Question

A retail company runs a stable, always-on microservices platform on a regional GKE Standard cluster. The workload has a predictable baseline of 40 e2-standard-8 nodes that run 24/7, with occasional bursts during promotions. The FinOps team notices the platform is billed entirely at on-demand rates and asks you to reduce the recurring compute spend on the steady-state nodes without risking availability during bursts. What should you do?

Accepted Answer

Purchase a compute-based committed use discount (CUD) sized to the 24/7 baseline capacity, and keep the burst nodes on on-demand autoscaling.

Answer

Move the entire cluster, including burst capacity, onto Spot VMs to obtain the deepest possible discount.

Answer

Buy a resource-based committed use discount sized to peak burst capacity so all nodes, including bursts, are covered.

Answer

Enable sustained use discounts by pinning all nodes to a single zone so they accumulate uninterrupted runtime.