Question

A retail company runs a stateless web frontend on a GKE Standard cluster. Traffic is highly variable, with sharp spikes during flash sales. The team observes that during off-peak hours the cluster keeps many partially-utilized nodes running, but during spikes the Horizontal Pod Autoscaler adds pods faster than the cluster can schedule them because CPU-based scaling reacts too slowly. Leadership wants to reduce steady-state cost while ensuring the frontend can absorb sudden spikes without request failures. What should the DevOps engineer implement?

Accepted Answer

Configure the HPA to scale on a custom requests-per-second metric with a lower target, enable the cluster autoscaler with the optimize-utilization profile, and provision a small buffer using balloon (low-priority pause) pods.

Answer

Increase the HPA target CPU utilization to 90% and disable the cluster autoscaler so nodes stay at a fixed count sized for peak traffic.

Answer

Replace the HPA with a scheduled CronJob that scales the Deployment up before known sale windows and down afterward.

Answer

Move the frontend to a single large node pool of committed-use N2 instances sized for peak load to guarantee capacity at a discount.