Overview
Designing data processing systems is the second-largest section of the Professional Data Engineer exam at about 22%, split into four official objectives. Objective 1.1 covers access control, encryption, PII, data sovereignty and project structure. Objective 1.2 covers data cleaning, pipeline monitoring, disaster recovery, ACID guarantees and validation. Objective 1.3 covers mapping business needs to architecture, multi-cloud portability and cataloging. Objective 1.4 covers migration planning with BigQuery Data Transfer Service, Database Migration Service, Transfer Appliance and Datastream.
The exam is current as of October 2026, with no retirement announced. Google lists 40-50 questions in two hours and publishes no passing score. Its certification page warns that the exam will soon be updated for recent branding changes, so this guide gives today's product names next to the names printed in the guide, such as Knowledge Catalog for Dataplex.
How it's tested
Domain 1 questions describe an organization, its data, its rules and its goals, then ask for the design that satisfies all of them. Distractors are usually secure or scalable but miss one stated requirement, such as region, key ownership, downtime limit or the need to stay portable across clouds.
-
Constraint stacking
A scenario lists compliance, location and access rules together, and only one option honors every rule without adding needless operational effort.
-
Reliability trade-off
You pick consistency, availability or recovery settings that match a stated recovery objective or transactional need.
-
Migration path
Data size, network bandwidth, source type and allowed downtime point to one transfer or replication service.
A bank must move 400 TB of on-premises archives to Google Cloud within two weeks over a 100 Mbps link, and keep an operational MySQL database in sync during cutover. The link would take months for the archive, so Transfer Appliance handles the bulk copy. Database Migration Service or Datastream keeps the live database replicated with minimal downtime. Choosing a single online transfer for both ignores the bandwidth math.
Objectives 4
Official objectives from the Google Cloud outline, in outline order.
1.1 Designing for security and complianceIAM and organization policies, encryption and keys, PII, sovereignty, regulation, governance structure, dev and prod separation.
- What it means
- This objective asks you to protect data by design. You choose IAM roles and organization policy constraints, decide between Google-managed and customer-managed encryption keys, plan how PII is found and protected, keep data in permitted regions, and lay out projects, datasets and tables so governance and environment separation are simple to enforce.
- What to understand
- Grant predefined roles at the narrowest level that works, often the dataset or table in BigQuery, and use groups rather than individuals. Organization policies such as resource location restrictions stop resources being created in the wrong region. CMEK through Cloud KMS gives the customer control over key rotation and revocation; Cloud EKM keeps keys outside Google. Sensitive Data Protection (called Cloud DLP in the guide) discovers and de-identifies PII through masking, tokenization or format-preserving encryption. VPC Service Controls reduce exfiltration risk. Separate projects for development and production keep permissions and billing apart.
- How it shows up
- Expect a regulated customer needing key control, regional storage or PII de-identification; pick the least-privilege, policy-enforced design.
- Common confusion
- Assuming default Google encryption satisfies a requirement for customer-controlled keys; it does not give revocation control.
1.2 Designing for reliability and fidelityData preparation and cleaning, pipeline monitoring and orchestration, disaster recovery, ACID and availability choices, validation.
- What it means
- Reliability means pipelines keep running and recover from failure; fidelity means the data they produce is correct. The objective covers cleaning data with Dataform, Dataflow or Cloud Data Fusion, prompting LLMs to help generate queries, monitoring and orchestrating pipelines, planning disaster recovery, choosing transactional guarantees and validating data.
- What to understand
- Dataform manages SQL transformations in BigQuery with dependencies, assertions and version control. Dataflow cleans data at scale in batch or streaming. Cloud Data Fusion offers a visual, code-free pipeline builder. Disaster recovery is planned against RPO and RTO: multi-region datasets, cross-region replication and snapshots each cover different failure cases. Spanner gives strong consistency with global scale, Bigtable gives single-row atomicity only, and BigQuery supports multi-statement transactions. Validation checks such as Dataform assertions or Knowledge Catalog data quality scans catch bad records before they reach consumers.
- How it shows up
- Look for a stated RPO, RTO or correctness rule and choose the storage or tool that meets it with the least custom code.
- Common confusion
- Expecting Bigtable to provide multi-row transactions; it guarantees atomicity only within a single row.
1.3 Designing for flexibility and portabilityMapping business needs to architecture, multi-cloud and residency portability, data staging, cataloging, profiling and discovery.
- What it means
- This objective tests whether a design can change with the business. You translate current and future requirements into architecture, keep data and applications portable when multi-cloud or residency rules apply, and make data findable through staging zones, catalogs and profiling.
- What to understand
- Open formats and engines give portability: Apache Iceberg tables in Lakehouse for Apache Iceberg (BigLake in the guide), Apache Beam pipelines that run on Dataflow or elsewhere, and Spark on Managed Service for Apache Spark (Dataproc in the guide). BigQuery Omni queries data held in AWS or Azure without copying it. Knowledge Catalog (Dataplex Catalog in the guide) indexes metadata across BigQuery and Cloud Storage, while data profiling scans summarize column statistics. A raw, curated and consumption zone pattern keeps staging clear.
- How it shows up
- A company fears lock-in or must keep some data on another cloud; prefer open formats or BigQuery Omni over copying everything.
- Common confusion
- Treating portability as moving all data to one cloud; residency rules may require the data to stay where it is.
1.4 Designing data migrationsAssessing stakeholders and current systems, planning migration and validation with Google Cloud transfer and replication services.
- What it means
- Migration design starts with people and systems: who uses the data, which processes depend on it and what technology runs today. You then plan the route to the target state and prove it worked, using BigQuery Data Transfer Service, Database Migration Service, Transfer Appliance, Google Cloud networking and Datastream.
- What to understand
- BigQuery Data Transfer Service schedules loads from SaaS sources, Amazon S3 and other warehouses such as Teradata and Amazon Redshift. Database Migration Service moves MySQL, PostgreSQL, SQL Server and Oracle databases to Cloud SQL or AlloyDB with continuous replication. Datastream captures change data from operational databases into BigQuery or Cloud Storage. Transfer Appliance ships large datasets physically when bandwidth is short, and Storage Transfer Service handles online object copies. Validation compares row counts, checksums or aggregates between source and target before cutover.
- How it shows up
- Calculate whether the link can move the volume in time; that usually decides between online transfer and Transfer Appliance.
- Common confusion
- Confusing Datastream, which streams change data, with Database Migration Service, which migrates a database to a managed target.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
CMEK
Customer-managed encryption keys held in Cloud KMS.
- Remember
- Customer controls rotation and can disable the key.
- Tested as
- Fits rules that demand key ownership.
- Trap
- Thinking default encryption meets key-control rules.
-
Organization policy constraints
Org-wide guardrails such as allowed resource locations.
- Remember
- Prevents rather than detects violations.
- Tested as
- Residency and sovereignty scenarios.
- Trap
- Relying on IAM alone to restrict regions.
-
Sensitive Data Protection
Discovery, inspection and de-identification of sensitive data (Cloud DLP in the guide).
- Remember
- Masking, tokenization, format-preserving encryption.
- Tested as
- PII handling questions.
- Trap
- Deleting PII columns when analysts still need joins.
-
VPC Service Controls
Service perimeter around Google Cloud APIs.
- Remember
- Blocks data exfiltration across the perimeter.
- Tested as
- Paired with IAM for regulated data.
- Trap
- Treating it as a replacement for IAM.
-
RPO and RTO
Tolerable data loss and tolerable downtime.
- Remember
- They decide replication and backup design.
- Tested as
- DR scenarios state one or both.
- Trap
- Choosing backups when RPO is near zero.
-
Dataform assertions
Built-in data quality tests on SQL workflows.
- Remember
- Fail the run when a rule is broken.
- Tested as
- Fidelity and validation questions.
- Trap
- Validating only after data is published.
-
BigQuery Omni
BigQuery queries on data stored in AWS or Azure.
- Remember
- No copy into Google Cloud needed.
- Tested as
- Multi-cloud and residency cases.
- Trap
- Assuming all data must move first.
-
Transfer Appliance
Physical device for offline bulk transfer.
- Remember
- Used when bandwidth makes online copy too slow.
- Tested as
- Hundreds of TB with a deadline.
- Trap
- Ignoring bandwidth arithmetic.
-
Database Migration Service
Managed migration to Cloud SQL or AlloyDB with continuous replication.
- Remember
- Minimal-downtime cutover.
- Tested as
- Lift of operational databases.
- Trap
- Using it for analytics loads into BigQuery.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Encryption key options
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Google-managed keys | Default encryption at rest, no setup. | No customer control over keys. | Assuming it satisfies key-ownership rules. |
| CMEK (Cloud KMS) | Keys created and controlled by the customer in Google Cloud. | Customer rotates, disables and audits. | Forgetting service agents need key access. |
| Cloud EKM | Keys held in an external key manager. | Keys never stored in Google Cloud. | Choosing it when CMEK already meets the rule. |
Migration services
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| BigQuery Data Transfer Service | Scheduled loads into BigQuery from SaaS, S3 and warehouses. | Target is always BigQuery. | Using it for operational databases. |
| Database Migration Service | Database moves to Cloud SQL or AlloyDB. | Keeps source and target in sync until cutover. | Expecting it to load BigQuery. |
| Datastream | Serverless change data capture. | Ongoing replication into BigQuery or Cloud Storage. | Treating it as a one-time bulk copy tool. |
| Transfer Appliance | Offline shipping of large data. | Bypasses network limits. | Picking it for small or continuous loads. |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Default encryption is enough
- What candidates pick
- Rely on Google-managed encryption at rest
- Why it looks right
- All data is encrypted anyway.
- Why it's wrong
- The requirement is customer control and revocation, which only CMEK or Cloud EKM provide.
Clue to watch for Customer must control, rotate or revoke keys.
-
Delete the PII
- What candidates pick
- Drop sensitive columns before loading
- Why it looks right
- It removes the risk completely.
- Why it's wrong
- Analysts often still need to join or count on those fields; tokenization keeps utility while protecting values.
Clue to watch for Data must stay usable for analytics.
-
Online copy at any size
- What candidates pick
- Use Storage Transfer Service for hundreds of terabytes over a slow link
- Why it looks right
- It is managed and needs no hardware.
- Why it's wrong
- At the stated bandwidth the copy would miss the deadline; Transfer Appliance avoids the network.
Clue to watch for Large volume, limited bandwidth, fixed date.
-
One project for everything
- What candidates pick
- Keep dev and prod datasets in the same project
- Why it looks right
- Fewer projects to manage.
- Why it's wrong
- Shared projects blur IAM and billing boundaries and let test jobs touch production data.
Clue to watch for Separate teams or environments are mentioned.
-
Copy everything to Google Cloud
- What candidates pick
- Replicate all AWS data into BigQuery
- Why it looks right
- One platform seems simpler.
- Why it's wrong
- Residency or cost rules may forbid copying; BigQuery Omni queries the data in place.
Clue to watch for Data must remain in another cloud.
Answer framework
Use this sequence on long Domain 1 design scenarios before comparing the answer options.
- List every hard constraint in the scenario before reading the options: region, key control, PII handling, allowed downtime, recovery objectives and any multi-cloud requirement.
- Remove any option that breaks even one hard constraint, even if it looks faster, cheaper or more familiar to the team described in the question.
- For security needs, choose least-privilege IAM roles granted to groups at the narrowest scope, then add guardrails such as location constraints, CMEK or service perimeters.
- For migrations, compare data volume with available bandwidth and allowed downtime, then match the need to bulk offline transfer, scheduled loads or change data capture.
- Among remaining options, prefer managed services and open formats that reduce operations work and keep the design portable.
Practice question
Want more like this? The GCP PDE practice test has questions for every domain.
Checklist
0 of 12
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Start with security because it appears inside almost every design question. Build a small project with two BigQuery datasets, grant a group access to only one, create a CMEK key in Cloud KMS and attach it to a dataset, then disable the key and observe what fails. Run a Sensitive Data Protection inspection on sample data containing fake email addresses and try a tokenization transform.
Next, write a one-page table of recovery options: BigQuery time travel and snapshots, multi-region datasets, Cloud SQL replicas, Spanner multi-region configurations and Cloud Storage dual-region buckets, each with typical RPO and RTO. For migrations, practice the bandwidth calculation until it takes under a minute, then map each transfer service to its source type and target. Read the guide's subsection 1.3 slowly and study how Knowledge Catalog, Iceberg tables and BigQuery Omni support portability. Finish by answering long scenario questions and, for each wrong option, naming the exact constraint it violates. That habit is the core Domain 1 skill.
FAQ
How much of the GCP PDE exam is Designing data processing systems?
Designing data processing systems carries about 22% of the exam, second only to Ingesting and processing the data. On a 40-50 question exam that is roughly 9 to 11 questions.
Do I need to memorize IAM role names?
Know the common predefined BigQuery roles such as Data Viewer, Data Editor, Job User and Data Owner, and understand when to grant at dataset versus project level. Exhaustive role lists are not needed.
Why does this guide say Knowledge Catalog when the exam guide says Dataplex?
Google renamed Dataplex Universal Catalog to Knowledge Catalog in April 2026. The exam guide had not been updated yet, so expect either name and learn both.
Is CMEK always the right security answer?
No. Choose CMEK only when the scenario requires customer control over keys. Otherwise default encryption is sufficient and adds no key management work.
Which migration service should I study most?
Study all four named in the guide, but the most common decision is between online transfer and Transfer Appliance, and between Database Migration Service and Datastream.
Are case studies part of the current exam?
The current exam guide does not list case studies. Questions are standalone scenarios in multiple choice and multiple select format.
Other Data & Analytics certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 1 knowledge
You have covered security, reliability, portability and migration design. Check whether you can spot the constraint that eliminates each wrong option under exam timing.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the GCP PDE AI tutor Get any concept explained your way