Our AWS bill for Fargate has tripled as we scaled our microservices. I'm considering migrating to EKS with Spot Instances to lower our compute costs. However, I’m worried about the operational complexity of managing Kubernetes clusters. Is the cost saving in 2023 significant enough to justify the extra DevOps hours required for K8s management?
3 answers
We made this jump in early 2023, and while we saved about 40% on our EC2 bill by using EKS with Karpenter for intelligent autoscaling and Spot instances, our "people cost" went up. You need at least one dedicated engineer who really understands K8s networking and security. Fargate is great because it's "set and forget," but you pay a premium for that convenience. If your monthly cloud spend is under $10k, stay on Fargate. If you are pushing $50k+, the savings from EKS will easily cover the salary of a DevOps specialist to manage it.
Did you look into "Fargate Spot" before deciding on EKS? It offers up to 70% discounts and might be a middle ground without the K8s complexity.
We stayed on Fargate but optimized our "Right-Sizing." Most of our tasks were over-provisioned by 50% on RAM. Fixing that was easier than migrating to Kubernetes.
Right-sizing is key! Linda mentioned Karpenter for EKS, but even on Fargate, using a tool like AWS Compute Optimizer can highlight where you are literally throwing money away.
James, we tried Fargate Spot, but the interruptions were too frequent for our stateful services. With EKS and Spot, we have more granular control over how we handle node terminations using termination handlers. It's more work to set up, but the stability is much higher for our specific workload. Plus, EKS gives us access to the wider CNCF ecosystem which we really needed for our observability stack. Posted by: David Clarke Date: 24-09-2023