Question bank
Professional Data Engineer
41 Professional Data Engineer questions with the correct answer and a written explanation, organised across 4 exam domains.
- 41 questions
- 4 domains
- Foundation level
certilumQuestion index
Jump straight to any question. Answers and explanations are inline.
Building and operationalizing data processing systems
12 questions in this domain
- Question 1Single Choice
Which Google Cloud service is a unified, serverless data-processing option for both batch and stream workloads?
- ACloud DataflowCorrect answer
- BCloud Dataproc
- CCloud Composer
- DCloud Data Fusion
Why this is the answer
Cloud Dataflow is the Google Cloud service that implements the Apache Beam model and supports both batch and streaming data processing in a fully managed, serverless way.
- Question 2Multiple Select
Which two Google Cloud services can be used to build managed data-processing pipelines?
- ACloud DataflowCorrect answer
- BCloud DataprocCorrect answer
- CCloud Storage
- DCloud IAM
Why this is the answer
Cloud Dataflow and Cloud Dataproc are managed services for data processing. Cloud Storage stores data and Cloud IAM governs access, so they are not processing services.
- Question 3Interactive
In the pipeline code shown, which transform is used to keep only the even numbers?
- Abeam.FilterCorrect answer
- Bbeam.Create
- Cbeam.Map
- Dbeam.CombineGlobally
Why this is the answer
beam.Filter applies a predicate and keeps elements for which the predicate returns true, so it selects even numbers in the pipeline.
- Question 4Single Choice
Which Apache Beam component is responsible for reading data from an external source in a Dataflow pipeline?
- AI/O connectorCorrect answer
- BPipeline
- CTransform
- DRunner
Why this is the answer
An I/O connector reads data from an external source and can also write results to a sink; transforms process data within the pipeline.
- Question 5Multiple Select
Which two Google Cloud services are designed for low-latency streaming data ingestion and processing?
- ACloud Pub/SubCorrect answer
- BCloud DataflowCorrect answer
- CCloud Storage
- DCloud SQL
Why this is the answer
Cloud Pub/Sub ingests streaming messages, while Cloud Dataflow processes those messages in a streaming pipeline. Cloud Storage and Cloud SQL do not provide streaming ingestion.
- Question 6Interactive
In the command shown, what is the name of the cluster that receives the Spark job?
- Aexample-clusterCorrect answer
- Bus-central1
- Corg.example.WordCount
- Dgs://bucket/input.txt
Why this is the answer
The --cluster flag identifies the Dataproc cluster where the Spark job is submitted; example-cluster is the cluster name.
- Question 7Single Choice
Which fully managed object storage service is typically used as a data lake for batch pipelines on Google Cloud?
- ACloud StorageCorrect answer
- BCloud SQL
- CCloud Spanner
- DBigQuery
Why this is the answer
Cloud Storage is Google Cloud's durable, object-based storage and is commonly used as a landing zone or data lake for batch processing.
- Question 8Multiple Select
Which two statements accurately describe Cloud Dataflow?
- AIt is fully managed and serverless.Correct answer
- BIt can automatically scale workers.Correct answer
- CIt requires a persistent cluster.
- DIt only supports batch data.
Why this is the answer
Cloud Dataflow is a fully managed, serverless service and can autoscale. It supports both batch and streaming workloads and does not need a persistent cluster.
- Question 9Interactive
The SQL statement creates a partitioned table. Which expression defines the partitioning column?
- ADATE(event_time)Correct answer
- Bevent_id
- Cevent_time
- Dproject.dataset.events
Why this is the answer
PARTITION BY DATE(event_time) uses the DATE function on event_time to define the partitioning column for the table.
- Question 10Single Choice
Which Google Cloud tool is a managed Apache Airflow environment used to orchestrate data workflows?
- ACloud ComposerCorrect answer
- BCloud Dataflow
- CCloud Dataproc
- DCloud Scheduler
Why this is the answer
Cloud Composer is Google Cloud's managed Apache Airflow service, used for authorizing, scheduling, and monitoring data workflows and pipelines.
- Question 11Multiple Select
Which two are valid reasons to choose Cloud Dataproc for a data-processing workload?
- ARun existing Hadoop/Spark jobs.Correct answer
- BMigrate on-premises Hadoop workloads.Correct answer
- CExecute serverless streaming pipelines.
- DProvide object storage for raw data.
Why this is the answer
Cloud Dataproc is ideal for running Hadoop and Spark jobs and for lifting existing on-premises workloads. Serverless streaming is better handled by Dataflow; object storage is Cloud Storage.
- Question 12Interactive
The command creates a Dataproc cluster. How many worker nodes does it request?
- A1
- B2Correct answer
- C3
- D4
Why this is the answer
The --num-workers flag sets the desired number of worker nodes in the cluster, and it is set to 2.
Designing data processing systems
9 questions in this domain
- Question 13Single Choice
Which Google Cloud service is a fully managed serverless data warehouse that can run SQL queries directly on large datasets?
- ABigQueryCorrect answer
- BCloud Storage
- CPub/Sub
- DCloud Spanner
Why this is the answer
BigQuery is Google Cloud's serverless data warehouse. It supports SQL queries and petabyte-scale analysis while separating storage from compute.
- Question 14Multiple Select
Which two statements correctly describe topics and subscriptions in Cloud Pub/Sub?
- AOne topic can have multiple subscriptions.Correct answer
- BA subscription pushes messages to the topic.
- CMultiple subscriptions can receive the same published message independently.Correct answer
- DA topic is created automatically every time a message is published.
Why this is the answer
A Pub/Sub topic is a named channel to which publishers send messages. Each subscription receives copies of messages, so multiple subscriptions can independently consume the same stream.
- Question 15Interactive
Look at the SQL query in the code block. Which join type is used to combine the two tables?
- ALEFT OUTER JOINCorrect answer
- BINNER JOIN
- CRIGHT OUTER JOIN
- DCROSS JOIN
Why this is the answer
The query uses LEFT JOIN, a synonym for LEFT OUTER JOIN, which returns all rows from the left table and only matching rows from the right.
- Question 16Single Choice
What is the main purpose of partitioning tables in BigQuery?
- AIt reduces the amount of data scanned by pruning partitions.Correct answer
- BIt automatically encrypts data at rest.
- CIt converts streaming data into structured data.
- DIt prevents users from updating table schemas.
Why this is the answer
Partitioning divides a table into segments based on a column, often date. Queries can prune scanned partitions, improving performance and lowering costs.
- Question 17Multiple Select
Which two practices help reduce cost when designing a Google Cloud batch data processing pipeline?
- AUse preemptible VMs for fault-tolerant Dataproc workloads.Correct answer
- BStore rarely accessed raw data in Cloud Storage Coldline.Correct answer
- CKeep a large pool of always-on Dataflow workers idle.
- DCopy every batch result into multiple world-wide buckets.
Why this is the answer
Preemptible VMs cost much less for short-lived batch jobs, and Cloud Storage Coldline is a low-cost storage class for archival data.
- Question 18Interactive
Examine the Apache Beam pipeline in the code block. Which transform produces a collection of word and count pairs?
- Abeam.combiners.Count.PerElement()Correct answer
- Bbeam.GroupByKey()
- Cbeam.CombineValues()
- Dbeam.io.ReadFromText()
Why this is the answer
Count.PerElement() is a beam combinator that counts how many times each element appears and outputs key-count pairs.
- Question 19Single Choice
Which Google Cloud service is primarily used to orchestrate data processing workflows with managed Apache Airflow?
- ACloud ComposerCorrect answer
- BDataflow
- CDataproc
- DCloud Functions
Why this is the answer
Cloud Composer is a managed Apache Airflow service used to schedule, monitor, and orchestrate complex data pipelines.
- Question 20Multiple Select
Which two schema design practices are recommended for BigQuery?
- AUse nested and repeated fields to represent arrays and relationships.Correct answer
- BDenormalize data when it reduces join complexity.Correct answer
- CStore timestamps as strings to improve readability.
- DCreate one table for every query result.
Why this is the answer
BigQuery performs well with denormalized data and native support for nested and repeated fields, which can reduce joins and improve query performance.
- Question 21Interactive
Inspect the Python snippet in the code block. What action does it perform in a Google Cloud data processing workflow?
- AIt uploads a string as an object to a Cloud Storage bucket.Correct answer
- BIt downloads a blob into memory from Cloud Storage.
- CIt deletes an object from a Cloud Storage bucket.
- DIt creates a Pub/Sub message from a Cloud Storage object.
Why this is the answer
The code gets the bucket and blob, then calls upload_from_string, which writes the given string as the content of the object.
Ensuring solution quality
12 questions in this domain
- Question 22Single Choice
Which Google Cloud service is used to create and enforce data quality rules across distributed data assets in a data mesh?
- ADataplexCorrect answer
- BDataflow
- CBigQuery
- DCloud Composer
Why this is the answer
Dataplex centralizes data governance across distributed storage systems. It lets you define and enforce data quality rules for BigQuery and Cloud Storage.
- Question 23Multiple Select
When operating production data pipelines, which two practices are part of a data quality strategy in Google Cloud? Choose two.
- AAutomated quality checksCorrect answer
- BPipeline monitoringCorrect answer
- CManual schema changes
- DDisabling audit logs
Why this is the answer
Automated quality checks verify completeness, freshness, and accuracy. Pipeline monitoring detects failures and delays. Manual changes and disabled logs reduce visibility and quality.
- Question 24Interactive
A data engineer runs the SQL query shown in the code block. What is the function of this query in BigQuery?
- ACounts rows in the tableCorrect answer
- BDeletes the table
- CUpdates column names
- DCreates a view
Why this is the answer
The SELECT COUNT(*) statement counts every row in the specified table. This common check is used to verify that data has been loaded correctly.
- Question 25Single Choice
Which Google Cloud service should be used to collect operational metrics and send alerts when a data pipeline is unhealthy?
- ACloud MonitoringCorrect answer
- BCloud Logging
- CData Catalog
- DDataplex
Why this is the answer
Cloud Monitoring collects performance metrics, builds dashboards, and sends alerts. This helps data engineers maintain solution quality by detecting unhealthy pipelines quickly.
- Question 26Multiple Select
Which two Google Cloud services allow data engineers to manage, search, and govern metadata for data assets? Choose two.
- AData CatalogCorrect answer
- BDataplexCorrect answer
- CCloud Storage
- DCloud Functions
Why this is the answer
Data Catalog provides metadata search and tagging. Dataplex extends governance with domains, policies, quality, and access controls, helping teams manage data assets across the organization.
- Question 27Interactive
A shell command runs after a data quality check in a Cloud Composer task. What is the effect of the command in the code block?
- AIt stops the pipeline with an errorCorrect answer
- BIt restarts the pipeline
- CIt logs a warning and continues
- DIt deletes the output data
Why this is the answer
The exit 1 command terminates the script with a nonzero status. This signals that quality validation failed and stops the pipeline from continuing with bad data.
- Question 28Single Choice
What mechanism in Google Cloud controls which users or services are allowed to access data resources and perform actions?
- AIAM rolesCorrect answer
- BDataflow jobs
- CCloud Scheduler
- DBigQuery slots
Why this is the answer
Identity and Access Management (IAM) controls who can perform actions on resources. Configuring IAM roles is a core part of governing Google Cloud data assets.
- Question 29Multiple Select
Which two actions should be taken when loading data into a data lake to avoid introducing poor-quality data? Choose two.
- AValidate schema on ingestCorrect answer
- BCheck for nulls and duplicatesCorrect answer
- CSkip data type checks
- DIgnore record source
Why this is the answer
Validating schemas and checking nulls or duplicates at ingestion prevents bad data from entering the lake. Skipping checks or ignoring source produces lower-quality datasets.
- Question 30Interactive
What does the gcloud command in the code block retrieve from Cloud Monitoring when a data engineer executes it?
- ADataflow job errorsCorrect answer
- BPub/Sub message backlogs
- CCloud Storage bucket sizes
- DBigQuery slot usage
Why this is the answer
The gcloud command filters metrics by the Dataflow job error_count metric descriptor. It returns information about errors in Dataflow jobs, which is useful for quality monitoring.
- Question 31Single Choice
Which Google Cloud service inspects data and identifies sensitive information such as personally identifiable information in stored data?
- ACloud DLPCorrect answer
- BCloud Monitoring
- CCloud Composer
- DCloud Shell
Why this is the answer
Cloud DLP inspects data streams and repositories to find sensitive patterns. It can classify, mask, or de-identify protected data, reducing compliance risk and supporting governance.
- Question 32Multiple Select
Which two practices help ensure that datasets used for analysis and machine learning remain accurate over time? Choose two.
- ADetect schema driftCorrect answer
- BProfile data for anomaliesCorrect answer
- CRemove all null values blindly
- DDisable database logging
Why this is the answer
Schema drift detection catches unexpected changes in data structure, while profiling identifies anomalies. Both practices help keep analytical datasets accurate over time.
- Question 33Interactive
The Apache Beam pipeline in the code block reads text data and filters it before writing output. Which values does it remove?
- AEmpty stringsCorrect answer
- BDuplicate keys
- CHigh-value records
- DInvalid timestamps
Why this is the answer
The Beam Filter step keeps records only if they are not equal to an empty string. Empty strings are therefore removed before the remaining data is written to output.
Operationalizing machine learning models
8 questions in this domain
- Question 34Single Choice
What is the primary purpose of the Vertex AI Feature Store on the Google Cloud platform?
- ACentral repository for storing, sharing, and reusing ML featuresCorrect answer
- BTool for manually labeling large training datasets
- CService for creating custom visualization dashboards
- DInfrastructure for running SQL queries on transactional data
Why this is the answer
Vertex AI Feature Store provides a centralized place to store, share, and reuse ML features, reducing duplicate engineering work and supporting online and offline serving.
- Question 35Multiple Select
Which practices are recommended when operationalizing machine learning models on the Google Cloud platform in production?
- AAutomating retraining using Vertex AI PipelinesCorrect answer
- BRegistering model artifacts in Vertex AI Model RegistryCorrect answer
- CMonitoring model performance and data driftCorrect answer
- DDeploying directly from local notebooks without version tracking
Why this is the answer
Operationalization relies on automation, versioning, and monitoring. MLOps best practices include pipeline automation, central model registry, and continuous drift monitoring.
- Question 36Interactive
Analyze the Python snippet and identify the parameter that defines the target endpoint for online prediction.
- Aendpoint_nameCorrect answer
- Binstances
- Cpredict
- Daiplatform
Why this is the answer
The endpoint_name argument passes the fully qualified endpoint resource path, determining which deployed model receives the online prediction request.
- Question 37Single Choice
What is the main purpose of using batch prediction in Vertex AI on Google Cloud?
- AGenerating predictions for a large dataset asynchronouslyCorrect answer
- BServing real-time predictions with minimal latency
- CTraining models on tabular data
- DVisualizing model evaluation metrics
Why this is the answer
Batch prediction processes large amounts of data offline and asynchronously, making it suitable for jobs that do not require immediate responses.
- Question 38Multiple Select
Which items can be associated with a model version in Vertex AI Model Registry on Google Cloud?
- AModel artifactCorrect answer
- BServing containerCorrect answer
- CVersion aliasCorrect answer
- DBigQuery destination table
Why this is the answer
A model version includes the serialized model artifact, a serving container specification, and optional version aliases that support safe deployment and serving.
- Question 39Interactive
Analyze the JSON configuration and identify the field that specifies the source of the input data for a batch prediction job.
- AinputPathCorrect answer
- BoutputPath
- Cmodel
- DpredictionsFormat
Why this is the answer
The inputPath field points to the storage location of the input data, allowing the batch prediction job to read instances from the specified path.
- Question 40Single Choice
What is the primary function of the Vertex AI Model Registry in MLOps workflows on Google Cloud?
- AManaging the lifecycle and versions of trained modelsCorrect answer
- BStoring raw training data in a data warehouse
- CGenerating feature transformations for models
- DExecuting scheduled SQL queries
Why this is the answer
Vertex AI Model Registry provides a central repository to manage model artifacts, versions, and deployment lifecycle, helping teams control releases.
- Question 41Multiple Select
Which signals indicate that a deployed machine learning model may need retraining on Google Cloud?
- APrediction accuracy drops over timeCorrect answer
- BData drift is detected in input featuresCorrect answer
- CPrediction skew is observed against training dataCorrect answer
- DCloud storage costs increase suddenly
Why this is the answer
A model should be retrained when performance degrades or the data distribution shifts. Storage costs alone do not indicate a modeling problem.
Frequently asked questions
Everything about the Professional Data Engineer question bank and how to use it.