GCP Professional Cloud Architect Practice Question

Your company is replacing a thumbnail-generation service that currently runs on Compute Engine. When a user uploads a video to a Cloud Storage bucket, Node.js code must extract thumbnails within 40 seconds and write them to a different bucket. Traffic is highly bursty-from a few files per hour to as many as 50 000 per minute-and the team wants to pay only for actual execution time without managing servers or pre-provisioning capacity. Which Google Cloud compute option best meets these requirements?

Rewrite the service for App Engine Standard Node.js, configuring automatic scaling to handle traffic spikes.
Deploy the Node.js code as a Cloud Function (2nd gen) with a Cloud Storage finalize trigger; the platform auto-scales to zero and charges only for invocation time.
Create a GKE Autopilot cluster and deploy the code as a Deployment with a Horizontal Pod Autoscaler triggered by Pub/Sub events from Cloud Storage.
Package the code in a container and run it on Cloud Run with min instances set to zero, using Eventarc to deliver Cloud Storage notifications.

Report Issue

Answer Description

Cloud Functions (2nd gen) is designed for event-driven workloads. It offers a native Cloud Storage finalize trigger, so no intermediary Pub/Sub topic or additional service is required. Instances start automatically when events arrive, scale horizontally to match demand (and down to zero after 15 minutes of idleness), and you are billed in 100-millisecond increments only for execution time.

Cloud Run can also scale to zero, but you must package and manage a container image and configure Eventarc (or a custom Pub/Sub push) to receive Cloud Storage events, adding operational overhead.

App Engine Standard primarily handles HTTP requests and cannot be triggered directly by Cloud Storage events; you would need an extra component to forward events. While it can scale to zero with default automatic scaling, it is not the most direct or cost-efficient fit for this event-driven pipeline.

GKE Autopilot removes node management but still incurs per-pod charges for requested CPU and memory even when idle, and it requires additional components-such as Pub/Sub subscriptions and Horizontal Pod Autoscalers-to ingest and scale on events.

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.