GKE Deployment processes Pub/Sub messages per Pod; publish rate varies and sometimes spikes. Which autoscaling feature lets your Deployment scale to handle message volume automatically?

The Horizontal Pod Autoscaler (HPA) based on an external metric is the correct choice because it directly scales your GKE Deployment based on the Pub/Sub message backlog or other Pub/Sub-specific metrics. This allows the autoscaler to react to the actual work queue, ensuring your application can handle varying message volumes and spikes efficiently. Vertical Pod Autoscaler (VPA) adjusts CPU and memory requests/limits for individual pods, not the number of pods. While VPA in Auto mode automatically applies these adjustments and Recommendation mode suggests them, neither scales the number of pods to handle increased message volume. HPA based on resource utilization scales based on CPU or memory usage, which might not directly correlate with Pub/Sub message volume, leading to suboptimal scaling for message-driven workloads.