9 Ways to Spin Up an EKS Cluster: Part 3 - eksctl

Eksctl is powerful command-line tool, developed and backed by AWS, streamlines the process of creating, managing, and scaling Kubernetes clusters on Amazon EKS (Elastic Kubernetes Service). With a user-friendly interface and robust functionality, eksctl eliminates the complexity traditionally associated with Kubernetes management.

An official, open-source command-line utility developed in Go, eksctl automates the creation, management, and operation of Amazon EKS clusters. It abstracts complex provisioning, like VPCs, IAM roles, and worker nodes, into simple commands or declarative YAML configurations.

In my previous post I showed how to spin up an EKS cluster with pure shell and AWS CLI.

This used to be the easiest way of getting to a cluster without leaving your terminal. But pretty early in EKS history (2017) some smart folks from a company named Weaveworks(RIP) realized it was too cumbersome to do this using the aws cli subcommand and that EKS is complex enough to deserve a command-line client of its own. That's how eksctl was born.

A few months ago Weaveworks (who brought us a plethora of great OSS tools like Flux, Flagger and Weave) was shut down. But AWS announced full support for eksctl in 2019 - so eksctl is now the de-facto standard EKS CLI tool.

To reflect this joint maintainership, the project has since moved out of the Weaveworks GitHub organization to a dedicated top-level organization at github.com/eksctl-io, which AWS and the community now maintain together. The old Weaveworks URL automatically redirects to the new location, so existing automation continues to work.

The great thing about eksctl is that it allows one to create and manage clusters not only using one-off commands with arguments but also with YAML configuration files - in a true and familiar IaC way.

Key features of eksctl include:

• Automated provisioning: Spin up a fully functional EKS cluster with managed EC2 or Fargate nodes using a single command.
• Declarative configs: Manage cluster configurations using YAML files for reproducible and version-controlled environments.
• Addon management: Easily install and manage VPC CNI, CoreDNS, and kube-proxy.
• Lifecycle operations: Handle node scaling, cluster upgrades, and cluster deletion effortlessly.

We'll check out both options but first let's install eksctl and generate an SSH key so we can connect to the nodes in the clusters we create if needed. Please note - I'm not endorsing SSH connections to your EKS nodes. Do avoid this if possible - so as not to cause inadvertent configuration drift. But sometimes we still need this for troubleshooting, especially in training environments. So let's have the SSH key handy.

This article is part of series "How to Spin Up an AWS Cluster".

Take a look at the different methods: 

Way 1 - Create an EKS Cluster in AWS Management Console

Way 2 - Create an EKS Cluster in AWS cli

Way 3 - Create an EKS Cluster with eksctl

Way 4 - Create an EKS Cluster with CloudFormation

Way 5 - Create an EKS Cluster with python and boto3

Way 6 - Create an EKS Cluster with AWS CDK

Way 7 - Create an EKS Cluster with Terraform

Way 8 - Create an EKS Cluster with Pulumi

Way 9 - Create an EKS Cluster with Crossplane

2. Create an EKS cluster - eksctl with a config file.

The config file I provide here gives us everything we defined at the command line and more. As mentioned - it also allows us to install Karpenter in the same eksctlexecution - thus giving us an industry-standard auto-scaling EKS cluster with just-in-time node provisioning. You can grab this file in Github too.

apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

metadata:
  name: way3
  region: eu-central-1
  version: "1.30"
  tags:
    karpenter.sh/discovery: way3
iam:
  withOIDC: true

managedNodeGroups:
  - name: ng-way3-1
    labels: { role: worker }
    instanceType: t2.medium
    desiredCapacity: 2
    minSize: 1
    maxSize: 3
    tags:
      nodegrouprole: way3
    volumeSize: 20
    iam:
      withAddonPolicies:
        externalDNS: true
        certManager: true
        awsLoadBalancerController: true
        albIngress: true
        ebs: true
        efs: true
        imageBuilder: true
        cloudWatch: true
    ssh:
      allow: true # will use ~/.ssh/id_rsa.pub as the default ssh key

karpenter:
  version: '0.37.0'
  createServiceAccount: true
  withSpotInterruptionQueue: true

An attentive eye will also notice I've also defined some additional stuff such as CloudWatch logging of the control plane, EBS and EFS access. Consider removing these lines if you don't need them. Also you'll notice that not only it installs Karpenter, it also takes care of setting up the SpotInterruptionQueue, which allows Karpenter to replace spot instances before they die. And there are many additional options available. So yes - this is a very scalable approach, which takes care of more or less everything one might need in an EKS cluster.

Execute this plan with:

eksctl create cluster -f cluster.yaml

This again creates a CloudFormation execution that, granted we have all the necessary permissions, should complete successfully.

Let's check that Karpenter got installed:

kubectl get pod -A
NAMESPACE     NAME                         READY   STATUS    RESTARTS   AGE
karpenter     karpenter-79db484bbf-flzzq   1/1     Running   0          32s
karpenter     karpenter-79db484bbf-nfhsp   1/1     Running   0          32s
kube-system   aws-node-8h4ln               2/2     Running   0          17m
kube-system   aws-node-vq8wj               2/2     Running   0          18m
kube-system   coredns-6f6d89bcc9-qx497     1/1     Running   0          24m
kube-system   coredns-6f6d89bcc9-wwjtp     1/1     Running   0          24m
kube-system   kube-proxy-8mnd2             1/1     Running   0          18m
kube-system   kube-proxy-c5zkp             1/1     Running   0          17m

Yup, here it is!

The upside of using the config file is of course the ability to manage stuff in a somewhat idempotent way. So for example if we want to change our node group config - we can update the following lines:

- name: ng-1
    labels: { role: worker }
    instanceType: t2.medium
    desiredCapacity: 1
    minSize: 1
    maxSize: 5

and then run eksctl update nodegroup -f cluster.yaml - this will update our NodeGroup autoscaling range.

And of course eksctl provides us with a plethora of addtional commands that come very handy for ongoing management of EKS clusters:

eksctl -h
The official CLI for Amazon EKS

Usage: eksctl [command] [flags]

Commands:
  eksctl anywhere                        EKS anywhere
  eksctl associate                       Associate resources with a cluster
  eksctl completion                      Generates shell completion scripts for bash, zsh or fish
  eksctl create                          Create resource(s)
  eksctl delete                          Delete resource(s)
  eksctl deregister                      Deregister a non-EKS cluster
  eksctl disassociate                    Disassociate resources from a cluster
  eksctl drain                           Drain resource(s)
  eksctl enable                          Enable features in a cluster
  eksctl get                             Get resource(s)
  eksctl help                            Help about any command
  eksctl info                            Output the version of eksctl, kubectl and OS info
  eksctl register                        Register a non-EKS cluster
  eksctl scale                           Scale resources(s)
  eksctl set                             Set values
  eksctl unset                           Unset values
  eksctl update                          Update resource(s)
  eksctl upgrade                         Upgrade resource(s)
  eksctl utils                           Various utils
  eksctl version                         Output the version of eksctl

One of the more significant recent additions is support for EKS Hybrid Nodes. This capability lets you run on-premises and edge applications on customer-managed infrastructure while using the same EKS clusters, features, and tools you already use in the AWS Cloud.

eksctl can provision and manage clusters that include hybrid nodes, so teams extending Kubernetes workloads beyond the cloud can keep a single, consistent workflow. This feature requires a recent release of eksctl, so make sure you are running an up-to-date version before enabling it.

Thanks for reading and may your clusters be lean!

P.S. now you got a cluster - why not start managing its cost and performance for free with PerfectScale by DoiT - the leading Kubernetes cost optimization solution?

Join now to build clusters you can be proud of: perfectscale.io.