Question

A data engineering team runs many short-lived batch jobs on a GKE Standard cluster. Monitoring shows the cluster autoscaler keeps nodes running with average CPU utilization around 25%, because pods are spread thinly across nodes and rarely consolidate. Each job requests modest CPU/memory, and jobs are tolerant of brief restarts. The team wants to reduce node-hour cost without changing the jobs' resource requests. Which change most directly improves node packing and lowers cost?

Accepted Answer

Enable the GKE optimize-utilization autoscaling profile so the cluster autoscaler bin-packs pods more aggressively and scales down underutilized nodes faster

Answer

Switch the node pool to larger machine types so each node can hold more pods and reduce total node count

Answer

Enable Vertical Pod Autoscaler in Auto mode to raise the pods' CPU and memory requests

Answer

Increase the maximum node count in the node pool autoscaler configuration to allow more parallelism