Question

Difficulty: HardCapacity Planning and Infrastructure Workload Scaling Optimization

A healthcare data platform processes streaming clinical records using Google Cloud Pub/Sub and a worker pool running on Compute Engine Managed Instance Groups (MIGs). During sudden peak emergency intake windows, message backlogs accumulate rapidly, yet worker instances fail to scale out in time to prevent processing delays. An operational audit identifies two core constraints: the MIG autoscaler is currently driven by average CPU utilization—which remains low while tasks wait in queue—and recent manual attempts to add capacity hit an error indicating that the region's vCPU quota limit of 500500 vCPUs was reached. Which combination of architectural actions should the Cloud Architect implement to optimize capacity planning and resolve the workload scaling bottlenecks?

  1. Submit a regional vCPU quota increase request in advance for the target region, and reconfigure the MIG autoscaling policy to scale dynamically based on a Cloud Monitoring metric tracking Pub/Sub unacknowledged message count per instance.Answer
  2. B
    Lower the target CPU utilization threshold from 80% to 20% on the existing MIG autoscaler, and provision additional HA VPN tunnels to increase network bandwidth during emergency windows.
  3. C
    Migrate the database layer to Cloud Spanner across multi-regions to bypass regional vCPU quota limits automatically, while retaining CPU-based autoscaling for the compute instances.
  4. D
    Re-architect the processing worker pool to run on Google Kubernetes Engine (GKE) Autopilot, relying on cluster autoscaling to automatically bypass GCP regional project quotas without requesting quota adjustments.

Answer

The Cloud Architect should submit a regional vCPU quota increase request in advance for the target region and reconfigure the MIG autoscaling policy to scale based on a Cloud Monitoring custom metric tracking Pub/Sub unacknowledged message count per instance.
The solution requires addressing both the quota bottleneck and the metric misalignment. Requesting a regional vCPU quota increase in advance ensures sufficient ceiling for Compute Engine instances. Reconfiguring the MIG autoscaler to use Pub/Sub unacknowledged message counts via Cloud Monitoring ensures that instance scaling responds directly to workload queue depth rather than lagging CPU indicators.

Step-by-Step Solution

1
Analyze the capacity ceiling constraint
Identified that scaling is blocked at 500500 vCPUs due to GCP regional quota enforcement.
Compute Engine resource allocation cannot exceed project quotas regardless of autoscaling configuration.
2
Address quota pre-provisioning requirements
Determined that a regional vCPU quota request must be submitted prior to anticipated peak demand.
Quota increases require evaluation and approval time from Google Cloud operations.
3
Analyze the scaling metric bottleneck
Recognized that CPU utilization is a lagging indicator for message-queue-driven workloads.
Tasks accumulate in Pub/Sub before worker instances experience high CPU utilization.
4
Configure queue-based autoscaling metric
Selected Cloud Monitoring metric measuring unacknowledged messages per worker instance for MIG autoscaling.
Custom queue metrics allow proactive horizontal scaling aligned directly with workload backlog growth.

Key Concept

Workload Scaling Optimization and Quota Management
Rate this question