Prepare and test your skills
Prepare and test your skills
Free sample question, answered and explained below. The Professional Cloud DevOps Engineer practice bank has 1,455 more like it. Try 10 free: instant score, no account.
Your team is building an automated continuous training (CT) and continuous deployment (CD) pipeline for a mission-critical tabular classification model on Google Cloud. You need to implement a solution using Vertex AI Pipelines that continuously evaluates candidate models and deploys them to production.
The pipeline implementation must meet the following operational requirements:
Which architecture and pipeline design should you implement?
google.VertexEndpoint using ModelDeployOp, and configure a post-deployment ModelEvaluationForecastingOp step to automatically initiate an endpoint rollback if live inference accuracy drops below the threshold.This approach unconditionally deploys the newly trained model directly to the live production endpoint using ModelDeployOp immediately after training, then attempts to run a post-deployment validation step using ModelEvaluationForecastingOp to roll back the endpoint if errors are detected.
ModelEvaluationForecastingOp is designed exclusively for time-series forecasting models, not tabular classification models. Using it for a classification workload will cause pipeline execution errors.ModelDeployOp does not natively perform automatic rollbacks based on subsequent component failures without complex custom orchestration.ModelEvaluationForecastingOp vs. classification).The workload was a forecasting task and the deployment strategy used an isolated canary staging endpoint that does not serve live production traffic prior to verification.
In Vertex AI Pipelines, model evaluation should always precede model uploading and endpoint deployment in the pipeline DAG.
VertexAITextEmbeddings to update endpoint configurations.This option uses an Apache Beam data processing pipeline to stream model prediction outputs into an AlloyDB for PostgreSQL table, executes database stored procedures to compute evaluation metrics, and calls VertexAITextEmbeddings to manage endpoint deployments.
VertexAITextEmbeddings is used in vector search and retrieval-augmented generation (RAG) pipelines for generating text embeddings, not for deploying tabular classification models to Vertex AI Endpoints.The requirement was to build a real-time semantic search indexing pipeline that transforms unstructured product catalogs into vector embeddings stored in AlloyDB.
AlloyDB and Beam MLTransform are powerful tools for vector embedding workflows, but tabular model evaluation and deployment gating should be handled natively within Vertex AI Pipelines components.
ModelUploadOp via the REST API while bypassing Vertex ML Metadata.This option relies on printing metrics from pipeline containers into standard output, extracting those metrics using Cloud Logging metric filters, and triggering an asynchronous Pub/Sub and Cloud Run workflow to upload the model via REST API calls.
google.VertexModel, google.VertexDataset).The deployment target was an unmanaged external system running on-premises that could not interact directly with Google Cloud Pipeline Components.
Vertex AI Pipelines natively integrates with Vertex ML Metadata precisely to avoid the need for custom log scrapers and asynchronous metric extraction webhooks.
ModelEvaluationClassificationOp to generate evaluation metrics, log metadata to Vertex ML Metadata, and gate deployment to a google.VertexEndpoint via ModelDeployOp inside a dsl.Condition block.This architecture uses Vertex AI Pipelines orchestrated via the Kubeflow Pipelines (KFP) SDK and Google Cloud Pipeline Components (GCPC). It integrates automated model evaluation using ModelEvaluationClassificationOp, logs all pipeline artifacts natively into Vertex ML Metadata, and enforces a quality deployment gate using native KFP conditional logic (dsl.Condition) before invoking ModelDeployOp.
google.VertexModel and google.VertexEndpoint preserve end-to-end data and model lineage.ModelEvaluationClassificationOp ingests ground-truth test data and candidate batch prediction outputs to compute standard classification metrics (such as ROC-AUC, accuracy, precision-recall, and sliced metrics).ModelDeployOp component inside a dsl.Condition block allows the pipeline DAG to dynamically evaluate whether the metric output meets or exceeds the required threshold. The model is uploaded via ModelUploadOp and deployed to the existing target google.VertexEndpoint only if the condition evaluates to true.google.VertexModel) across components without custom boilerplate.This approach aligns with Google Cloud recommended MLOps practices. It eliminates manual intervention, leverages managed components designed specifically for tabular classification evaluation, and avoids brittle external glue code or post-deployment rollback scripts.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated Validated