Question

A retail company runs a stateless web frontend on a GKE Standard cluster. Traffic is highly variable: near-zero overnight, moderate during business hours, and 5x spikes during flash sales. The platform team observes that node utilization averages only 25% because a single large-machine node pool stays provisioned at peak capacity around the clock. Finance wants to cut compute spend without introducing risk of failed customer checkouts during spikes. Which change best optimizes cost while preserving spike-handling capacity?

Accepted Answer

Enable the cluster autoscaler on a right-sized node pool with appropriate Pod resource requests, and configure the Horizontal Pod Autoscaler to scale replicas based on CPU/traffic.

Answer

Purchase a 3-year committed use discount sized for the flash-sale peak capacity to lower the per-vCPU rate on the existing static node pool.

Answer

Migrate the entire node pool to Spot VMs to reduce cost, since the frontend is stateless.

Answer

Increase the machine size of the existing nodes and reduce the node count to two, relying on manual scaling before scheduled sales.