What instance families are people using for production workloads on Amazon EKS? Our cluster currently runs mostly Node.js microservices on T3 instances, and so far they've worked well. Is it reasonable to stay with T3, or would moving to Graviton or another instance family provide better performance or lower costs?
3 Answers
Graviton nodes can be worth testing for cost savings if your images and dependencies support ARM64. I’d start with a separate node group and move a few compatible services over before changing the whole cluster.
T3 instances are a reasonable choice for many production clusters, especially when the services are relatively lightweight. The main thing to watch is whether your workloads regularly exhaust their CPU credits or need more memory. Monitoring components can sometimes become the first resource-heavy workloads as the cluster grows.
Check the CPUCreditBalance metric in CloudWatch over at least the past month. If it stays healthy, T3 is probably meeting your needs and the decision becomes mostly about pricing. Also verify whether unlimited mode is enabled, since sustained CPU usage there can create additional charges.
If you switch from x86 to ARM, remember to update your CI runners and build pipelines too. Building images for the wrong architecture can lead to much longer build times or images that won’t run on the new nodes.
T3 is fine until the CPU credit balance starts hitting the cap and performance becomes unpredictable. Graviton can work well, but you’ll need ARM64-compatible container images and a bit more build setup. For light Node.js services without sustained CPU usage, there’s no urgent reason to move.