Spot Instances vs. Savings Plans: The EKS Cost Kill-Box Part 1

Here’s the architecture for a 30% reduction in your AWS bill without sacrificing 99.9% uptime.

1. The Strategy: The 70/30 Split

Don't put everything on Spot. And don't put everything on Savings Plans.

• 70% Spot Instances: For your stateless workloads, worker nodes, and dev environments. Managed Node Groups make this trivial now.
• 30% Savings Plans: For your stateful services (DBs, Cache) and the absolute "minimum base" of your production traffic.

2. Handling Interruptions with Karpenter

The biggest fear with Spot is the 2-minute interruption notice. Karpenter handles this natively with Spot Termination Handling. When AWS reclaims a Spot instance, Karpenter immediately:

1. Identifies the replacement.
2. Drains the old node.
3. Spins up a new one before the 2 minutes are up.

spec:
  requirements:
    - key: "karpenter.sh/capacity-type"
      operator: In
      values: ["spot"]
    - key: "kubernetes.io/arch"
      operator: In
      values: ["arm64"] # Use Graviton for an extra 20% savings

3. The Math

Verdict: If your EKS bill is >$5k/month and you aren't using Spot for at least 50% of your compute, you are donating money to Jeff Bezos.