GCP Professional Cloud Architect Practice Question

An e-commerce company runs a GKE cluster where stateless worker pods perform product-image rendering jobs that last 20-40 minutes. Each job reads source images from Cloud Storage, writes results back to the same bucket, and automatically retries if a pod terminates. Marketing usage spikes unpredictably, and leadership asks you to cut compute costs by at least 60 % without jeopardizing the four-hour turnaround time for completed renders. Which change best satisfies these requirements?

Move the workload to a GPU-based node pool that benefits from sustained-use discounts and disable the retry mechanism to avoid duplicated work after disruptions.
Purchase three-year committed use discounts for the existing n2-highcpu-16 node pool and leave the workload placement unchanged.
Keep the current node pool but raise the cluster autoscaler utilization target from 60 % to 95 % to reduce the number of nodes kept running.
Create an additional node pool composed of Spot VMs, configure the rendering deployment to schedule onto that pool, and set a PodDisruptionBudget that allows zero running pods during preemptions so they are simply rescheduled.

Report Issue

Answer Description

The rendering tasks are short-lived, stateless, and already tolerate restarts through automatic retry logic, making them ideal for Spot VMs that can be preempted with short notice. Creating a dedicated Spot VM node pool allows the cluster autoscaler to add low-cost capacity during spikes and scale to zero when idle, typically reducing VM spend by 60 %-91 %. A permissive PodDisruptionBudget that allows all replicas to be evicted ensures the scheduler can reschedule pods quickly after preemption. Multi-year committed use discounts lower rates only for always-running nodes and therefore cannot achieve the required savings for bursty traffic. Simply raising the autoscaler utilization target reduces idle nodes but rarely yields a 60 % cost reduction when demand is unpredictable. Migrating to GPU nodes without adjusting provisioning increases hourly rates and removes the current retry safety net, risking higher costs and SLA violations. Deploying the workload on a Spot VM node pool is therefore the most effective, low-risk solution.

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.