "StepFunctions-StartSyncExecution"
        credentials:
          Fn::GetAtt: [HttpApiRole, Arn]
        requestParameters:
          Input: "$request.body"
          StateMachineArn:
            Fn::GetAtt: [DeleteKeyStateMachine, Arn]
        payloadFormatVersion: "1.0"
        type: "aws_proxy"
        connectionType: "INTERNET"
x-amazon-apigateway-importexport-version: "1.0"

Now we should have an API and we can test to invoke it from Postman and there should be new values added to the KeyValueStore.

Start by checking the KeyValueStore which should show empty key value pairs.

Calling the API from Postman, we supply the ket value pair in the body of the call.

After a call to the API we should now see the value in the key value store.

With that we have a simple API in place to add new values.

CloudFront setup

Now we have all the bits and pieces in place to create our CloudFront distribution and CloudFront function. There are a couple of things to understand about Cloudfront functions. There are not as broad support as in Lambda@Edge. To fully understand the limitation I recommend that you read some of my other posts. Protecting a Static Website with JWT and Lambda@Edge or Migrating from Lambda@Edge to CloudFront Functions there is also a very good post by AWS Introducing CloudFront Functions – Run Your Code at the Edge with Low Latency at Any Scale.

The code for our function is not that complicated and looks something like this. The function split the URL and use that as the key. It will then either redirect to the mapped value from the KeyValueStore or if the key doesn't exists, it will redirect to our base url for Site B.

import cf from 'cloudfront';

const kvsId = '${RedirectKeyValueStore.Id}';
const kvsHandle = cf.kvs(kvsId);

async function handler(event) {
    const request = event.request;
    const headers = request.headers;
    const key = request.uri.split('/')[1]
    let base = "${BaseSiteUrl}";
    let value = ""; // Default value
    try {
        value = await kvsHandle.get(key);
    } catch (err) {
        console.log(`Kvs key lookup failed.`);
    }

    let newurl = base + value
    const response = {
        statusCode: 302,
        statusDescription: 'Found',
        headers:
            { "location": { "value": newurl } }
        }

    return response;
}

So we add the function to out template. Unfortunately we must have the code inline in the template, it's not pretty but at the same time it forces us to keep our functions short.

  RedirectFunction:
    Type: AWS::CloudFront::Function
    Properties:
      AutoPublish: true
      FunctionCode: !Sub |
        import cf from 'cloudfront';

        const kvsId = '${RedirectKeyValueStore.Id}';
        const kvsHandle = cf.kvs(kvsId);

        async function handler(event) {
            const request = event.request;
            const headers = request.headers;
            const key = request.uri.split('/')[1]
            let base = "${BaseSiteUrl}";
            let value = ""; // Default value
            try {
                value = await kvsHandle.get(key);
            } catch (err) {
                console.log(`Kvs key lookup failed.`);
            }

            let newurl = base + value
            const response = {
                statusCode: 302,
                statusDescription: 'Found',
                headers:
                    { "location": { "value": newurl } }
                }

            return response;
                
        }
      FunctionConfig:
        Comment: Function for url redirect
        KeyValueStoreAssociations:
          - KeyValueStoreARN: !GetAtt RedirectKeyValueStore.Arn
        Runtime: cloudfront-js-2.0
      Name: !Sub ${ApplicationName}-${Environment}-redirect-function

We associate the function with the KeyValueStore in the field KeyValueStoreAssociations. What is important to remember is that a function MUST be associated with the KeyValueStore to be able to read it. A function can also only be associated with ONE KeyValueStore. A KeyValueStore can however be associated with several functions.

Finally, we are at a point where we can add our CloudFront distribution and set the CloudFront function to be invoked by the viewer request. Before we continue we need to look at the integration points.

CloudFront integration points

There are four different integration points for Lambda@Edge and you can only set one Lambda function as target for each. The integration is done on the cache behavior so if you have several cache behaviors you can setup integration with different functions for each of them. So how does each integration point work?

Viewer Request The Lambda function is invoked as soon as CloudFront receives the call from the client. This function will be invoked on every request.
Origin Request The Lambda function is invoked after the cache and before CloudFront calls the origin. The function will only be invoked if there is a cache miss.
Origin Response The Lambda function is invoked after CloudFront receives the response from the origin and before data is stored in the cache.
Viewer Response The Lambda function is invoked before CloudFront forwards the response to the client.

However for CloudFront functions, the only available integration points are viewer request and viewer response. We will use the viewer request integration point.

Create CloudFront distribution

When creating the CloudFront distribution I only create a fake origin which only purpose is to integrate with the CloudFront Function. In our use case the function will always return an redirect and we'll never hit the origin.

  CloudFrontDistribution:
    Type: AWS::CloudFront::Distribution
    Properties:
      DistributionConfig:
        Aliases:
          - !Ref DomainName
        Comment: !Sub Distribution for the ${ApplicationName} ${Environment}
        CustomErrorResponses:
          - ErrorCode: 403
            ResponseCode: 200
            ResponsePagePath: "/404.html"
        DefaultCacheBehavior:
          AllowedMethods:
            - "GET"
            - "HEAD"
            - "OPTIONS"
          Compress: False
          CachePolicyId: 4135ea2d-6df8-44a3-9df3-4b5a84be39ad #Managed Cache Policy 'CachingDisabled'
          FunctionAssociations:
            - EventType: viewer-request
              FunctionARN: !GetAtt RedirectFunction.FunctionMetadata.FunctionARN
          TargetOriginId: function-redirect-origin
          ViewerProtocolPolicy: redirect-to-https
        DefaultRootObject: index.html
        Enabled: True
        Origins:
          - DomainName: !Sub handle-redirect.${DomainName}
            Id: function-redirect-origin
            CustomOriginConfig:
              OriginProtocolPolicy: match-viewer
        PriceClass: PriceClass_100
        ViewerCertificate:
          AcmCertificateArn: !Ref SSLCertificateArn
          SslSupportMethod: sni-only
      Tags:
        - Key: Application
          Value: !Ref ApplicationName
        - Key: Name
          Value: !Sub ${ApplicationName}-${Environment}

With that part deployed our solution is complete. All access to Site A will now be redirected to Site B. Either we redirect to a mapped url found in the KeyValueStore or we redirect to the top of Site B.

Final Words

In this post I showed a way to use CloudFront Functions with KeyValueStore to do redirects between different domains. The KeyValueStore is a very useful part of CloudFront that can be used for different use-case. I have been using CloudFront Functions for other solutions and I have often needed a simple way to store environment variables. The solution presented in this blog could easily be turned into a solution to create, update, and delete exactly that.

I also want to give a shout out to Elias Brange, a good friend that one week before me presented a different useful solution using a similar setup. Elias creates and deploys his solution using SST giving you a different approach to your IaC. We had some good laughs as we wrote similar blogs without knowing it. Hat of to Elias that managed to publish before me.

Don't forget to follow me on LinkedIn and X for more content, and read rest of my Blogs