What does the GCP Professional Data Engineer exam cover?
The Google Cloud Professional Data Engineer exam covers designing, building, operationalizing, securing, and monitoring data processing systems on GCP. It emphasizes BigQuery, Dataflow, Pub/Sub, Dataproc, and machine learning integration, testing candidates on both batch and streaming pipeline architectures as well as data governance and compliance practices.
The Google Cloud Professional Data Engineer certification validates expertise in building data systems that extract business value from large, complex datasets. As organizations increasingly shift analytics workloads to the cloud, certified GCP data engineers command significant salary premiums and are positioned at the intersection of data engineering, cloud infrastructure, and applied machine learning. Payscale data for 2025 shows GCP Professional Data Engineer certified professionals earning a median base salary of $145,000 in the United States [1].
This guide presents the full exam domain structure, service-level knowledge requirements, preparation resources, and the mental models needed to answer scenario-based questions accurately. Material is drawn from Google's official exam guide [2], the BigQuery documentation [3], Apache Beam documentation [4], and the Tutorials Dojo data engineer exam review [5].
Exam Logistics
| Attribute | Detail |
|---|---|
| Exam cost | $200 USD |
| Exam duration | 120 minutes |
| Number of questions | 50-60 multiple-choice and multiple-select |
| Validity period | 2 years |
| Delivery | Remote proctored or test center |
| Prerequisites | None (Google recommends 3+ years data engineering experience) |
| Recommended prior knowledge | SQL, Python or Java, distributed systems concepts |
The Professional Data Engineer exam is notably more difficult than the associate-level ACE exam. Candidates without a data engineering background consistently report underestimating the depth of BigQuery and Dataflow knowledge required. Plan for 10-16 weeks of preparation if data engineering is not your primary daily work.
Exam Domains
| Domain | Title | Approximate Weight |
|---|---|---|
| 1 | Designing data processing systems | 22% |
| 2 | Ingesting and processing data | 25% |
| 3 | Storing the data | 20% |
| 4 | Preparing and using data for analysis | 15% |
| 5 | Maintaining and automating data workloads | 18% |
Domain 1: Designing Data Processing Systems (22%)
This domain tests your ability to select and architect the right GCP services for a given data engineering scenario.
Batch vs. Streaming Architecture Decisions
The most fundamental decision in data engineering is whether a workload is batch or streaming. The GCP Professional Data Engineer exam presents scenarios and requires identifying the correct processing model:
- Batch processing: Data is collected over a period, then processed at once. Use Dataflow batch pipelines (Apache Beam), Dataproc (Apache Spark), or BigQuery batch queries. Appropriate when latency of hours or days is acceptable.
- Streaming processing: Data is processed as it arrives, continuously. Use Dataflow streaming pipelines with Pub/Sub as the message bus. Appropriate when real-time dashboards, alerts, or downstream systems require sub-minute latency.
- Micro-batch: A hybrid where streaming data is buffered and processed in small batches every few seconds. Dataflow supports this via windowing operations.
Service Selection Framework
| Use Case | Primary Service | Alternative |
|---|---|---|
| Serverless SQL analytics | BigQuery | -- |
| Streaming ETL pipeline | Dataflow (Beam) | -- |
| Managed Spark/Hadoop | Dataproc | -- |
| Message queuing | Pub/Sub | -- |
| Relational OLTP | Cloud SQL or Cloud Spanner | -- |
| Wide-column NoSQL | Cloud Bigtable | -- |
| Document NoSQL | Firestore | -- |
| In-memory cache | Memorystore (Redis) | -- |
| Data orchestration | Cloud Composer (Airflow) | -- |
Schema Design
BigQuery schema design questions appear frequently. Key concepts:
- Denormalized schemas are preferred in BigQuery for analytical performance; normalization that reduces cost on OLTP systems increases query complexity and cost in columnar stores
- Nested and repeated fields (STRUCT and ARRAY types) reduce the need for joins in BigQuery
- Partitioning strategies: ingestion-time partitioning, column partitioning (DATE/TIMESTAMP/INTEGER), and range partitioning
- Clustering: sorting data within partitions by up to four columns to reduce bytes scanned for filtered queries
Domain 2: Ingesting and Processing Data (25%)
The highest-weighted domain covers pipeline construction using GCP services, with heavy emphasis on Dataflow and Pub/Sub.
Pub/Sub Architecture
Pub/Sub is GCP's managed asynchronous messaging service. Key concepts for the exam:
- Topics receive messages from publishers; subscriptions deliver messages to consumers
- Pull subscriptions: the consumer polls for messages; appropriate for batch consumers or consumers with variable load
- Push subscriptions: Pub/Sub pushes messages to an HTTPS endpoint; appropriate for Cloud Run and App Engine consumers
- Message ordering: Pub/Sub Lite and the ordering key feature in standard Pub/Sub enable ordered delivery when required
- Message retention: up to 7 days; unacknowledged messages are redelivered
- Dead-letter topics: configure to handle messages that fail processing after a set number of attempts
"Pub/Sub guarantees at-least-once delivery. This means consumers must implement idempotency to handle duplicate messages correctly. Designing idempotent Dataflow pipelines is a critical skill for this exam." -- Google Cloud Pub/Sub documentation [6]
Dataflow and Apache Beam
Dataflow is the managed service for Apache Beam pipelines. The exam tests both conceptual understanding and specific Beam concepts:
- PCollections: the distributed dataset abstraction in Beam; both batch and streaming pipelines use PCollections
- Transforms: operations applied to PCollections. Common transforms: ParDo (element-wise), GroupByKey, Combine, Flatten, Partition
- Windows: for streaming pipelines, windows group elements by time. Types: Fixed (tumbling), Sliding (hopping), Session (gap-based)
- Triggers: determine when to emit windowed results. Event time triggers fire based on watermarks; processing time triggers fire on wall clock
- Watermarks: estimate how far behind real-time the pipeline's event time is. Late data beyond the watermark is handled by allowed lateness and side output
Dataflow templates allow non-engineers to run parameterized pipelines without code deployment. Flex templates (container-based) support custom dependencies; classic templates have limitations.
Dataproc
Dataproc is Google's managed Hadoop and Spark service. Key differentiators from Dataflow:
- Use Dataproc when migrating existing Spark or Hadoop workloads without rewriting in Apache Beam
- Dataproc clusters can be ephemeral: create for a job, delete immediately after; use Cloud Storage for persistent data instead of HDFS
- Dataproc Serverless eliminates cluster management for Spark batch workloads
- Dataproc on GKE runs Spark workloads on existing GKE clusters
Domain 3: Storing the Data (20%)
Storage selection requires matching access patterns, latency requirements, and scale to the correct GCP service.
BigQuery Storage
BigQuery stores data in a columnar format called Capacitor across Google's distributed file system. Key storage concepts:
- Storage billing: BigQuery charges for active storage (data modified in the last 90 days) and long-term storage (data not modified for 90+ days at 50% discount)
- Authorized views: allow sharing query results with users who do not have access to the underlying tables
- Row-level security: row access policies restrict which rows different users can query from the same table
- Column-level security: policy tags on sensitive columns enforce access control integrated with Cloud DLP
Bigtable
Cloud Bigtable is a wide-column NoSQL database optimized for high-throughput, low-latency read/write workloads. Key design principles:
- Row key design is the most critical performance factor; poor row key design causes hotspotting
- Avoid sequential row keys (timestamps, sequential IDs) as the first component; prefix with a hash or reverse the string
- Column families group related data and have independent garbage collection policies
- Bigtable scales linearly by adding nodes; each node handles approximately 10,000 operations per second for reads, 10,000 for writes
Storage Class Selection Summary
| Service | Best For | Latency | Scale |
|---|---|---|---|
| BigQuery | Analytical SQL | Seconds | Petabytes |
| Bigtable | High-throughput NoSQL | Milliseconds | Petabytes |
| Cloud SQL | Relational OLTP | Milliseconds | Terabytes |
| Cloud Spanner | Global relational OLTP | Milliseconds | Petabytes |
| Firestore | Document NoSQL | Milliseconds | Terabytes |
| Cloud Storage | Object/file storage | Variable | Exabytes |
Domain 4: Preparing and Using Data for Analysis (15%)
This domain bridges data engineering and data science, testing knowledge of BigQuery ML, Looker, and data quality practices.
BigQuery ML
BigQuery ML enables training and serving machine learning models using SQL syntax directly within BigQuery. The exam tests:
- Supported model types: linear regression, logistic regression, k-means clustering, matrix factorization, deep neural networks, XGBoost, and AutoML Tables integration
- The CREATE MODEL statement and model evaluation using ML.EVALUATE
- When to use BigQuery ML vs. Vertex AI: BigQuery ML for SQL-native, smaller-scale model training on BigQuery data; Vertex AI for custom training with TensorFlow/PyTorch, larger models, or production serving with SLAs
Looker and Data Studio
- Looker uses LookML, a proprietary modeling language, to define business logic and metrics in a governed data layer
- Looker Studio (formerly Data Studio) provides free, no-code dashboard creation connected to BigQuery and other sources
- The exam distinguishes between these tools based on governance requirements: Looker for enterprise governed metrics; Looker Studio for ad-hoc exploration
Data Quality and Validation
- Cloud Data Quality (Cloud DQ) enables defining and running data quality rules against BigQuery tables
- Dataplex provides a unified data management layer for organizing data lakes, enforcing data quality, and managing metadata across Cloud Storage and BigQuery
- Data Catalog provides metadata management and data discovery across GCP and hybrid environments
Domain 5: Maintaining and Automating Data Workloads (18%)
Pipeline Orchestration
Cloud Composer is the managed Apache Airflow service on GCP. Key concepts:
- DAGs (Directed Acyclic Graphs) define workflow dependencies in Python
- Operators: BigQueryOperator, DataflowJavaOperator, DataprocSubmitJobOperator, PubSubPublishMessageOperator
- Composer environments run on GKE; multiple environments are used for dev/staging/production separation
"Cloud Composer is the standard answer for orchestrating multi-step data pipelines on GCP that span multiple services. When a scenario involves dependencies between pipeline stages, automated scheduling, and retry logic, Cloud Composer is almost always the correct service." -- Tutorials Dojo Data Engineer exam review [5]
Monitoring Data Pipelines
- Dataflow monitoring via the Dataflow UI and Cloud Monitoring: track job state, element counts, wall time, and system lag for streaming jobs
- BigQuery slot utilization monitoring: identify queries consuming excessive slots and optimize with query plans
- Pub/Sub subscription metrics: oldest unacked message age (key SLO indicator for streaming pipelines)
Cost Optimization
- BigQuery on-demand pricing charges per byte scanned; use partitioning and clustering to minimize bytes scanned
- BigQuery reservations (flat-rate pricing) with slot commitments for predictable costs at high usage
- Dataflow Flex Resource Scheduling (FlexRS) reduces batch pipeline costs by 40% using preemptible VMs during off-peak hours
- Dataproc spot (preemptible) workers for cost-sensitive batch jobs with checkpointing
Preparation Resources
| Resource | Type | Best For |
|---|---|---|
| Google Cloud Skills Boost Data Engineer Path | Hands-on labs | Service-level practice |
| Dan Sullivan "Professional Data Engineer Study Guide" | Textbook | Comprehensive coverage |
| Tutorials Dojo PDE Practice Exams | Practice questions | Exam realism |
| Apache Beam documentation | Official docs | Dataflow depth |
| BigQuery documentation | Official docs | SQL and storage depth |
| "Designing Data-Intensive Applications" by Kleppmann | Book | Distributed systems foundation |
Hands-On Practice Priorities
Exam questions are grounded in real-world scenarios. Candidates who have actually built pipelines answer scenario questions faster and more accurately. Prioritize these hands-on exercises:
- Build a Pub/Sub to Dataflow to BigQuery streaming pipeline for a real dataset (use the New York Taxi dataset from BigQuery public datasets)
- Create a BigQuery partitioned and clustered table and measure the query cost reduction
- Run a Dataproc job using an ephemeral cluster with Cloud Storage as the data source
- Train a BigQuery ML logistic regression model and evaluate it using ML.EVALUATE
- Set up a Cloud Composer DAG that orchestrates a multi-step pipeline with error handling and retries
- Configure a Bigtable table with a correctly designed row key for a time-series scenario
References
[1] Payscale. "Google Cloud Certified Professional Data Engineer Salary." payscale.com. Accessed May 2026.
[2] Google Cloud. "Professional Data Engineer Exam Guide." cloud.google.com/certification/data-engineer. Accessed May 2026.
[3] Google Cloud. "BigQuery Documentation." cloud.google.com/bigquery/docs. Accessed May 2026.
[4] Apache Software Foundation. "Apache Beam Documentation." beam.apache.org. Accessed May 2026.
[5] Tutorials Dojo. "Google Professional Data Engineer Practice Exams." tutorialsdojo.com. Accessed May 2026.
[6] Google Cloud. "Pub/Sub Documentation: Subscriber Overview." cloud.google.com/pubsub/docs/subscriber. Accessed May 2026.
[7] Kleppmann, Martin. "Designing Data-Intensive Applications." O'Reilly Media, 2017.
[8] Google Cloud. "Dataplex Documentation." cloud.google.com/dataplex/docs. Accessed May 2026.