Google Cloud Professional Data Engineer Practice Test

Professional Data Engineer practice test: what's covered

• Professional Data Engineer

Free Professional Data Engineer sample questions

1. A gaming backend must store player profile lookups by user ID with single-digit millisecond latency at over 500,000 reads per second, with no complex joins. Which database fits best?

   • A. Cloud Bigtable, a wide-column NoSQL store tuned for high-throughput key-based access
   • B. BigQuery, because it scales storage and compute independently for analytics
   • C. Cloud SQL for PostgreSQL with several cross-region read replicas attached
   • D. Cloud Storage with JSON objects fetched directly by the player ID prefix
   Show answer

   Correct: A. Bigtable delivers single-digit millisecond latency for key-based reads at massive throughput, which is exactly this access pattern. BigQuery is an analytical warehouse, not a low-latency point-lookup store, so it is incorrect here. Cloud SQL does not scale to hundreds of thousands of reads per second on a single key, unlike Bigtable. Cloud Storage is object storage with no millisecond random-read database semantics.

2. Your team wants a single SQL interface and consistent governance over Parquet and Iceberg files in Cloud Storage, queried from BigQuery, Spark, and Trino. Which approach achieves this?

   • A. Define BigLake tables over the Cloud Storage data so access policies apply across engines
   • B. Load every file into native BigQuery tables and drop the Cloud Storage copies
   • C. Grant each engine its own bucket IAM role and manage policies separately
   • D. Export the files to Bigtable so all engines read through one row-key index
   Show answer

   Correct: A. BigLake tables provide a unified governance and performance layer over open formats in Cloud Storage, enforcing fine-grained access across BigQuery, Spark, and Trino. Loading into native tables abandons the open-format requirement and duplicates data, so it is incorrect. Per-bucket IAM cannot give column or row-level controls, unlike BigLake. Bigtable is a NoSQL store and cannot query Parquet or Iceberg files.

3. You are choosing managed databases for a global retail platform. Which statements about GCP storage services are accurate? (Choose TWO)

   • A. Cloud Spanner offers horizontally scalable relational SQL with strong consistency across regions
   • B. Firestore is a serverless document database well suited to mobile and web app state
   • C. Bigtable enforces multi-statement ACID transactions across arbitrary row ranges
   • D. Cloud SQL automatically shards a single database across multiple regions for writes
   • E. BigQuery is the recommended primary OLTP store for high-frequency transactional writes
   Show answer

   Correct: A and B. Spanner combines relational SQL with horizontal scale and strong global consistency, and Firestore is a serverless document database ideal for app state. Bigtable provides only single-row atomicity, not arbitrary multi-row ACID transactions, so that option is incorrect. Cloud SQL does not auto-shard writes across regions, unlike Spanner. BigQuery is an analytical warehouse and is not an OLTP write store, whereas Spanner or Cloud SQL would be.

4. An enterprise is migrating an on-premises Hadoop and Spark estate to GCP and wants minimal code rewrites in the first phase. Which initial strategy is most appropriate?

   • A. Lift the existing Spark and Hive jobs onto Dataproc, then modernize incrementally toward BigQuery and Dataflow
   • B. Rewrite every job as Apache Beam pipelines on Dataflow before any data is moved
   • C. Reimplement all logic as BigQuery SQL and decommission Spark on day one
   • D. Run the existing cluster unchanged on Compute Engine VMs with no managed services
   Show answer

   Correct: A. Dataproc runs open-source Spark and Hive with minimal changes, enabling a phased lift-and-modernize migration. Rewriting everything to Beam first delays migration and adds risk, so it is incorrect for an initial low-rewrite phase. Reimplementing all logic as SQL on day one is not minimal-effort, unlike Dataproc. Running unmanaged VMs forgoes managed-service benefits and is not the recommended path.

5. A telemetry system must absorb spiky bursts of up to one million events per second, decouple producers from consumers, and feed a streaming transform. Which ingestion service belongs at the front of this pipeline?

   • A. Pub/Sub, which buffers and decouples publishers from subscribers at global scale
   • B. Cloud Storage, by appending each event as a new object in a single bucket
   • C. Cloud SQL, by inserting each event row inside a synchronous transaction
   • D. BigQuery legacy streaming inserts called directly from every producer device
   Show answer

   Correct: A. Pub/Sub is built to buffer high-throughput bursts and decouple producers from downstream consumers, which is exactly the front-of-pipeline need. Cloud Storage object-per-event is not a real-time streaming buffer, so it is incorrect. Cloud SQL cannot sustain a million synchronous inserts per second, unlike Pub/Sub. Calling BigQuery streaming inserts from every device couples producers tightly and lacks buffering.

Last updated: 2026-10-19

Professional Data Engineer practice test FAQ

Related practice tests

• Google Cloud ACE practice test
• Google Cloud Professional Cloud Architect practice test
• Google Professional Cloud DevOps Engineer practice test
• Google Professional Cloud Network Engineer practice test
• Google Cloud PCSE practice test
• Google Cloud GenAI-Leader practice test