Node Autoscaling and Karpenter

So far, we’ve focused on scaling workloads by adding and removing pods based on demand. However, you can have KEDA configured to automatically create new pods; it won’t matter if there’s nowhere for your pods to run. This is where node autoscaling comes in.

Node autoscaling is the other half of the equation. When your workload autoscaling creates new pods that can’t be scheduled because your cluster is at capacity, you need a way to add nodes. And when those pods get removed and you’re left with underutilized nodes, you need a way to consolidate and remove them to stop wasting resources. This is what tools such as Cluster Autoscaler and Karpenter do: they manage the infrastructure layer so your workload autoscaling can actually work.

We’ll start with Cluster Autoscaler, which has been around for years and is still what most teams use. You’ll learn how it works, how to set it up properly (especially...