Minimal private-only EKS cluster

Sources

• aws/containers-roadmap#298 (comment)
• https://github.com/vranystepan/codepipeline-eks-example
• snippets were collected all over the internet 😂

Variants

• ./managed/ with managed EKS groups
• ./self-managed/ with classic self-managed ASGs, use this variant if you want to control maximum of the configuration (special kubelet args such as labels and taints, ...)

Key components

• VPC endpoint for s3 for Docker images of base EKS components
• VPC endpoints for ECR for Docker images of base EKS components
• VPC endpoint for EC2
• VPC with enableDnsHostnames and enableDnsSupport enabled

Process

1. setup remote backend for terraform
2. terraform plan
3. terraform apply
4. extend!

Result

NAME                                           STATUS   ROLES    AGE     VERSION
ip-10-20-1-122.eu-central-1.compute.internal   Ready    <none>   2m50s   v1.15.10-eks-bac369

kubectl get pods --all-namespaces
NAMESPACE     NAME                       READY   STATUS    RESTARTS   AGE
kube-system   aws-node-k7znj             1/1     Running   0          2m43s
kube-system   coredns-5b6dbb4b59-nqqqv   1/1     Running   0          6m29s
kube-system   coredns-5b6dbb4b59-wkcf9   1/1     Running   0          6m29s
kube-system   kube-proxy-mz4ff           1/1     Running   0          2m43s

It basically means that Kubernetes nodes were able to pull all the assets from VPC endpoins. Noice!

IAM roles for K8S workload

If you need IAM roles for Kubernetes workload, use IAM roles for Kubernetes Service Accounts. That's the best and most secure way how to handle this. See more in our blog post: https://blog.pipetail.io/posts/2020-04-13-more-eks-tips/

Authors

Stepan Vrany, buy me a beer if you want 😉

Need any help?

[email protected], @MstrsObserver or @MarekBartik. Simple.

License

MIT