AWS EKS Pricing Calculator

Fargate charges per pod based on requested vCPU ($0.04048/vCPU-hr) and memory ($0.004445/GB-hr). A 1 vCPU / 2 GB pod costs ~$0.049/hr or ~$36/mo running 24/7.

Usually yes. Self-managed control planes need 3+ EC2 instances for HA ($100+/mo), plus your team's time for upgrades and maintenance. $73/mo for a managed, HA control plane is a good deal.

NAT Gateway ($33+/mo per AZ), ALB/NLB (~$16/mo base + usage), EBS storage ($0.08/GB for gp3), CloudWatch logging, and data transfer. These can add 30-50% to the base cluster cost.

Yes. Fargate Spot offers up to 70% discount. Pods can be interrupted with 2 minutes notice. Good for batch processing, CI/CD, and non-critical workloads. Configure via Fargate profiles.

Yes. EC2 Reserved Instances and Savings Plans apply to EKS worker nodes. Compute Savings Plans also cover Fargate. Commit 1-3 years for 30-60% savings on steady workloads.

Use Karpenter for smart autoscaling, spot instances for non-critical workloads, right-size pod requests, use VPC Endpoints instead of NAT Gateway for AWS services, and consolidate clusters where possible.

Karpenter is an open-source node provisioner for Kubernetes. It selects optimal instance types based on pod requirements, supports spot instances, and consolidates nodes to minimize waste. Replaces Cluster Autoscaler.

At $73/mo per cluster, fewer clusters saves money. One large cluster with namespaces for isolation is more cost-efficient. But separate clusters for prod/staging or for compliance isolation is common.

When workloads are spiky, short-lived, or small (under ~10 steady pods). Also when the engineering cost of managing nodes exceeds the compute savings. For steady 20+ pod workloads, EC2 is usually cheaper.

Warden monitors your EKS-hosted services from outside your cluster. It catches ingress, load balancer, and DNS issues that internal monitoring can miss. 10-second checks from multiple global regions.