Professional Data Engineer Practice Exam 2025: Latest Questions

A retail company wants to ingest clickstream events from a website and generate near-real-time metrics (events per minute by page) with end-to-end latency under 2 minutes. The solution must handle traffic spikes and allow reprocessing if a bug is found in the pipeline logic. What architecture should you choose?

A data team needs to load CSV files from Cloud Storage into BigQuery every day. The schema changes occasionally (new nullable columns added). They want the simplest approach that minimizes custom code while keeping a history of loads. What should they do?

Your team trained a classification model in Vertex AI. The business requires an explanation for each prediction to satisfy internal audit requirements. What is the most appropriate approach on Google Cloud?

A BigQuery dataset contains sensitive customer PII (email, phone). Analysts should be able to query aggregated results, but only a small security team can view raw PII. What is the recommended way to enforce this in BigQuery?

You run a Dataflow streaming pipeline that writes to BigQuery. The pipeline occasionally crashes and, after restart, you notice duplicate rows in the BigQuery table. You need to make the sink effectively exactly-once. What should you do?

A company has a multi-tenant analytics platform where each tenant must only see their own rows in shared BigQuery tables. Tenants are authenticated via Google identity, and analysts run ad-hoc queries. What is the best way to implement tenant isolation in BigQuery?

Your organization has a curated BigQuery dataset used by many downstream dashboards and ML pipelines. A new ingestion process sometimes introduces nulls and invalid values, causing broken dashboards. You need automated, repeatable data quality checks that run as part of the pipeline and can fail the workflow when checks do not pass. What should you implement?

You trained a model and deployed it to a Vertex AI endpoint. After a few weeks, prediction quality drops because input feature distributions have shifted. You want to detect drift and trigger investigation using managed services. What should you do?

A financial institution must process streaming transactions with exactly-once semantics, strong ordering per account, and the ability to replay a full history for audits. They also require low-latency fraud feature computation and long-term retention. Which design best meets these requirements on Google Cloud?

A global media company wants a governed data mesh on Google Cloud. Different domains publish datasets, but a central platform team must enforce consistent classification, lineage, and policy-based access controls across BigQuery, Cloud Storage, and Pub/Sub. They also want self-service discovery for analysts. What is the best approach?

A team wants to share curated analytics datasets across multiple projects. They need consistent access control and want to avoid copying data. Analysts should be able to query the shared datasets from BigQuery in their own projects. What should they do?

A retail company uses BigQuery for analytics. They accidentally deleted a table containing yesterday’s sales and need to restore it quickly with minimal operational work. What is the recommended approach?

You have a daily Dataflow batch pipeline reading from Cloud Storage and writing to BigQuery. It sometimes writes duplicate rows when the job is retried after transient failures. How should you best prevent duplicates in BigQuery?

A company needs to enrich streaming events with reference data that changes daily (e.g., product catalog). The pipeline must keep low latency and handle reference data updates without redeploying. Which approach is best in Dataflow?

A data science team deployed a Vertex AI model endpoint and wants to detect model drift by comparing recent prediction feature distributions to training data. They also want alerts when drift exceeds thresholds. What should they use?

Your organization stores raw and curated data in BigQuery and needs to ensure that analysts cannot accidentally query raw datasets containing PII unless they are explicitly approved. You need centralized governance and consistent policy enforcement across projects. What should you implement?

A batch pipeline writes partitioned BigQuery tables daily. Some downstream jobs fail because partitions occasionally arrive late and overwrite previously loaded data. You need an approach that supports late-arriving data while keeping downstream tables consistent and reproducible. What is the best design?

You are training a model using Vertex AI with a BigQuery table as the source. Training succeeds, but online predictions show lower accuracy than expected. Investigation reveals that training used a different feature transformation than serving. What is the most effective way to prevent this class of issue?

A financial services company must process streaming transactions with exactly-once outcomes in BigQuery for downstream risk calculations. Transactions may arrive out of order and occasionally be duplicated by upstream systems. Latency should be under a few seconds. Which design best meets these requirements?

A company operates multiple data pipelines (Dataflow, Dataproc, and BigQuery jobs). They need an enterprise-grade approach to monitor data quality (schema changes, null spikes, freshness SLAs) and trace lineage from raw sources to curated tables. They also need to surface this in a governed catalog for auditors. What is the best solution?

A data engineering team runs a nightly batch pipeline that writes curated Parquet files to Cloud Storage. Analysts query the data from BigQuery using external tables. Recently, query performance has degraded and costs increased due to repeated full scans. The team wants faster queries while keeping the same ingestion process and minimizing operational overhead. What should they do?

A streaming pipeline publishes events to Pub/Sub. A Dataflow streaming job reads from the subscription, enriches events, and writes results to BigQuery. The downstream BigQuery table shows occasional duplicate rows after Dataflow worker restarts. The business requires exactly-once writes at the logical record level. What is the best approach?

A Vertex AI model is deployed to an endpoint and performs well on offline validation data. After deployment, the team suspects data drift because prediction quality is declining. They want a managed solution that monitors feature distribution changes over time and alerts when drift exceeds thresholds, with minimal custom code. What should they use?

A regulated enterprise stores sensitive PII in BigQuery. Data scientists need to run aggregate analyses and build features, but company policy prohibits direct access to raw PII columns. The security team also wants to ensure users cannot infer individual identities from results. What is the best solution?

A data platform uses multiple microservices that publish events to Pub/Sub. A Dataflow pipeline consumes the events and writes to BigQuery. During incident reviews, the team struggles to trace a single business transaction across services, Pub/Sub, Dataflow, and BigQuery loads. They want improved end-to-end observability with correlation IDs and centralized log analysis. What should they implement?