[rudi_mk]

I run 200+ production clusters across EKS, GCP and MSA. In a nutshell - running your own clusters and being responsible for every aspect - autoscaling, storage, upgrades - it's not fun. Using managed Kubernetes allows you to run a cluster knowing all those bells and whistles are already taken care of. Hard to go wrong with the following setup:

1. A cluster VPC with public and private subnets.

2. A managed control plane on EKS.

3. You can opt between self-managed and managed nodegroups. To be fair, self-managed nodegroups give you more control, but you'll need to be careful about updating them yourself.

4. Use AWS controllers - the AWS Ingress Controller, the EBS CSI Driver - stuff like this will ensure your cluster can provision and manage load balancers, storage and so on.

5. For starters use cluster-autoscaler to dynamically manage compute capacity on your nodegroups. As your platform builds up and scales, you can look at Karpenter as a viable alternative.

[reacharavindh]

I don’t think the parent poster meant EKS when they talk about “in-house” and “provision new hardware, servers etc”. I’d think more in the lines of buying a Dell/Lenovo server, racking it in their server room or collocated space, and running Kubernetes by themselves on their own hardware - not the cloud.