@wheatstalk/cdk-ecs-keycloak

This CDK construct allows you to spin up a high availability Keycloak cluster on AWS ECS using Fargate capacity.

Usage no npm install needed!

<script type="module">
  import wheatstalkCdkEcsKeycloak from 'https://cdn.skypack.dev/@wheatstalk/cdk-ecs-keycloak';
</script>

README

CDK-based Keycloak Cluster on ECS

This CDK construct allows you to spin up a high availability Keycloak cluster on AWS ECS using Fargate capacity.

What is Keycloak?

Keycloak is an open-source identity and access management server. With it, you can add authentication to your apps and secure services with little fuss.

But why Keycloak on ECS over Cognito?

  • Account linking doesn't break down in Keycloak when the user signed up with a social account and now wants to set a password
  • User-oriented account management console
  • More user federation, identity brokering, and social login features
  • Store and access the user's federated access tokens on other services

Examples

All defaults

The simplest example shows all defaults below.

import * as keycloak from '@wheatstalk/cdk-ecs-keycloak';
import * as cdk from '@aws-cdk/core';

// Create a CDK app and a stack
const app = new cdk.App();
const stack = new cdk.Stack(app, 'Keycloak-test');

// Add a Keycloak cluster to the stack
new keycloak.KeycloakCluster(stack, 'Keycloak');

The initial deployment takes a few minutes, after which a URL to Keycloak will be shown in the terminal. After a few minutes of settling, Keycloak will be available at that URL.

By default, all the following are created:

  • A VPC (public and private subnet with nat gateways)
  • An Internet-facing load balancer serving HTTP (not HTTPS) traffic
  • An ECS cluster with Keycloak running on it
  • An Aurora Serverless MySQL cluster

Auto-scaling with internal HTTPS and Keycloak configuration

To demonstrate more functionality, here is an example cluster that does the following:

  • Requests Fargate tasks sized 0.5vCPU / 1GB RAM
  • Controls ECS deployment min and max health settings
  • Sets up Infiniscan clustering by increasing cache owner count
  • Publishes an Application Load Balancer with internal HTTPS
  • The Application Load Balancer in this example upgrades HTTP connections to HTTPS
  • Sets up auto-scaling by interacting directly with the ECS service
// Create a Keycloak cluster on Fargate
const keycloakCluster = new keycloak.KeycloakCluster(this, 'Keycloak', {
  // Fargate task sizes
  cpu: 512,
  memoryLimitMiB: 1024,
  // Service options
  minHealthyPercent: 50,
  maxHealthyPercent: 200,
  keycloak: {
    // Set distributed inficaches owners to two
    cacheOwnersCount: 2,
  },
  // Use an HTTPS load balancer with internal HTTPS from the load balancer to Keycloak.
  httpsPortPublisher: keycloak.PortPublisher.httpsAlb({
    certificates: [certificate],
    // Redirect HTTP traffic to HTTPS
    upgradeHttp: true,
  }),
});

// Auto-scale the service
const autoScaling = keycloakCluster.service.autoScaleTaskCount({
  maxCapacity: 5,
  minCapacity: 3,
});

autoScaling.scaleOnCpuUtilization('Target40', {
  targetUtilizationPercent: 40,
  scaleInCooldown: cdk.Duration.minutes(30),
  scaleOutCooldown: cdk.Duration.minutes(10),
});

Use a database instance

You may opt to use a database instance instead of an Aurora Serverless cluster.

new keycloak.KeycloakCluster(this, 'Keycloak', {
  databaseProvider: keycloak.DatabaseProvider.databaseInstance({
    engine: rds.DatabaseInstanceEngine.mysql({
      version: rds.MysqlEngineVersion.VER_5_7,
    }),
    instanceType: ec2.InstanceType.of(ec2.InstanceClass.T3, ec2.InstanceSize.MICRO),
  }),
});

Use Postgres instead of MySQL

new keycloak.KeycloakCluster(this, 'Keycloak', {
  databaseProvider: keycloak.DatabaseProvider.databaseInstance({
    engine: rds.DatabaseInstanceEngine.postgres({
      version: rds.PostgresEngineVersion.VER_11_9,
    }),
    instanceType: ec2.InstanceType.of(ec2.InstanceClass.T2, ec2.InstanceSize.MICRO),
  }),
});

Provide your VPC, Database and Application Load Balancer

You may provide resources you've created, such as VPCs, Clusters, CloudMap namespaces, databases and load balancers.

In the following for example we re-use a VPC, database and Application Load Balancer listener. As a result the Cluster construct will:

  • Create a new ECS cluster in the given VPC
  • Use the given database information
  • Create an ingress rule in the database's security group
  • Publish its HTTP port on the given load balancer listener
// Your resources
const vpc = new ec2.Vpc(...);
const rdsDb = new rds.DatabaseInstance(...);
const loadBalancer = new elbv2.ApplicationLoadBalancer(...);
const listener = loadBalancer.addListener(...);

new keycloak.KeycloakCluster(this, 'Keycloak', {
  // Provide an existing VPC so the cluster and database can opt to reuse it
  vpcProvider: keycloak.VpcProvider.fromVpc(vpc),
  // Bring your own database
  databaseProvider: keycloak.DatabaseProvider.fromDatabaseInfo({
    // Provide an RDS-compatible secret with credentials and connection
    // info (required)
    credentials: rdsDb.secret!,
    // Inform Keycloak of the database vendor (required)
    vendor: keycloak.KeycloakDatabaseVendor.MYSQL,
    // Add an ingress rule to the database security group (optional as long
    // as the Keycloak tasks can connect to the database)
    connectable: rdsDb,
  }),
  // Bring your own load balancer
  httpPortPublisher: keycloak.PortPublisher.addTarget({
    // Your load balancer listener
    listener,
    // Only publish certain paths
    conditions: [elbv2.ListenerCondition.pathPatterns([
      '/auth/*',
    ])],
    // Set your listener rule priority
    priority: 1000,
  }),
});

Customize the container image

You may build and use a custom container image to add a custom theme, run scripts, add custom realm configs, or bundle your own Keycloak SPIs.

new keycloak.KeycloakCluster(this, 'Keycloak', {
  keycloak: {
    image: ecs.ContainerImage.fromAsset(pathToDockerBuildContext, {
      buildArgs: {
        FROM: 'jboss/keycloak:12.0.2',
      },
    }),
  },
});

Publish container ports through a Network Load Balancer

You may use the PortPublisher pattern to publish container ports through a Network Load Balancer.

new keycloak.KeycloakCluster(this, 'Keycloak', {
  // Publish the container's HTTP web port in a NLB on port 8080
  httpPortPublisher: keycloak.PortPublisher.nlb({
    port: 8080,
  }),
  // Publish the container's HTTPS port in an NLB on port 8443
  httpsPortPublisher: keycloak.PortPublisher.nlb({
    port: 8443,
    healthCheck: false,
  }),
  // Publish the Wildfly Admin Console on port 9990 (not recommended in
  // production)
  adminConsolePortPublisher: keycloak.PortPublisher.nlb({
    port: 9990,
    healthCheck: false,
  }),
});