GCP Professional Data Engineer (PDE) Study Guide 2026

The Google Cloud Professional Data Engineer (PDE) exam is 2 hours, 40-50 multiple choice and multiple select questions, and USD 200 (plus tax). Google does not publish a passing score. The certification is valid for 2 years, and a 1-hour, 20-question renewal exam (USD 100) is available.

Practice this exam in timed mode with the StudyKits Professional Data Engineer app. The Google Cloud certification hub lists every Google Cloud exam guide and practice app.

Its five sections are designing data processing systems (~22%), ingesting and processing data (~25%), storing data (~20%), preparing and using data for analysis (~15%), and maintaining and automating data workloads (~18%). BigQuery runs through all of them. Facts are from the Google pages under Sources, checked October 15, 2026.

What is the PDE exam format?

The Professional Data Engineer exam has 40-50 questions and you get 2 hours. Google does not publish a passing score. The exam costs USD 200 (plus tax) and the certification is valid for 2 years (a shorter renewal exam is available). Google’s exam page notes the exam “will soon be updated to reflect recent branding changes”, so check the current guide before you book.

Google recommends 3+ years of industry experience including 1+ years designing and managing solutions on Google Cloud. If you have already passed the Professional Cloud Architect exam, you have a strong foundation for the PDE.

The exam covers five sections: designing data processing systems (~22%), ingesting and processing the data (~25%), storing the data (~20%), preparing and using data for analysis (~15%), and maintaining and automating data workloads (~18%).

Which Google Cloud data services must you know?

BigQuery: The Heart of GCP Data Engineering

BigQuery is central to the PDE exam. It is a serverless, multi-cloud data warehouse that separates compute from storage, allowing each to scale independently. Nearly every data architecture on GCP uses BigQuery as the analytics layer.

Architecture concepts you must understand:

• Dremel execution engine — how BigQuery processes queries across thousands of workers
• Colossus storage — how data is stored in a columnar format called Capacitor
• Separation of storage and compute — why this architecture enables massive scalability
• Slots — units of computational capacity. Know the difference between on-demand pricing and capacity-based pricing (BigQuery editions with reserved slots)

Optimization techniques the exam tests:

• Partitioning — time-based partitioning (by ingestion time or a timestamp column), range partitioning, and integer partitioning. Partitioning reduces the amount of data scanned per query.
• Clustering — organizing data within partitions by up to four columns. Clustering further reduces data scanned and improves query performance.
• Materialized views — precomputed results that BigQuery automatically refreshes. Know when to use materialized views vs standard views vs tables.
• BI Engine — in-memory analysis service for sub-second query performance on dashboards.
• Query optimization — avoiding SELECT *, using approximate aggregation functions (APPROX_COUNT_DISTINCT), and structuring queries to minimize data processed.

Security and governance:

• Column-level security with policy tags
• Row-level security with row access policies
• Data masking for sensitive fields
• Authorized views and authorized datasets for controlled data sharing
• BigQuery Data Transfer Service for automated data movement

Dataflow: Unified Batch and Stream Processing

Dataflow is Google Cloud’s fully managed service for running Apache Beam pipelines. It handles both batch and streaming data processing with the same programming model.

Key concepts:

• Apache Beam programming model — PCollections, Transforms (ParDo, GroupByKey, CoGroupByKey, Combine, Flatten), and Pipelines
• Windowing — fixed windows, sliding windows, session windows, and global windows for grouping streaming data
• Watermarks — tracking event time progress and handling late-arriving data
• Triggers — controlling when results are emitted (event time triggers, processing time triggers, data-driven triggers)
• Exactly-once processing — how Dataflow guarantees no duplicate processing
• Autoscaling — horizontal autoscaling based on pipeline backlog
• Flex Templates — parameterized, containerized pipeline templates

When to use Dataflow:

• Real-time streaming analytics from Pub/Sub
• Batch ETL from Cloud Storage to BigQuery
• Data enrichment and transformation pipelines
• When you need unified batch and streaming with the same code

Dataproc: Managed Spark and Hadoop

Dataproc is a managed service for running Apache Spark, Hadoop, Flink, and Presto clusters. It is the right choice when you have existing Spark or Hadoop workloads that you want to migrate to Google Cloud.

Key concepts:

• Cluster types — standard (1 master), high availability (3 masters), and single node
• Autoscaling policies for worker nodes
• Ephemeral clusters — creating clusters for specific jobs and deleting them when done (the recommended pattern)
• Dataproc on GKE — running Spark workloads on GKE for better resource utilization
• Initialization actions — scripts that customize cluster setup
• Optional components — Jupyter, Zeppelin, Hive, Presto
• Integration with Cloud Storage as the persistent data layer (instead of HDFS)

Dataflow vs Dataproc decision tree:

• New pipeline with no existing code? Choose Dataflow.
• Existing Spark or Hadoop code? Choose Dataproc.
• Need unified batch and streaming? Choose Dataflow.
• Need fine-grained cluster control or specific Hadoop ecosystem tools? Choose Dataproc.
• Want fully serverless? Choose Dataflow (or Dataproc Serverless for Spark).

Pub/Sub: Real-Time Messaging

Cloud Pub/Sub is a fully managed messaging service for event-driven architectures and real-time data ingestion.

Key concepts:

• Topics and subscriptions — publishers send messages to topics, subscribers receive messages from subscriptions
• Push vs pull subscriptions — push delivers to HTTP endpoints, pull requires subscribers to request messages
• At-least-once delivery — messages may be delivered more than once. Design for idempotency.
• Ordering — Pub/Sub does not guarantee order by default. Use ordering keys when order matters.
• Dead-letter topics — handling messages that cannot be processed
• Message retention — configurable up to 31 days
• Pub/Sub Lite — lower-cost option for high-volume, latency-tolerant workloads
• Exactly-once delivery — available with pull subscriptions and specific client library configurations

Cloud Composer: Managed Airflow

Cloud Composer is Google’s managed Apache Airflow service for orchestrating data pipelines.

Key concepts:

• DAGs (Directed Acyclic Graphs) — defining workflow dependencies
• Operators — GCP-specific operators for BigQuery, Dataflow, Dataproc, and Cloud Storage
• Sensors — waiting for external conditions (file arrival, API response)
• Environment sizing — small, medium, large environments and when to scale
• Cloud Composer 2 vs Cloud Composer 3 — architecture differences and improvements
• When to use Composer vs Cloud Workflows vs Cloud Scheduler

When to choose Cloud Composer:

• Complex pipeline orchestration with dependencies
• Existing Airflow DAGs you want to migrate
• Need for retries, SLAs, and complex scheduling
• Multi-step workflows involving multiple GCP services

Additional Data Services

Cloud Storage:

• Storage classes: Standard, Nearline, Coldline, Archive
• Object lifecycle management
• Transfer Service for bulk data migration
• Signed URLs for temporary access

Firestore:

• Document database for real-time applications
• Real-time listeners for data synchronization
• Offline support for mobile applications

Bigtable:

• Wide-column NoSQL for low-latency, high-throughput workloads (IoT, time-series, ad tech)
• Row key design for performance
• Replication for high availability

Cloud Spanner:

• Globally distributed relational database with strong consistency
• Horizontal scaling with relational semantics
• When to use Spanner vs Cloud SQL vs AlloyDB

Memorystore:

• Managed Redis and Memcached for caching
• Session management and real-time analytics

How does PDE compare with AWS DEA-C01?

If you have studied for or passed the AWS Data Engineer Associate (DEA-C01), understanding the service mapping helps:

GCP Service	AWS Equivalent	Key Difference
BigQuery	Redshift + Athena	BigQuery is serverless by default
Dataflow	Glue (Spark) + Kinesis Analytics	Dataflow uses Apache Beam, unified batch/stream
Dataproc	EMR	Both manage Spark/Hadoop clusters
Pub/Sub	Kinesis Data Streams + SNS	Pub/Sub is serverless with no shard management
Cloud Composer	MWAA	Both are managed Airflow
Cloud Storage	S3	Very similar feature sets
Bigtable	DynamoDB	Bigtable is wide-column, DynamoDB is key-value/document
Cloud Spanner	Aurora (global)	Spanner has true global strong consistency
Dataflow templates	Glue job bookmarks	Different approaches to reusable pipelines
Data Catalog	Glue Data Catalog	Similar metadata management

The biggest philosophical difference: GCP leans heavily into BigQuery as the central analytics service, while AWS distributes analytics across Redshift, Athena, and Glue. If you understand AWS data services, you already understand the concepts — you just need to learn the GCP-specific implementations.

How much machine learning is on the PDE exam?

The PDE exam includes questions about preparing data for machine learning and using ML services within data pipelines.

Key ML services to know:

• BigQuery ML — training and deploying ML models directly in BigQuery using SQL
• Vertex AI — managed ML platform for custom model training, AutoML, and model serving
• TensorFlow on Dataflow — running TensorFlow models within Dataflow pipelines for real-time inference
• Feature Store — centralized repository for ML features

You do not need to be a machine learning expert, but you need to understand how data engineers prepare data for ML workflows and how ML models integrate into data pipelines.

Which security and governance topics are tested?

• IAM — roles for BigQuery, Dataflow, Dataproc, Pub/Sub. Understand predefined roles vs custom roles.
• Data Catalog — metadata management, data discovery, policy tags for column-level security
• Cloud DLP (Data Loss Prevention) — identifying and protecting sensitive data (PII, credit card numbers) in datasets
• VPC Service Controls — preventing data exfiltration from BigQuery and other data services
• CMEK (Customer-Managed Encryption Keys) — encrypting data with your own KMS keys in BigQuery, Cloud Storage, Dataflow, and Dataproc
• Audit logging — Cloud Audit Logs for tracking data access

What does an 8-week PDE study plan look like?

Weeks 1-2: BigQuery Mastery

BigQuery is the foundation. Spend two full weeks on it.

• Architecture, pricing models, optimization techniques
• Partitioning, clustering, materialized views
• Security: column-level, row-level, data masking
• BigQuery ML basics
• Hands-on: load, transform, and query datasets in BigQuery
• 20 practice questions per day

Weeks 3-4: Data Processing with Dataflow and Dataproc

• Apache Beam programming model and Dataflow execution
• Windowing, watermarks, and triggers for streaming
• Dataproc cluster management and autoscaling
• When to use Dataflow vs Dataproc
• Hands-on: build a streaming pipeline from Pub/Sub to BigQuery with Dataflow
• 25 practice questions per day

Weeks 5-6: Ingestion, Storage, and Orchestration

• Pub/Sub architecture and patterns
• Cloud Composer DAGs and operators
• Cloud Storage, Bigtable, Spanner, Firestore selection criteria
• Data migration strategies with Transfer Service and DMS
• Hands-on: orchestrate a multi-step pipeline with Cloud Composer
• 30 practice questions per day

Weeks 7-8: Security, ML Integration, and Practice Exams

• IAM, VPC Service Controls, Cloud DLP, CMEK
• Data Catalog and governance
• ML integration: BigQuery ML, Vertex AI, Feature Store
• Take full-length practice exams
• Review weak areas
• 40 practice questions per day
• Schedule your exam

What exam strategy works for PDE?

• BigQuery questions will make up a large portion of the exam. If you know BigQuery deeply, you have a strong foundation.
• For “which service should you use” questions, focus on the specific requirements. Real-time vs batch, structured vs unstructured, and latency requirements are the key differentiators.
• Google favors serverless solutions. When in doubt, choose BigQuery, Dataflow, or Pub/Sub over self-managed alternatives.
• Pay attention to cost optimization questions. Ephemeral Dataproc clusters, BigQuery partitioning, and Cloud Storage lifecycle policies are common cost-saving answers.
• For ML questions, remember that data engineers prepare and deliver data — they do not need to build complex ML models.

How can you practise for PDE?

Use this guide as a map, check each topic against the current exam guide, and build hands-on experience with BigQuery, Dataflow and the rest of the data stack.

The StudyKits Data Engineer app has original practice questions with explanations, organised by the exam guide’s sections.

Frequently asked questions

How many questions are on the Professional Data Engineer exam?

40-50, in 2 hours. That is fewer than the 50-60 on most other Google professional exams.

What is the PDE passing score?

Google does not publish one.

Is the PDE exam changing?

Google’s exam page says it “will soon be updated to reflect recent branding changes”, with no date. Check the guide before you book.

How do I renew the Data Engineer certification?

It is valid for 2 years. Renew with the full exam, the 1-hour, 20-question renewal exam (USD 100), or designated Google Skills courses, which add 1 year.

Keep reading

• Professional Data Engineer vs ML Engineer (Google Cloud)

Sources

Checked October 15, 2026:

• Google Cloud Professional Data Engineer
• PDE exam guide (PDF)
• Google Cloud certification FAQ (validity)
• Google Cloud certification renewal