Skip to content
Reference architecture of Amazon EKS cloudformation templates and AWS CDK samples
Makefile Python TypeScript JavaScript
Branch: master
Clone or download

Amazon EKS with AWS Cloudformation and AWS CDK

Reference architecture of Amazon EKS cloudformation templates and AWS CDK samples

This project aims to help you provision a ready-to-use Amazon EKS cluster by simply launching a cloudformation template with nested stacks or simply cdk deploy with AWS CDK with our provided CDK samples.


  • Creates both Amazon EKS cluster and NodeGroup in a single cloudformatoin template with nested stacks.
  • Abstracts away the CLI control in the Makefile - simply make create-eks-cluster, make update-eks-cluster and make delete-eks-cluster. That's all.
  • Fully support the latest Autoscaling Group features to hybrid on-demand and spot instances with mixed types and purchase options.
  • The cloudformation stack will help you automate the configuration on aws-auth-cm ConfigMap with AWS Lambda-backed custom resource.
  • No need to provision SpotFleet anymore.
  • On-demand instances will have node label lifecycle=OnDemand
  • Spot instances will have node label lifecycle=Ec2Spot and a spotInstance=true:PreferNoSchedule taint
  • NVIDIAGPU=1 node label will be created for all GPU instances on nodegroup creation
  • support passing ExtraNodeLabels(#3)
  • Support private subnets
  • Support non-RFC1918 IP/CIDR VPC subnets
  • Select the latest EKS-optimized AMI for Amazon Linux from a Lambda-backed cloudformaiton custom resource. This help you always use the latest EKS-Optimized AMI.
  • Support awslabs/amazon-eks-serverless-drainer to help you drain the pods on terminating spot instances to protect your online workload. Just pass EnableNodeDrainer=yes to the make command to install the plug-in straight from SAR.
  • Support the latest Amazon EC2 Instance Connect with web SSH capabilities(details)


Create AmazonEKSAdminRole IAM Role

Create a AmazonEKSAdminRole IAM Role manually and we will use thie role to

  1. deploy the cloudformaiton stacks
  2. execute the Lambda function as custom resource to help you configure the aws-auth ConfigMap so the nodes in the nodegroup can register themselves to the control plane.
  3. kubecl will call Amazon EKS control plane as this IAM role for RBAC Auth.
$ aws iam create-role --role-name AmazonEKSAdminRole --assume-role-policy-document file://assume-role-policy.json
$ aws iam attach-role-policy --role-name AmazonEKSAdminRole --policy-arn arn:aws:iam::aws:policy/AmazonEC2FullAccess
$ aws iam attach-role-policy --role-name AmazonEKSAdminRole --policy-arn arn:aws:iam::aws:policy/service-role/AWSLambdaBasicExecutionRole
$ aws iam put-role-policy --role-name AmazonEKSAdminRole --policy-name EKSAdminExtraPolicies --policy-document file://eks-admin-iam-policy.json

get the role arn string. We will use the arn later.

$ aws iam get-role --role-name AmazonEKSAdminRole --query 'Role.Arn' --output text


Create a file as your configuration file like this and leave Makefile untouched.

SSH_KEY_NAME ?= 'aws-pahud'
EKS_ADMIN_ROLE ?= arn:aws:iam::903779448426:role/AmazonEKSAdminRole
REGION ?= us-west-2 
VPC_ID ?= vpc-c7635ea0 
SUBNET1 ?= subnet-0f247a2b289129708
SUBNET2 ?= subnet-0da5823de42ae81f9
SUBNET3 ?= subnet-02858661648d4840a
InstanceTypesOverride ?= 't3.medium,m4.large,m5.large'
OnDemandBaseCapacity ?= 2
NodeAutoScalingGroupMinSize ?= 10 
NodeAutoScalingGroupDesiredSize ?= 12
NodeAutoScalingGroupMaxSize ?= 20
NodeVolumeSize ?= 50

The parameters in file will override the content of Makefile for customization.

  1. CLUSTER_STACK_NAME - the stack name and cluster name. default: eksdemo
  2. SSH_KEY_NAME - Your existing SSH keypair name in AWS EC2 console. default: aws-pahud (please update this value)
  3. REGION - The region code to deploy your EKS cluster. default: ap-northeast-1
  4. VPC_ID - The VPC ID to deploy your nodegroup
  5. EKS_ADMIN_ROLE - The AmazonEKSAdminRole arn described above.
  6. SUBNET1 The 1st subnet ID to deploy your nodegroup
  7. SUBNET2 The 2nd subnet ID to deploy your nodegroup
  8. SUBNET3 The 3rd subnet ID to deploy your nodegroup
  9. InstanceTypesOverride Instance types for your ASG
  10. OnDemandBaseCapacity On-Demand base capacity
  11. NodeAutoScalingGroupMinSize ASG min size
  12. NodeAutoScalingGroupDesiredSize ASG desired size
  13. NodeAutoScalingGroupMaxSize ASG max size
  14. NodeVolumeSize Default node volume size

OK. Let's create the complete Amazon EKS cluster and nodegroup

$ make create-eks-cluster

You may override the default values like this

$ REGION=ap-northeast-1 EKS_ADMIN_ROLE=arn:aws:iam::903779448426:role/AmazonEKSAdminRole CLUSTER_STACK_NAME=eksdemo10 make create-eks-cluster

or if you intend to run your nodegroup in private subnets and disable the auto-assign-public-ip completely for your nodes.

$ ASGAutoAssignPublicIp=no make create-eks-cluster

To specify a specific Amazon EKS cluster version(1.11 , 1.12 or 1.13):

$ ClusterVersion=1.13 make create-eks-cluster    

(if you don't specify ClusterVersion, it will create the latest version for you)


    "StackId": "arn:aws:cloudformation:ap-northeast-1:903779448426:stack/eksdemo10/b3ebf5c0-3395-11e9-bfb3-0a4b1943673a"

Behind the scene, a cloudformation stack with nested stacks will be created:

  1. eksdemo - the primary stack
  2. eksdemo-CL- - the control plane and security group
  3. eksdemo-AMI- - EKS-Optimized AMI info for Amazon EKS
  4. eksdemo-NG- - the nodegroup
  5. eksdemo-CM- - the custom resource to update the aws-auth ConfigMap
  6. eksdemo-CM-xxx- - the custom resource lambda function as AWS::Serverless::Application resource from SAR(Serverless Application Repository)
  7. eksdemo-Drainer-* - the eks-lambda-drainer app installed from SAR public repo.

Test and Validate

Now cloudformation stack is created. The Amazon EKS cluster will only be able to administratered via kubectl as AmazonEKSAdminRole IAM role. However, according to our assume-role-policy.json, only the following identities are allowed to assume to this role:

  1. Lambda service(
  2. Cloudformation service(

We need to grant our current IAM identity to assume this role(i.e. AmazonEKSAdminRole)

Let's check our current identity

$ aws sts get-caller-identity
    "Account": "903779448426", 
    "UserId": "AIDAJQENSMB5TSS54VEB2", 
    "Arn": "arn:aws:iam::903779448426:user/pahud"

(please note your Account and Arn string would be different from mine)

Let's edit assume-role-policy.json file from the local repo:

  "Version": "2012-10-17",
  "Statement": [
      "Effect": "Allow",
      "Principal": {
        "Service": ""
      "Action": "sts:AssumeRole"
      "Sid": "",
      "Effect": "Allow",
      "Principal": {
        "Service": ""
      "Action": "sts:AssumeRole"
      "Effect": "Allow",
      "Principal": {
        "AWS": "arn:aws:iam::903779448426:root"
      "Action": "sts:AssumeRole"

This will allow all IAM user from AWS Account ID 903779448426 be able to assume this role.

If you prefer to restrict to a single IAM user, for example pahud:

"Principal": {
    "AWS": "arn:aws:iam::903779448426:user/pahud"

And of course you can specify multiple IAM users in Principal

"Principal": {
    "AWS": "arn:aws:iam::903779448426:user/pahud",
    "AWS": "arn:aws:iam::903779448426:user/emilie"

OK let's update the assume role policy

$ aws iam update-assume-role-policy --role-name AmazonEKSAdminRole --policy-document file://assume-role-policy.json

Try assume this role with aws assume-role like this

$ aws sts assume-role --role-arn arn:aws:iam::903779448426:role/AmazonEKSAdminRole --role-session-name test
    "AssumedRoleUser": {
        "AssumedRoleId": "AROAJCL4US4TE3MXZM272:test", 
        "Arn": "arn:aws:sts::903779448426:assumed-role/AmazonEKSAdminRole/test"
    "Credentials": {
        "SecretAccessKey": "...", 
        "SessionToken": "...", 
        "Expiration": "2019-02-20T08:19:58Z", 
        "AccessKeyId": "..."

If you get the response like this then you are allowed to assume role to AmazonEKSAdminRole.

Download required binaries

download the latest kubectl binary

For example, in Linux

$ curl -o kubectl
$ chmod +x $_
$ sudo mv $_ /usr/local/bin/

check Amazon EKS document about install kubectl and getting started and download the two binaries of latest version.

Generate or update kubeconfig

run update-kubeconfig

$ aws --region ap-northeast-1 eks update-kubeconfig --name eksdemo --role-arn arn:aws:iam::903779448426:role/AmazonEKSAdminRole


Updated context arn:aws:eks:ap-southeast-1:903779448426:cluster/eksdemo in /home/ec2-user/.kube/config

try list the nodes

$ kubectl get no
NAME                                          STATUS   ROLES    AGE   VERSION   Ready    <none>   1m   v1.13.7-eks-c57ff8   Ready    <none>   1m   v1.13.7-eks-c57ff8

Your cluster is ready now.

Update the cluster

update from 1.12 to 1.13

$ ClusterVersion=1.13 make update-eks-cluster

update from 1.11 to 1.12

$ ClusterVersion=1.12 make update-eks-cluster

Node Labels, Taints and Tolerations

By default, all the on-demand instances will have lifecycle=OnDemand label while spot instances will have lifecycle=Ec2Spot. Use the node selector to better schedule your workload

Additionally, all the spot instances have a spotInstance=true:PreferNoSchedule taint. To deploy your Pod on spot instances, use the node label selector to specify lifecycle=Ec2Spot, otherwise the pod will not be scheduled on the spot instances unless it has relevant toleration. (Taint and Toleration in Kubernetes).

Nodegroup rolling update with the latest AMI

From time to time you may need to update your nodegroup with the latst Amazon EKS-optimized AMI, especially when there's a new security patch like this and a new AMI is released. No bother to specify the latest AMI ID - just kick off the nodegroup rolling update with the latest Amazon EKS-optimized AMI.

# force AMI check and update(if any)
$ UPDATE_AMI=1 make update-eks-cluster

Spot Termination Handling

To enable the awslabs/amazon-eks-serverless-drainer support as the plug-in and install the drainer from SAR, just pass EnableNodeDrainer=yes to the make command. e.g.

$ EnableNodeDrainer=yes make create-eks-cluster

A Drainer cloudformation nested stack will be created and install the eks-lambda-drainer from SAR for you. All the spot instance termination will be watched and handled by the provided Lambda function. Easy peasy!

You can’t perform that action at this time.