Professional Data Engineer Free Practice Questions
This is the free Google Professional Data Engineer practice question bank —
170 of 332 total questions, each with a full explanation, free to
read with no signup required. Updated 2026-08-04.
Every answer is verified against official Google documentation —
see our methodology.
Question 1
You are using Bigtable to persist and serve stock market data for each of the major indices. To serve the trading application, you need to access only the most recent stock prices that are streaming in. How should you design your row key and tables to ensure that you can access the data with the simplest query?
A. Create one unique table for all of the indices, and then use the index and timestamp as the row key design.
B. Create one unique table for all of the indices, and then use a reverse timestamp as the row key design.
C. For each index, have a separate table and use a timestamp as the row key design.
D. For each index, have a separate table and use a reverse timestamp as the row key design.
Show Answer
Correct Answer: A
Explanation: Cloud Bigtable schema design recommends using a single table for related data and avoiding row keys that start with a (normal or reversed) timestamp because they create hotspots. Prefixing the row key with the index distributes writes. Although reversed timestamps are useful for reading the newest records first, options B and D start the row key with a reversed timestamp, which is discouraged. Among the available choices, using a single table with a row key beginning with the index is the best match.
Question 1
You monitor and optimize the BigQuery instance for your team. You notice that a particular daily report that uses a large JOIN operation is consistently slow. You want to examine the query's execution plan to identify potential performance bottlenecks within the JOIN as quickly as possible. What should you do?
A. Use the bg query --dry_run command to review the estimated number of bytes read and review query syntax.
B. Run a query on the INFORMATION_SCHEMA.JOBS_BY_PROJECT view filtering by the job_ID and analyze total_bytes_processed.
C. Review the BigQuery audit logs in Cloud Logging.
D. Leverage BigQuery's Query History view and analyze the execution graph.
Show Answer
Correct Answer: D
Explanation: The Query History execution graph (query plan) provides stage-by-stage execution details, including JOIN, shuffle, skew, slot usage, and other execution metrics that help identify performance bottlenecks quickly. A dry run only estimates bytes processed, INFORMATION_SCHEMA.JOBS views expose job metadata rather than the detailed execution graph, and audit logs are for auditing and operations rather than query plan analysis.
Question 2
Your company stores vital operational sales data in a BigQuery dataset in us-central1. Your company requires a disaster recovery plan to restore this data to us-east1 with a recovery point objective (RPO) of 24 hours and a recovery time objective (RTO) of 4 hours if us-central1 experiences an outage. You need to implement the disaster recovery plan while keeping costs and complexity to a minimum. What should you do?
A. Set up continuous queries and Pub/Sub to stream data changes from BigQuery tables in us-central1 to us-east1.
B. Manually export data to a CSV file in a multi-regional Cloud Storage bucket daily and use bq load to restore to us-east1.
C. Configure BigQuery cross-region dataset replication from ns-central1 to us-east1.
D. Take daily BigQuery table snapshots in us-central1.
Show Answer
Correct Answer: C
Explanation: BigQuery cross-region dataset replication is designed for disaster recovery across regions with low operational overhead. Replication to us-east1 satisfies the 24-hour RPO requirement and enables much faster recovery than manual exports or snapshots, while avoiding the complexity of building a custom streaming pipeline. Daily exports and snapshots do not provide a cross-region ready-to-use dataset with the required recovery characteristics.
Question 3
You are using Dataflow to build a streaming data pipeline to analyze user website click activity from Pub/Sub. You need to calculate the number of clicks for each user site visit. A site visit is defined as a period of activity followed by 30 minutes of inactivity for a specific user. What should you do?
A. Use tumbling windows with a 30-minute window.
B. Use hopping windows with a 30-minute window, and a 1-minute period.
C. Use hopping windows with a 30-minute window, and a 30-minute period.
D. Use session windows with a 30-minute gap duration.
Show Answer
Correct Answer: D
Explanation: A session window is specifically designed to group events separated by periods of inactivity. Defining a 30-minute gap duration means all clicks from the same user are counted in the same session until there has been 30 minutes with no activity, which matches the site visit definition. Tumbling and hopping windows use fixed time boundaries and do not adapt to user inactivity.
Question 4
You are defining the data governance strategy for a new BigQuery table with medical and financial data. You want a scalable solution that ensures the clinical researchers can access patient medical data without financial information, while allowing the accounting team to access only financial data with minimal patient identifiers. What should you do?
A. Implement column-level security policies in BigQuery tables with IAM permissions.
B. Create separate tables for personally identifiable information (PII), financial data, and anonymized medical data. Use IAM permissions to control access to each table.
C. Implement row-level security policies in BigQuery tables with IAM permissions.
D. Create separate datasets with authorized views exposing only approved data.
Show Answer
Correct Answer: A
Explanation: Column-level security in BigQuery, implemented with Data Catalog policy tags and enforced through IAM, is designed to restrict access to specific columns within the same table. This provides a scalable solution where clinical researchers can see medical columns but not financial columns, while accounting can access financial columns and only the minimal identifier columns they require. Row-level security filters records rather than columns, and splitting data into separate tables or relying on authorized views adds more maintenance overhead and is less scalable for this use case.
Question 5
You are designing a data lake on Google Cloud to store vast amounts of customer interaction data from various sources, such as websites, mobile apps, and social media. You need to ensure that this data, which arrives in different formats, is consistently cataloged and easy for data analysts to discover and use. You also want to perform basic data quality checks and transformations before the data is consumed by downstream applications. You need an automated and managed data governance solution. What should you do?
A. Use Cloud Storage as the central repository. Use Vertex AI to classify and process the data and perform data quality checks.
B. Stream all the data directly into BigQuery, where it is automatically cataloged and governed.
C. Use Cloud Storage and BigQuery as repositories. Use Dataplex Universal Catalog for metadata discovery, data quality checks, and transformations.
D. Use Cloud Storage as the central repository. Use a Cloud Run function to catalog, transform the data, and perform data quality checks.
Show Answer
Correct Answer: C
Explanation: Dataplex is Google Cloud's managed data fabric and governance service for data lakes. It provides a universal catalog for metadata discovery, supports data quality scanning, and integrates with Cloud Storage and BigQuery while enabling managed governance and transformations. The other options either use services not intended for catalog/governance (Vertex AI, Cloud Run) or incorrectly imply BigQuery alone provides the required governance across heterogeneous lake data.
Question 6
Your retail company is concerned about their BigQuery analytics spend. The company runs several queries that require the use of the same aggregation for the store ID and real-time sales volume. You need to implement the optimal solution that minimizes analytics spend and returns faster results. What should you do?
A. Create a new table from a CSV file with the repeated aggregation for the other queries to reference for faster processing.
B. Create a materialized view to minimize repetitive computations.
C. Use join acceleration with primary and foreign keys to increase query joining to live data.
D. Leverage partitioning to minimize the number of bytes read.
Show Answer
Correct Answer: B
Explanation: Materialized views precompute and incrementally maintain aggregation results in BigQuery. When multiple queries repeatedly use the same aggregation, they reduce repeated computation, lower query costs, and improve performance compared with recomputing the aggregation each time. Creating a separate CSV table is manual and not automatically maintained, join acceleration is unrelated to repeated aggregations, and partitioning reduces scanned data but does not eliminate repeated aggregation work.
Question 7
You are preparing data to serve a sales demand prediction model. The training data undergoes several pre-processing steps, including scaling numerical features and one-hot encoding categorical features. The model is deployed on Vertex AI Endpoints. You need to prevent training-serving skew and ensure accurate predictions in production. You want a solution that is easy to implement.
What should you do?
A. Implement a custom handler within the Vertex AI Endpoint to automatically perform data transformations before the model makes a prediction.
B. Replicate the exact same pre-processing logic in the inference pipeline that was used during model training.
C. Store the raw, unprocessed data in a separate Cloud Storage bucket exclusively for serving.
D. Ensure the serving data is a smaller, random sample of the training data.
Show Answer
Correct Answer: B
Explanation: Training-serving skew is avoided by applying the same preprocessing transformations at inference that were used during training. Replicating the identical scaling and one-hot encoding logic ensures the model receives inputs in the same representation. The other options do not address consistency of preprocessing; while embedding preprocessing in serving can work, the key requirement is identical preprocessing, making B the best answer.
Question 8
You are building a system to process financial transactions. The system must handle a high throughput of concurrent user operations and each operation requires low-latency reads and writes to individual records. You need to identify a storage solution that guarantees ACID compliance for the processed transactions. You want a Google Cloud managed service. What should you do?
A. Select Bigtable.
B. Select BigQuery.
C. Select Cloud Storage.
D. Select Spanner.
Show Answer
Correct Answer: D
Explanation: Cloud Spanner is Google Cloud's fully managed relational database designed for high-throughput, low-latency transactional workloads with strong consistency and full ACID transactions. Bigtable is a NoSQL database and does not provide full relational ACID transactions across rows. BigQuery is an analytical data warehouse, not an OLTP database. Cloud Storage is object storage and is not suitable for transactional record processing.
Question 9
Your organization has thousands of images stored in Cloud Storage. The organization's data analysts need to categorize these images upon analyzing the image content and store the results in a structured format for integration with other BigQuery datasets. You need a solution that allows the analysts to perform this task with the least amount of coding. What should you do?
A. Use the Gemini 2.5 Pro model in Vertex AI to analyze the images and store the results in BigQuery.
B. Use the ML.UNDERSTAND_TEXT function along with object tables to analyze the images, and store the JSON output in a Cloud Storage bucket.
C. Use Vertex AI Vision API to analyze the images and store the results in BigQuery.
D. Use the ML.GENERATE_TEXT function along with object tables to analyze the images and store the JSON output in a BigQuery table.
Show Answer
Correct Answer: D
Explanation: ML.GENERATE_TEXT in BigQuery can use object tables that reference images in Cloud Storage and invoke Gemini models to analyze image content directly from SQL. The generated JSON can be written into a BigQuery table, making it easy to integrate with other BigQuery datasets while requiring minimal custom code. ML.UNDERSTAND_TEXT is for text understanding rather than image analysis, and using Vision API or Vertex AI directly would generally require more application code.
$19
Get all 332 questions with detailed answers and explanations
Instant download HTML + PDF delivered the moment payment clears.
Secure Stripe checkout we never see or store your card details.
7-day refund if files are defective see our refund policy.