GCP Professional Cloud Architect Practice Question

Your company is containerizing an internal image-rendering microservice that will be invoked over HTTP by other workloads running in several projects attached to the same Shared VPC. For most of the day the service is idle, but during marketing campaigns it must burst to thousands of concurrent requests and then scale back to zero to avoid unnecessary cost. Each request can take up to 20 minutes to finish. Operations wants to avoid managing clusters or virtual machines, and security requires that the service be reachable only over the private VPC, never from the public internet. Which Google Cloud compute option best satisfies these requirements with the least operational overhead and cost?

Implement the service as an HTTP Cloud Function with the maximum timeout and a Serverless VPC Access connector to the Shared VPC.
Create a GKE Autopilot cluster, expose the deployment with an internal HTTP(S) load balancer, and enable a Horizontal Pod Autoscaler that can scale down to zero.
Run the container in App Engine flexible environment with automatic scaling and firewall rules that block external IP ranges.
Deploy the container on Cloud Run, set ingress to internal-only, and configure minimum instances to 0 so it scales from zero to peak automatically.

Report Issue

Answer Description

Cloud Run can deploy any OCI-compliant container image, automatically scale the number of instances from zero to thousands based on incoming requests, and supports request time-outs up to 60 minutes. When the service's ingress setting is configured to "internal" (internal-only), it is reachable exclusively from clients inside the same VPC network or connected VPCs, fulfilling the private-access requirement without exposing public endpoints. Setting minimum instances to 0 means no charges accrue while the service is idle.

Cloud Functions is discarded because its maximum request duration is 9 minutes and it cannot be fully isolated from the public internet. GKE Autopilot still requires cluster lifecycle management and costs for system nodes; Horizontal Pod Autoscaler cannot scale deployments to zero. App Engine flexible instances do not scale to true zero, and you continue to pay for at least one provisioned VM even when traffic is absent. Therefore, Cloud Run with internal ingress and no minimum instances is the most suitable choice.

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.