Is EKS Auto Mode a Good Choice for Production Workloads?

We're looking into using Auto Mode because it seems AWS is lagging behind AKS and GKE in features like autoscaling and automatic patching. It feels a bit annoying that they charge extra for it. My company can deploy on any of the big cloud providers, but right now, we've been avoiding AWS due to the higher maintenance involved.

Some of my clients are using EKS Auto Mode, but there's a key limitation: with Auto Mode, the IMDSv2 hop limit is set to 1. This means you can't use instance profiles within your pods unless you're okay with using the host network, which isn’t ideal for security. If you'll need that feature, consider using IRSA or Pod Identity instead for better security practices.

EKS Auto Mode is basically Karpenter, and once you set it up, it's pretty zero maintenance. You can run it on Fargate or an ASG, and check in on it maybe once every nine months, which could save you a ton of cash!

If your team has some technical expertise, you might want to consider setting up Karpenter yourself. Just keep in mind that EKS Auto Mode comes with an additional 10% charge on all your instances, which can add up depending on how large your cluster is.

I'm considering Auto Mode mainly to avoid the hassle of manually upgrading kube-proxy, vpc-cni, and other EKS add-ons that AWS manages. Just a heads up, all your clusters and add-ons get upgraded simultaneously, so you can't test anything in a non-production environment before moving to production.