A single GKE cluster uses one 200-node pool for batch, stateful, and stateless workloads. Reduce cost without harming availability.

Enabling Horizontal Pod Autoscaler (HPA) allows stateless and compatible stateful workloads to scale pods up or down based on CPU utilization or other metrics, optimizing resource usage. Cluster autoscaler (node auto scaling) then dynamically adjusts the number of nodes in the node pool to match the pod demand, preventing over-provisioning and reducing costs. This combination ensures resources are only consumed when needed, maintaining availability. Creating a second cluster adds operational overhead and doesn't inherently reduce costs without autoscaling. Enforcing CPU/memory limits improves resource management but doesn't dynamically scale infrastructure. Preemptible VMs are unsuitable for stateful or critical workloads due to their ephemeral nature, risking availability.