DEV Community

EKS Auto Mode Unlocked for Existing Clusters with Terraform

Jatin Mehrotra on December 13, 2024

In the previous blog, I explained that EKS Auto mode is now supported by terraform-eks-module and illustrated how we can create new cluster with E...

Read full post

Deepak B • Jan 24

Thanks for the post Jatin, This is very helpful.

I have quick query on how do we create custom NodePool and Node Class for existing cluster as we have specific requirements from compute perspective.

Jatin Mehrotra • Jan 25 • Edited

THANK YOU @deepak_b. I am happy that it helped you.

Here is the documentation

docs.aws.amazon.com/eks/latest/use...
docs.aws.amazon.com/eks/latest/use...

Nodepool requires to have nodeclass in the cluster.

Let me know if you have any other questions would be happy to help and I am very curious to understand how you are using EKS auto for your needs and how it is helping you. Any challenges you are facing or you wish they have included some specific feature in EKS Auto? Please share your views.

Deepak B • Jan 29

Thanks Jatin.. was wondering on how can we add it to the terraform code for existing cluster so that when migrate our workload to Auto mode we make sure its using custom nodeclass and nodePool

Jatin Mehrotra • Jan 29

@deepak_b Thank you for your question

how can we add it to the terraform code for existing cluster so that when migrate our workload to Auto mode we make sure its using custom nodeclass and nodePool

There is one consideration before you can add custom nodeClass and nodePool

nodeClass and nodePool are the concepts of karpenter. Having said that it also means the EKS Auto when enabled on EKS installs karpenter and CRD to support nodeClass and nodePool
This means first you need to enable EKS Auto mode on your cluster then using terraform install nodeClass and nodePool
nodeClass and nodePool essentially are yaml manifest which can be applied using terraform using kubectl_manifest resource

Note: In order to use this resource you need to connect respective kubectl provider and configure it so that it can connect to eks cluster

provider "kubectl" {
  host                   = var.eks_cluster_endpoint
  cluster_ca_certificate = base64decode(var.eks_cluster_ca)
  token                  = data.aws_eks_cluster_auth.main.token
  load_config_file       = false
}

Hope it helps.

Deepak B • Jan 29

Thanks Jatin. This helps :)

Jatin Mehrotra • Jan 30

I am also curious how you are trying to use EKS Auto for what kind of use case and why did you guys decided to move to EKS Auto? Please share it if you have time

Deepak B • Feb 4

Sure Jatin, We are currently in POC phase, were exploring Karpenter for node autoscaling and better optimisation of resources that's when AWS launched EKS AUTO, this makes it more convenient as now its all managed by AWS. It would be interesting run chaos engineering with this setup and find out how app responds to scaling requests and obvious time taken for scaling request to first request processed by new pod.

Jatin Mehrotra • Feb 5

Sounds cool,Now it all makes sense.
I would be really interested to know how you are pilling the chaos engineering with this setup.

Just in case you can use AWS FIS for injecting fault into Kubernetes pods : docs.aws.amazon.com/fis/latest/use...

You can use the aws:eks:pod actions to inject faults into the Kubernetes Pods running in your EKS clusters.

Lets connect on Linkedin so that we can have more discussion on this linkedin.com/in/jatinmehrotra/

Deepak B • Feb 5

Yeah, we have been using FIS since ~2yrs now. Love, how its catching up with other leading tools.

Hari Karthigasu • Dec 30 '24

bootstrap_self_managed_addons = true saved my day :D, Thanks!

Jatin Mehrotra • Dec 31 '24

@harik8

I lost my few hairs for this :D Happy that it helped. Share it with Other EKS lovers or in your Linkedin Network if you think it can help others too