Is EKS Auto Mode Worth It for Production Use?

EKS Auto Mode is essentially Karpenter, which requires zero maintenance once it’s set up. After the initial configuration, you can run it on Fargate or its own ASG and not have to check on it often—you'll save a lot of money that way.

Some of my clients are using EKS Auto Mode. One thing to note is that it sets the IMDSv2 hop limit to 1, which impacts how you can use instance profiles within pods. You'll need to explore using IRSA with webIdentity or Pod Identity instead if you're planning on using that feature—it's a security measure on AWS's end.

If your team is technical enough, you might want to manage Karpenter yourself. Just a heads up, EKS Auto Mode adds an extra 10% cost on all instances in your cluster, which can really add up depending on your cluster's size.

I’m considering Auto Mode mainly to avoid manual upgrades for kube-proxy, vpc-cni, and other EKS add-ons. I've heard that while EKS manages these upgrades, they can drift apart if not monitored. However, all clusters and plugins upgrade simultaneously in Auto Mode, so it limits testing options in staging before going live.

We’re considering Auto Mode too since EKS and AWS lag behind AKS and GKE in some areas like autoscaling and automatic patching, which are often included for free with competitors. However, it's frustrating that AWS charges an extra 10% for these features. My company can deploy on any major cloud platform, and we’re keen on improving our EKS experience, but it’s not a top priority right now because most clients don’t have a strong preference.