Google Cloud Professional Data Engineer practice questions

Question 1

A data engineering team is designing a Pub/Sub pipeline and wants export subscriptions to land messages directly in Google Cloud resources without running a separate consumer. According to the subscription-type documentation, which two destinations can Pub/Sub export subscriptions write to directly? Select two.

Accepted Answer

A BigQuery table, written directly by a BigQuery export subscription

Accepted Answer

A Cloud Storage bucket, written by a Cloud Storage export subscription

Answer

A Spanner table, written directly by a Spanner export subscription

Answer

A Cloud SQL database, written directly by a Cloud SQL export subscription

Question 2

A staff engineer is evaluating Pub/Sub versus Pub/Sub Lite for a payments event stream where the design review demands the strongest message-durability guarantee and global availability. Given the constraints, which characterization is accurate?

Accepted Answer

Pub/Sub guarantees synchronous replication and offers global routing; Pub/Sub Lite trades global availability for cost.

Answer

Pub/Sub Lite alone guarantees synchronous replication, while standard Pub/Sub only replicates asynchronously per partition.

Answer

Both services guarantee synchronous replication, so the decision rests solely on each tier's per-message pricing.

Answer

Pub/Sub offers only asynchronous replication, while Pub/Sub Lite adds global routing for stronger durability overall.

Question 3

A platform team is designing an Object Lifecycle Management policy on a Cloud Storage bucket, and during testing one object simultaneously satisfies a Delete rule and two competing SetStorageClass rules. Given a scenario where several rules' conditions are met at once, which two statements describe how Cloud Storage resolves the conflict?

Accepted Answer

The Delete action takes precedence over any SetStorageClass action, so the matching object is deleted rather than reclassified.

Accepted Answer

Among competing SetStorageClass actions, the one moving the object to the class with the lowest at-rest storage pricing is applied.

Answer

Cloud Storage runs every matching action in sequence, first changing the object's storage class and then deleting the object on the same day.

Answer

Cloud Storage applies the SetStorageClass action that moves the object to the highest-performance storage class available.

Question 4

A senior data engineer is tasked with, in an on-call review, what the expected delay is between dropping a DAG file into the environment's bucket and the DAG becoming visible to the Airflow scheduler. Which option matches the Managed Airflow manage-DAGs documentation?

Accepted Answer

The synchronization and re-parsing of DAGs happen with a delay of approximately 1-2 minutes, depending on the number of files in the /dags folder and available resources.

Answer

The synchronization runs once per hour on a hard-coded schedule. The documentation says approximately 1-2 minutes, not hourly.

Answer

The synchronization is always instantaneous regardless of /dags folder size. The documentation says approximately 1-2 minutes, depending on /dags folder size and resources.

Answer

DAGs in the /dags folder are never automatically synced and require a manual gcloud command for every update. The documentation describes automatic sync with a 1-2 minute delay.

Question 5

A senior BI engineer is tasked with, during a design review, which workloads BI Engine cannot accelerate and which front-ends benefit most from it. Which option matches the BI Engine documentation?

Accepted Answer

BI Engine works effectively for BI dashboards and frequently queried tables, though it has limitations with wildcard queries, external tables, and row-level security features.

Answer

BI Engine only accelerates row-level security queries on external tables. The documentation explicitly lists row-level security and external tables among BI Engine's limitations.

Answer

BI Engine accelerates wildcard queries and external tables as a primary use case. The documentation explicitly lists wildcard queries and external tables among BI Engine's limitations.

Answer

BI Engine is documented as a write-acceleration cache for streaming inserts, not a read cache. BI Engine is a read-acceleration in-memory analysis service.

Question 6

A data engineering team is evaluating Cloud Data Fusion and needs to understand how compute is managed during a pipeline run so they can predict when clusters incur cost. According to the Cloud Data Fusion overview, how does Data Fusion handle the execution environment for a run?

Accepted Answer

It provisions an ephemeral Spark cluster at the start of the run and deletes it at the end

Answer

It runs every pipeline on one permanent cluster that is shared across all tenants in the region

Answer

It runs on a standing Dataflow Streaming Engine backend that is shared across Data Fusion tenants

Answer

It pushes all transformation logic down into BigQuery and never provisions any compute cluster

Question 7

A data platform team is designing an ELT layer in BigQuery and needs to expose Parquet files in Cloud Storage to analysts without loading the bytes into managed BigQuery storage. The team wants metadata managed inside BigQuery so dataset ACLs apply, while raw data continues to live in the bucket. Which BigQuery construct matches this design?

Accepted Answer

An external table, which stores metadata in BigQuery while data resides externally, exposing the Cloud Storage Parquet files through standard BigQuery dataset access controls.

Answer

A scheduled query that copies the Parquet files into a BigQuery native table every hour and drops the bucket-level dataset once the load completes. Copying the data into native storage abandons the requirement to keep raw data in the bucket.

Answer

A federated query that pushes the SQL down to the Cloud Storage REST API for every analyst session so BigQuery never has to hold any metadata about the files. Federated queries target operational databases, not Cloud Storage objects, and BigQuery does maintain metadata for files exposed through tables.

Answer

A BigQuery materialized view defined over a private REST endpoint so the Parquet files appear as a fully managed table inside BigQuery storage. Materialized views are precomputed from existing BigQuery sources and do not by themselves expose external Parquet files.

Question 8

A startup is designing a Firestore (Native mode) data model, and a developer is unsure whether every document in a collection must share the same set of fields. Which statement correctly describes Firestore's data model?

Accepted Answer

Firestore is schemaless, so documents in one collection may hold different fields and types, though consistent fields aid querying.

Answer

Every document within a collection must declare exactly the same fields, all of which are enforced by a stored schema definition object.

Answer

A single collection may directly contain both documents and other nested collections as its immediate members.

Answer

Field data types are fixed for the whole collection when it is created and cannot vary from one document to another.

Question 9

A senior data engineer is tasked with explaining why a maintained BigQuery reporting workload slows down at peak business hours even though the queries are unchanged. Which factor from the BigQuery performance best-practices documentation best explains the contention during peak load?

Accepted Answer

Capacity and concurrency, how many slots are available and how many other queries run during the same peak window.

Answer

Output materialization, the bytes a query writes, which is fixed by the result set and unaffected by other workloads.

Answer

Shuffling, the bytes passed between nodes, which depends only on the query plan and not on concurrent activity.

Answer

Input I/O, the bytes a query reads, which stays constant for the identical query regardless of when it runs.

Question 10

A platform team is rolling out a new Google Cloud data pipeline. A senior data engineer is being asked, in a design review for a new BigQuery streaming ingest pipeline, why new projects should prefer the Storage Write API to the legacy tabledata.insertAll streaming method. Which option matches the BigQuery legacy streaming API documentation?

Accepted Answer

For new projects, we recommend using the BigQuery Storage Write API instead of the tabledata.insertAll method, because the Storage Write API has lower pricing and more robust features, including exactly-once delivery semantics.

Answer

Both APIs charge identical prices and offer identical features, so new projects should pick whichever is convenient. The documentation states the Storage Write API has lower pricing and more robust features.

Answer

The legacy tabledata.insertAll method is the documented preferred path for new projects because it offers exactly-once semantics. The documentation recommends the Storage Write API for new projects.

Answer

tabledata.insertAll has been removed from BigQuery entirely. The documentation explicitly states that the legacy streaming API is still fully supported.

Question 11

A data team is rolling out a Pub/Sub ingest pipeline whose only consumer is BigQuery. The team wants to avoid running a Dataflow job just to forward messages and is looking for a Pub/Sub subscription type that writes directly to BigQuery. Which option matches the Pub/Sub subscriber documentation?

Accepted Answer

Export subscriptions, which directly write to Google Cloud resources without intermediate processing, supporting BigQuery, Bigtable, and Cloud Storage destinations for high-volume message delivery.

Answer

Push subscriptions that target a Cloud Run service that batches inserts into BigQuery. Push subscriptions deliver messages to an HTTPS endpoint and still require an application consumer in front of BigQuery.

Answer

Pull subscriptions, which the application code polls and writes into BigQuery in a custom worker loop. A pull subscription requires a custom consumer rather than direct delivery into BigQuery.

Answer

Push subscriptions that target a Pub/Sub-managed Cloud Function which then writes to BigQuery on every message. Pub/Sub does not provide an implicit Cloud Function endpoint that is distinct from a normal push subscription.

Question 12

An MLOps team is migrating a large batch pipeline to the service-based Dataflow shuffle and the architect is asked to summarize the documented benefits before production. Which two benefits match the Dataflow shuffle documentation? Select two.

Accepted Answer

It reduces the CPU, memory, and Persistent Disk storage consumed on the worker VMs.

Accepted Answer

It speeds up execution time for the majority of batch pipeline job types overall.

Answer

It guarantees exactly-once delivery for unbounded streaming sources that skip GroupByKey.

Answer

It removes any need to define GroupByKey or Combine transforms in the pipeline code.

Question 13

An organization runs a latency-sensitive Cloud SQL instance, and an administrator is tasked with identifying which class of Cloud SQL system update can cause downtime so it can be scheduled for a low-traffic window. Which statement matches Cloud SQL's update model?

Accepted Answer

Maintenance updates upgrade the OS and database program, require an instance restart, and so incur some schedulable downtime.

Answer

Hardware updates always force an instance restart, so these are precisely the updates that must be scheduled into a low-traffic window.

Answer

Online updates to the adjacent service agents require a restart and a brief period of downtime every time they are applied.

Answer

All three update categories run with zero downtime, so configuring a maintenance window is unnecessary for this instance.

Question 14

A data engineering team is designing a resilient streaming ingestion pipeline that reads from Pub/Sub and processes events in Dataflow. They must avoid permanently losing the small set of messages that repeatedly fail processing, and they must absorb large, unpredictable spikes in input volume without manually re-tuning the worker count. Select the two Google Cloud capabilities that match these requirements.

Accepted Answer

Configure a Pub/Sub dead-letter topic so messages exceeding the max delivery attempts are forwarded aside for later analysis.

Accepted Answer

Rely on Dataflow horizontal autoscaling so the job adjusts its worker count from the estimated work, re-evaluated periodically.

Answer

Replace Dataflow with a fixed-size Dataproc cluster provisioned for peak load, removing any need to react to volume spikes.

Answer

Schedule a BigQuery query through Cloud Scheduler to reprocess failed events on a nightly cron instead of during ingestion.

Question 15

A data governance team is tasked with explaining how to restrict access to sensitive PII columns in BigQuery so that only an HR group can read them, while leaving everyone else's access to non-sensitive columns intact. Which BigQuery capability matches this requirement?

Accepted Answer

Column-level access control, which provides fine-grained access to sensitive columns using policy tags, or type-based classification, of data, with access checks enforced at query time and policy tags managed in Data Catalog.

Answer

Drop the PII columns from BigQuery and store them in Memorystore for Redis behind a custom proxy. Removing columns sidesteps the documented BigQuery column-level access control feature.

Answer

Cluster the table by PII columns so BigQuery automatically restricts access. Clustering is a performance feature; it does not enforce access control.

Answer

Use a BigQuery scheduled query to copy non-PII columns into a separate table for everyone else. Per-column table splitting is exactly the anti-pattern column-level security is documented to avoid.

Google Cloud Professional Data Engineer

Sample questions

Sources

Similar exams