We Recommend Managed Node Groups over Fargate for EKS Add-Ons

docs.cloudposse.com

11 pointsmooreds6 comments

When talking about EKS Auto Mode, you wrote "Available only for Kubernetes 1.29+". The oldest kubernetes version you can create on AWS is 1.31 at the moment.

I feel the real problem here is mixing and matching kubernetes with ECS. They aren’t really meant to play together and were never designed to integrate together.

That's false. The only thing which slows things down is that Autoscaler is provisioning compute resources in a way which is dog slow.

If you add Karpenter (https://karpenter.sh) to the cluster it can scale based on the rules set out in the configuration and sends out requests directly to the underlying cloud orchestrator. Which in practice means it spins up new compute resources near instantly.

I see that they added support for Azure aswell recently (been a while since I used it)

who said anything about ecs?

I’m just showing my age since fargate was initially tied to ECS (and still mostly used with it). But the general issues still remain the same. It’s a container orchestrator trying to integrate with another container orchestrater.

I've actually had no issues running karpenter on fargate. Not sure what other uses cases there are for it, though.

Reply on news.ycombinator.com