Google Cloud Professional Data Engineer (PDE) is Google Cloud's certification for people who design, build and run data platforms: ingestion with Pub/Sub and Dataflow, storage in BigQuery and Bigtable, orchestration with Cloud Composer, and governance with Dataplex. The current exam guide also covers preparing data for AI, including embeddings and retrieval-augmented generation (RAG), BigQuery ML and prompting LLMs to generate queries.
This guide follows Google's official exam guide: all five sections with weights, the service choices that decide most questions, a 7-week plan and 15 original scenario questions.
Quick facts (Google Cloud, November 2026)
| Item | Detail |
|---|---|
| Length | 2 hours |
| Questions | 40–50, multiple choice and multiple select |
| Fee | USD 200 (plus tax where applicable) |
| Languages | English, Japanese |
| Delivery | Online-proctored or onsite-proctored |
| Prerequisites | None; 3+ years of industry experience including 1+ year designing and managing data solutions on Google Cloud recommended |
| Valid for | 2 years |
| Renewal | Standard exam, or a renewal exam (1 hour, 20 questions, USD 100) from 60 days before expiry |
Read the official Professional Data Engineer page and its exam guide before you book.
Who should take PDE?
- Data engineers working with BigQuery, Dataflow, Spark or Kafka.
- ETL, SQL and BI developers moving to cloud data platforms.
- ML and AI engineers who build the data pipelines behind models and RAG apps.
Freshers should first get comfortable with SQL, Python and one cloud, perhaps through the Associate Cloud Engineer. See the data engineer career path for the full roadmap.
Sections and weights
| Section | Weight |
|---|---|
| 1. Designing data processing systems | ~22% |
| 2. Ingesting and processing the data | ~25% |
| 3. Storing the data | ~20% |
| 4. Preparing and using data for analysis | ~15% |
| 5. Maintaining and automating data workloads | ~18% |
Section 1: Designing data processing systems (~22%)
- Security and compliance: IAM and organisation policies; encryption and key management; PII handling; data sovereignty; project, dataset and table design for governance; dev vs prod environments.
- Reliability and fidelity: cleaning data with Dataform, Dataflow, Cloud Data Fusion and LLM-generated queries; pipeline monitoring and orchestration; disaster recovery; ACID vs availability trade-offs; data validation.
- Flexibility and portability: multicloud and residency requirements; staging, cataloguing, profiling and discovery.
- Migrations: BigQuery Data Transfer Service, Database Migration Service, Transfer Appliance, Datastream and networking.
Section 2: Ingesting and processing (~25%)
- Planning: sources and sinks, transformation and orchestration logic, networking and encryption.
- Building: Dataflow (Apache Beam), Dataproc (Spark and Hadoop), Cloud Data Fusion, BigQuery, Pub/Sub and Kafka; batch and streaming transformations, including windowing and late-arriving data; AI data enrichment.
- Operating: Cloud Composer and Workflows for orchestration; CI/CD for pipelines.
| Situation | Usual choice |
|---|---|
| Serverless batch and streaming with one code base | Dataflow (Apache Beam) |
| Existing Spark or Hadoop jobs to move with little change | Dataproc |
| Event ingestion, decoupling producers and consumers | Pub/Sub |
| Change data capture from MySQL, PostgreSQL or Oracle into BigQuery | Datastream |
| Visual, low-code pipelines | Cloud Data Fusion |
| SQL transformations inside BigQuery with version control | Dataform |
| Scheduling many dependent jobs as DAGs | Cloud Composer (Airflow) |
Section 3: Storing the data (~20%)
- Choosing storage: BigQuery, BigLake, AlloyDB, Bigtable, Spanner, Cloud SQL, Cloud Storage, Firestore and Memorystore, by access pattern, cost and performance; data lifecycle.
- Data warehouse: data modelling and how far to normalise; mapping business requirements to access patterns.
- Data lake: discovery, access and cost controls; processing and monitoring the lake.
- Data platform: Dataplex, Dataplex Catalog, BigQuery and Cloud Storage; federated governance across domains.
-- BigQuery: partition by day and cluster by customer to cut scanned bytes
CREATE TABLE sales.orders
PARTITION BY DATE(order_ts)
CLUSTER BY customer_id AS
SELECT * FROM staging.orders_raw;Section 4: Preparing and using data for analysis (~15%)
- Visualisation: connecting BI tools; precalculated fields; BI Engine and materialized views; fixing slow queries; masking, IAM and Cloud DLP (Sensitive Data Protection).
- AI and ML: feature engineering, training and serving with BigQuery ML; preparing unstructured data for embeddings and RAG.
- Sharing: sharing rules, publishing datasets and reports, and BigQuery sharing (Analytics Hub).
For the RAG side, see the pgvector RAG tutorial and embeddings explained.
Section 5: Maintaining and automating workloads (~18%)
- Resources: cost per business need; capacity for critical jobs; persistent vs job-based Dataproc clusters.
- Automation: Composer DAGs; repeatable scheduling.
- Workload organisation: BigQuery editions and reservations; interactive vs batch queries.
- Monitoring: Cloud Monitoring and Logging, the BigQuery admin panel, errors, billing and quotas.
- Failures: fault tolerance and restarts; multi-region or multi-zone jobs; corrupt or missing data; replication and failover for Cloud SQL and Redis.
7-week study plan
| Week | Focus | Lab |
|---|---|---|
| 1 | BigQuery: storage, partitioning, clustering, cost, IAM | Load a public dataset; compare bytes scanned with and without partitions |
| 2 | Pub/Sub and Dataflow streaming: windows, triggers, late data | Stream events into BigQuery with 1-minute windows |
| 3 | Dataproc, Data Fusion, Dataform, Datastream | CDC from Cloud SQL to BigQuery with Datastream |
| 4 | Storage choices: Bigtable, Spanner, Cloud SQL, AlloyDB, Cloud Storage classes | Decision table for 10 workloads |
| 5 | Governance: Dataplex, Catalog, DLP, policy tags, column-level security | Mask a PII column for analysts |
| 6 | Composer DAGs, reservations, monitoring; BigQuery ML; embeddings | DAG that loads, validates and trains a BigQuery ML model |
| 7 | Sample questions and revision | Review every wrong answer; book the exam |
15 scenario questions with answers
These are original scenarios written for this guide, not exam questions.
1. Clickstream events must be processed in near real time and loaded into BigQuery with minimal operations. What do you build?
Pub/Sub into a Dataflow streaming pipeline writing to BigQuery.
2. Mobile events arrive up to 2 hours late. How should a Dataflow pipeline handle them?
Event-time windowing with watermarks, allowed lateness and triggers that update results.
3. Hundreds of existing Spark jobs must move to Google Cloud quickly. What do you use?
Dataproc, ideally with job-based (ephemeral) clusters.
4. Changes in an on-premises MySQL database must replicate continuously into BigQuery. What service?
Datastream (change data capture).
5. Queries on a 5 TB table always filter by date and customer. How do you cut cost?
Partition by date and cluster by customer ID.
6. An IoT app writes millions of time-series rows per second and reads by device and time. Which database?
Bigtable, with a row key designed around device and time.
7. A global app needs relational transactions with strong consistency across regions. Which database?
Spanner.
8. Analysts must see only masked phone numbers while the fraud team sees full values. How?
BigQuery policy tags with column-level security and dynamic data masking.
9. A dashboard runs the same aggregate query every minute and is slow. What helps?
A materialized view and/or BI Engine.
10. Twenty dependent daily jobs need retries, alerts and a visible dependency graph. What do you use?
Cloud Composer (Airflow) DAGs.
11. Analysts want to build a churn model without leaving SQL. What do you use?
BigQuery ML.
12. Product manuals must be searchable by meaning for a support chatbot. What preparation is needed?
Chunk the documents, generate embeddings, and store them with metadata in a vector-capable store for RAG retrieval.
13. Your company wants to share a curated dataset with partner companies without copying it. What do you use?
BigQuery sharing (Analytics Hub) listings.
14. Ad-hoc analyst queries slow down critical nightly ELT. How do you protect the ELT?
Separate BigQuery reservations (or editions) for ELT and ad-hoc workloads.
15. 300 TB must move from an on-premises data centre with a slow link. What do you use?
Transfer Appliance.
After PDE
- Prepare with data engineering interview questions and Snowflake vs Databricks for data engineers.
- Read AI for data engineers and GCP for AI engineers.
- See live data engineering jobs.
Cloud Soft Solutions in Ameerpet, Hyderabad trains data engineering with SQL, Python, Spark and cloud data platforms. See the data engineering and AI/ML course and all courses.
📞 Call +91 96660 19191 · WhatsApp · Flat No. 513, 5th Floor, Nilagiri Block, Aditya Enclave, Ameerpet, Hyderabad
Frequently Asked Questions
How long is the Google Professional Data Engineer exam?
2 hours, with 40 to 50 multiple-choice and multiple-select questions.
How much does the Professional Data Engineer exam cost?
USD 200, plus tax where applicable.
Which PDE section has the most weight?
Ingesting and processing the data, at about 25%.
Is there a prerequisite for the Data Engineer exam?
No. Google recommends 3+ years of industry experience, including 1+ year designing and managing data solutions on Google Cloud.
How long is the certification valid?
2 years. You can renew with the standard exam or a 1-hour, 20-question renewal exam from 60 days before expiry.
Does the PDE exam cover AI?
Yes. It includes preparing data for ML with BigQuery ML and preparing unstructured data for embeddings and RAG.
