A managed node group's scaling config has three numbers:

• desiredSize is the number of nodes right now. With the Cluster Autoscaler it moves on its own, up on pending pods and down on idle nodes, and AWS says not to change it by hand.
• minSize is the floor the autoscaler will not scale below.
• maxSize is the ceiling it will not scale above.

The autoscaler never touches the floor or the ceiling; those are yours. To keep spare capacity around, such as headroom for bursty deployments, raise minSize and the autoscaler stops taking those nodes away.

The catch: EKS validates minSize <= desiredSize <= maxSize against the desired size as it is right now. It does not lift the desired size to meet a new minimum the way a bare Auto Scaling group does, which has been an open request since 2022. Raising the minimum above the current desired size fails:

InvalidParameterException: Minimum capacity 100 can't be greater than desired size 60

So the desired size has to go up first, or in the same call:

aws eks update-nodegroup-config \
  --cluster-name my-cluster \
  --nodegroup-name my-nodegroup \
  --scaling-config minSize=100,desiredSize=100,maxSize=200

In Terraform the usual ignore_changes = [scaling_config[0].desired_size] makes this worse: Terraform keeps sending the desired size it last read, so bumping desired_size next to min_size in the config changes nothing and the apply fails with the same error. Run the CLI command above, then put the same min_size in the config so the next plan is clean. Raising only the desired size from the CLI and letting Terraform raise the floor leaves a window: by default the autoscaler starts removing idle nodes after about ten minutes.

The same rule applies at the top: a maxSize below the current desired size is rejected too.