Google Professional Data Engineer Practice Test

Answer choices

1. A. Choose at-least-once mode without any deduplication because duplicate results are impossible in a streaming job.
2. B. Disable checkpointing because retries are the only source of duplicate output in a managed pipeline.
3. C. Keep exactly-once streaming mode for the pipeline and verify that the sink also handles duplicate-safe writes.
4. D. Use a local worker file as the authoritative result because external storage is always exactly once.

Wrong-answer review

• A. Choose at-least-once mode without any deduplication because duplicate results are impossible in a streaming job.: This option addresses a different concern and does not provide the documented behavior required for exactly-once and sinks in the stated workload.
• B. Disable checkpointing because retries are the only source of duplicate output in a managed pipeline.: This option addresses a different concern and does not provide the documented behavior required for exactly-once and sinks in the stated workload.
• D. Use a local worker file as the authoritative result because external storage is always exactly once.: Worker-local state is not a durable or portable control across retries, worker replacement, or distributed execution.

Answer choices

1. A. Enable message ordering because ordered delivery normalizes field names and timestamp formats.
2. B. Version a shared event schema covering field meaning, types, required values, and timestamps.
3. C. Store all fields as unvalidated free-form strings so no producer can violate the schema.
4. D. Increase subscriber concurrency so incompatible messages are parsed in parallel.

Answer choices

1. A. Evaluate a capacity-based BigQuery edition and size its reservation from observed usage and performance needs.
2. B. Use Cloud Storage lifecycle rules to cap the number of BigQuery slots.
3. C. Grant every analyst reservation administration so each user can resize capacity independently.
4. D. Remove all cost controls because stable demand guarantees on-demand bills never change.

Answer choices

1. A. Establish retry mechanisms and dead-letter queues within Cloud Composer to handle task failures during DAG execution.
2. B. Implement automated testing within the CI/CD pipeline to validate and synchronize DAG code before deployment.
3. C. Configure Pub/Sub topics and subscriptions to ensure reliable data delivery and decoupling of data producers.
4. D. Define delivery, acknowledgement, and retention policies within Pub/Sub subscriptions to optimize data processing.

Answer choices

1. A. Use a single Pub/Sub ordering key because message order creates Spark clusters and waits for job completion.
2. B. Write the job names in a wiki and ask an operator to submit them in order every day.
3. C. Use BigQuery table clustering because clustering orders Spark job execution.
4. D. Create a Managed Service for Apache Spark Workflow Template that expresses the job DAG and managed cluster lifecycle.

Answer choices

1. A. Replace embeddings with storage-class metadata because Archive and Standard values encode document meaning.
2. B. Discard document identifiers after embedding because vector values contain the original document location.
3. C. Store the document embeddings with stable document identifiers and use a suitable vector-similarity retrieval method to return the relevant chunks.
4. D. Use alphabetical sorting of document titles as the similarity metric.

Answer choices

1. A. Let each domain define unrelated sensitivity schemes so local autonomy is maximized.
2. B. Use shared catalog standards and metadata with decentralized domain stewardship and explicit data ownership.
3. C. Avoid recording ownership because certified data should be self-explanatory.
4. D. Remove domain stewardship and require the central team to approve every schema change in every project.

Answer choices

1. A. Share the broadest available dataset first because access boundaries can be inferred after publication.
2. B. Ignore update behavior because subscribers always receive the same data regardless of publisher changes.
3. C. Use a public link because a listing cannot express who is authorized to consume a data product.
4. D. Define authorized data, update behavior, access controls, and revocation before creating the listing.

Answer choices

1. A. Use BigQuery because analytical SQL warehouses are optimized for high-frequency OLTP transactions.
2. B. Use Cloud Storage because globally available objects provide row-level relational transactions.
3. C. Use Spanner for the globally distributed relational transactional workload.
4. D. Use Memorystore because a cache automatically provides durable cross-region relational consistency.

Answer choices

1. A. Expose only approved data through a governed view or data product; keep raw-dataset IAM restricted.
2. B. Make the raw dataset public but encrypt the restricted column names.
3. C. Grant the analysts project-level Data Owner and ask them to avoid the restricted columns.
4. D. Copy the raw dataset to each analyst's project so every consumer can implement a personal security policy.