GCP Professional Cloud Architect Practice Question

Your team has a Kubeflow v2 pipeline running on Vertex AI Pipelines. Compliance now requires that each time a new CSV arrives in gs://retail-landing the pipeline must run automatically; every execution must store complete dataset, code, and hyper-parameter lineage for audit; and a human must approve or reject deployment to the production endpoint from the Cloud Console. Which architecture satisfies all requirements while minimising custom code maintenance?

Implement a Cloud Composer DAG that polls the landing bucket, triggers Dataflow preprocessing and Vertex AI training, then pauses using a Slack-based approval before deployment while maintaining a separate metadata database.
Configure Cloud Storage object-create notifications to Pub/Sub, invoke a Cloud Functions subscriber that starts a Dataproc training job and writes custom logs for traceability.
Send Cloud Storage notifications to Pub/Sub, trigger a Cloud Workflows execution that calls the Vertex AI Pipelines REST API with the pre-compiled pipeline, include a built-in Manual Approval step before deployment, and rely on Vertex ML Metadata for lineage.
Use Eventarc to trigger Cloud Build on each new object; the build runs gcloud ai custom-jobs create and relies on Cloud Build logs plus Cloud Deploy approvals for promotion.

Report Issue

Answer Description

Publishing Cloud Storage object-creation events to Pub/Sub and using a Cloud Workflows trigger keeps the glue layer serverless; the workflow simply calls the Vertex AI Pipelines REST API to launch the already compiled pipeline, so no custom runner code is needed. Vertex AI Pipelines automatically writes datasets, code versions, and hyper-parameters to Vertex ML Metadata, delivering the required lineage. A built-in Manual Approval component pauses the pipeline and surfaces an approval prompt in the Cloud Console before the deployment step. The other approaches either omit the explicit in-pipeline approval, lack automatic lineage capture, or require operating and maintaining extra infrastructure such as Airflow or extensive custom logging.

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.