gcp data engineer Intermediate Practice Exam: Medium Difficulty 2025

A retail company is building a data platform where events from mobile apps must be processed in near real time for fraud detection, while also supporting daily batch reporting. The team wants a single pipeline design pattern that minimizes duplicated logic and works for both streaming and batch. Which approach should they choose?

A logistics company needs to ingest telemetry from 50,000 vehicles. Messages arrive out of order and can be delayed by up to 10 minutes. The company computes per-vehicle rolling metrics every 1 minute and must ensure correctness when late events arrive. Which solution best addresses these requirements?

A media company ingests clickstream events into BigQuery using streaming inserts. Analysts complain about occasional duplicate events caused by client retries. The company needs to reduce duplicates without building a complex deduplication pipeline. What should they do?

A financial services company runs nightly ETL on Dataproc that reads from Cloud Storage and writes outputs to BigQuery. The job occasionally fails midway, leaving partially written outputs that break downstream reporting. The company wants reruns to be safe and not produce duplicates or partial data. What is the best approach?

A company is building a data lake on Cloud Storage for semi-structured logs. They want cost-effective storage, the ability to evolve schemas, and fast SQL analytics in BigQuery without ingesting everything into native BigQuery tables immediately. Which approach best meets these needs?

An IoT platform stores time-series device data and needs very low-latency lookups for the most recent readings per device as well as high write throughput. Analysts occasionally run large aggregations, but the primary requirement is operational read/write performance. Which storage solution is the best fit for the primary workload?

A healthcare organization uses BigQuery for analytics and must enforce that only the compliance team can see direct identifiers (e.g., full name), while analysts can see de-identified data. They want to enforce this at query time without duplicating datasets. What should they implement?

A data team maintains a large BigQuery table partitioned by event_date. Query costs are increasing because analysts often filter by customer_id but not always by date. The team wants to improve performance and reduce scanned data without changing analyst behavior significantly. What should they do?

A company has multiple upstream systems producing files to Cloud Storage. A Dataflow job reads new files, transforms them, and loads results into BigQuery. The team needs automated orchestration with dependency management, retries, and visibility into failures across steps (file arrival, transform, load). Which solution is most appropriate?

A streaming pipeline on Dataflow writes aggregated metrics to BigQuery. After a recent change, the pipeline occasionally lags and autoscaling adds workers, increasing cost. The team wants to detect regressions early and troubleshoot bottlenecks (e.g., hot keys, backpressure, slow sinks) using managed observability. What should they do?