GCP Professional Cloud Architect Practice Question

Your online learning platform runs a regional GKE Standard cluster. A Deployment currently scales with a Horizontal Pod Autoscaler (HPA) that targets 60 % CPU, but load tests show pods crash at 80 % memory while CPU stays below the threshold, so no scale-out occurs. You must redesign scaling so that pods add replicas when either CPU or memory exceeds 70 %, and node pools automatically grow and shrink with minimal operational effort. Which solution best meets these goals?

Replace the HPA with a KEDA scaler that watches a Cloud Monitoring memory metric and turn on node auto-provisioning instead of the cluster autoscaler.
Keep the CPU-based HPA, add a Vertical Pod Autoscaler in recommend mode to right-size memory requests, and disable the cluster autoscaler to prevent conflicts.
Migrate the service to Cloud Run, set concurrency to 1, and cap the revision at 1 000 instances to obtain automatic scaling.
Create an HPA that targets 70 % on both cpu and memory resource metrics, and enable the cluster autoscaler on the node pool.

Report Issue

Answer Description

GKE's HPA can use multiple built-in resource metrics simultaneously. Defining one HPA that targets 70 % for both cpu and memory causes a scale-out when either metric is above the target, eliminating the current blind spot. Because the metrics server is enabled by default on GKE, no additional components are needed. Enabling the cluster autoscaler on the node pool lets the cluster add or remove nodes as replica counts change, meeting the cost-efficiency requirement with native, fully managed capabilities.

KEDA adds a third-party component and is unnecessary when resource metrics are sufficient. A Vertical Pod Autoscaler only tunes requests and, with the cluster autoscaler disabled, would not add nodes or replicas quickly enough. Migrating to Cloud Run changes the platform, may still need memory tuning, and introduces higher re-architecting effort that the question explicitly wants to avoid. Therefore, using a single HPA with both cpu and memory metrics plus the cluster autoscaler is the optimal answer.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.