- Domain 2 Overview: Why It's the Biggest Slice of the Exam
- Continuous Integration and Delivery for Application Workloads
- CI/CD for Infrastructure: Terraform, Config Sync, and Policy Enforcement
- Continuous Testing and Delivery for Machine Learning Workloads
- Continuous Testing Strategies You Need to Know
- Google Cloud Tooling You'll Be Tested On
- What Domain 2 Questions Actually Look Like
- A Focused Study Sequence for Domain 2
- Frequently Asked Questions
- Domain 2 is tied for the largest domain on the exam at roughly 25% of questions.
- It spans three distinct pipeline types: application code, infrastructure-as-code, and ML workloads.
- Continuous testing (unit, integration, canary, and automated rollback triggers) is tested across all three pipeline types.
- Expect scenario questions comparing Cloud Build, Cloud Deploy, Artifact Registry, and third-party tools like Jenkins or GitLab.
Domain 2 Overview: Why It's the Biggest Slice of the Exam
If you've already skimmed the CDE Exam Domains 2026: Complete Guide to All 5 Content Areas, you know that Google Cloud splits the Professional Cloud DevOps Engineer exam into five domains, with Domain 2 - Building and implementing CI/CD pipelines, including continuous testing, for application, infrastructure, and machine learning workloads - tied with Domain 4 as the largest at approximately 25% of the exam. That weighting alone tells you this domain deserves the most study hours on your calendar.
Unlike Domain 1's organizational bootstrapping focus, Domain 2 is deeply hands-on. It asks whether you can actually design, build, and troubleshoot delivery pipelines across three very different workload types: standard application services, infrastructure configuration, and machine learning models. Each of these has its own toolchain quirks, testing philosophy, and failure modes, and Google expects you to know all three cold.
Continuous Integration and Delivery for Application Workloads
This is the most familiar territory for most candidates, but the exam digs into specifics that generic DevOps knowledge won't cover. You need fluency with the full lifecycle: source trigger, build, test, artifact storage, deployment strategy, and rollback.
Application CI/CD Essentials
Candidates must understand how source changes flow through build and deployment automation on Google Cloud, and how to choose the right deployment strategy for the risk profile of a service.
- Cloud Build triggers from Cloud Source Repositories, GitHub, and GitLab, including branch and tag-based triggering
- Build configuration with cloudbuild.yaml, including substitutions, custom builders, and build step sequencing
- Artifact Registry for storing container images, language packages, and OS packages with vulnerability scanning
- Cloud Deploy delivery pipelines and progressive rollout strategies (canary, blue-green, rolling)
- Rollback mechanics and how to structure a pipeline so a bad release can be reverted quickly
A recurring exam pattern is presenting a scenario where a deployment strategy causes an incident, then asking which pipeline change would have caught the problem earlier. This mirrors the practical troubleshooting emphasis you'll also see reinforced in CDE Domain 4: Implementing observability practices and troubleshooting issues, since delivery and observability are tightly linked in real Google Cloud environments.
CI/CD for Infrastructure: Terraform, Config Sync, and Policy Enforcement
Infrastructure pipelines behave differently than application pipelines because the "deployment" is a change to declarative state rather than a running binary. The exam expects you to know how to safely automate this.
Infrastructure-as-Code Pipeline Essentials
You must be able to describe a safe, auditable path from a Terraform or config change to an applied state change in a live environment.
- Terraform plan/apply automation through Cloud Build, including remote state management and locking
- Policy-as-code validation gates (e.g., policy checks before apply) to catch non-compliant resources
- Config Sync and Anthos Config Management for GitOps-style Kubernetes configuration
- Environment promotion patterns: dev to staging to production with approval gates between stages
- Drift detection and how automated pipelines should react when live state diverges from source-controlled state
Key Takeaway
Memorize the difference between a "plan" step (read-only, safe to automate broadly) and an "apply" step (state-changing, usually gated by manual approval or a stricter automated check) - the exam frequently tests this distinction.
Continuous Testing and Delivery for Machine Learning Workloads
This is the section that most surprises candidates coming from a traditional ops background. ML pipelines introduce artifacts beyond code: datasets, trained models, and evaluation metrics, each of which needs its own versioning and validation approach.
ML Pipeline Essentials
Understand how continuous training and continuous delivery extend standard CI/CD patterns to accommodate model artifacts and data quality checks.
- Vertex AI Pipelines for orchestrating training, evaluation, and deployment steps
- Model registry concepts: versioning trained models and promoting them between environments
- Automated evaluation gates that compare a new model's metrics against a baseline before promotion
- Data validation steps to catch schema or distribution drift before training runs
- Canary and shadow deployment patterns for serving new model versions safely
You don't need to be a data scientist to pass this domain, but you do need to recognize the pipeline shape: data validation, training, evaluation gate, model registration, and staged rollout. Expect at least a few questions that ask you to identify the missing stage in a described ML pipeline.
Continuous Testing Strategies You Need to Know
Continuous testing is called out explicitly in the domain title, so it isn't a side topic - it's tested across all three workload types above. The exam wants you to match the right testing layer to the right stage of a pipeline.
| Testing Layer | Typical Pipeline Stage | What It Catches |
|---|---|---|
| Unit tests | Build/commit stage | Logic errors in individual functions or modules |
| Integration tests | Post-build, pre-deploy | Broken interactions between services or components |
| Policy/compliance checks | Pre-apply (infrastructure) | Non-compliant or insecure resource configurations |
| Model evaluation gates | Post-training (ML) | Model quality regressions vs. baseline metrics |
| Canary analysis | Post-deploy, production | Regressions only visible under real traffic |
A well-designed pipeline layers these tests so that cheap, fast checks run first and expensive, slow checks (like full canary analysis) run only after earlier gates pass. Expect exam scenarios where a team is testing too late (catching issues only in production) or too little (skipping integration tests), and you're asked to identify the fix.
Google Cloud Tooling You'll Be Tested On
The exam is tool-aware but not tool-obsessed - you need working knowledge of the primary Google Cloud services plus an understanding of how third-party tools fit into a Google Cloud-centric pipeline.
- Cloud Build: the core CI engine; know triggers, build configs, and private pool usage for VPC-restricted builds
- Cloud Deploy: managed continuous delivery for GKE and other runtimes, including delivery pipeline YAML and promotion sequences
- Artifact Registry: centralized artifact storage with vulnerability scanning and regional replication
- Binary Authorization: attestation-based deployment gating to ensure only verified images reach production
- Cloud Source Repositories, GitHub, and GitLab integrations: source triggers and webhook-based automation
- Vertex AI Pipelines: orchestration for ML training and deployment workflows
What Domain 2 Questions Actually Look Like
Domain 2 questions on the real exam tend to follow a few recognizable patterns, consistent with the multiple-choice and multiple-select format used throughout all 50-60 questions on the exam:
- Diagnose-the-gap scenarios: a pipeline description is given, and you must identify what testing or gating step is missing before production deployment.
- Tool-selection scenarios: given constraints (e.g., need for VPC isolation, need for GitOps, need for canary rollout), select the appropriate Google Cloud service or configuration.
- Rollback and recovery scenarios: a deployment has caused a regression, and you must choose the fastest safe recovery path.
- Cross-workload comparisons: a question tests whether you know that an ML model promotion gate and an infrastructure policy gate serve analogous but distinct purposes.
If you're still building a feel for the exam's overall difficulty and phrasing, How Hard Is the CDE Exam? Complete Difficulty Guide 2026 breaks down what makes these scenario-based questions harder than straightforward recall, and CDE Pass Rate 2026: What the Data Shows discusses what the available data suggests about candidate outcomes.
A Focused Study Sequence for Domain 2
Given that Domain 2 is worth about 25% of the exam, it deserves a dedicated multi-week block within your broader prep schedule rather than being folded into general review. Here's a sequence that respects the domain's three distinct sub-areas.
Application CI/CD Fundamentals
- Build and trigger a sample pipeline with Cloud Build and Artifact Registry
- Configure a Cloud Deploy pipeline with a canary rollout stage
- Practice identifying rollback triggers in sample scenarios
Infrastructure Pipelines
- Automate Terraform plan/apply through Cloud Build with a manual approval gate
- Study Config Sync and policy-as-code enforcement patterns
- Review drift detection and environment promotion strategies
ML Pipelines and Continuous Testing
- Walk through a Vertex AI Pipelines example end to end
- Map testing layers (unit, integration, evaluation gate, canary) to pipeline stages
- Take a timed practice set focused only on Domain 2 questions
This sequence pairs well with the broader roadmap in the CDE Study Guide 2026: How to Pass on Your First Attempt, which lays out how to balance all five domains across your total prep window. Since Domain 2 and Domain 4 are tied as the exam's heaviest sections, many candidates study them back-to-back - you can preview that pairing in CDE Domain 4: Implementing observability practices and troubleshooting issues.
Who Hires for These Skills
Domain 2 skills map directly to job titles like DevOps Engineer, Platform Engineer, Release Engineer, and MLOps Engineer. Employers hiring for these roles specifically want candidates who can build pipelines that span application, infrastructure, and increasingly ML deployment - which is exactly why Google structured this domain the way it did. If you're evaluating whether the certification translates into real job opportunities, CDE Jobs outlines common titles and responsibilities, while CDE Salary Guide 2026: Complete Earnings Analysis and Is the CDE Certification Worth It? Complete ROI Analysis 2026 dig into the broader value proposition of holding this credential.
For a refresher on what the certification represents at a high level before you go deeper into domain content, see What Is CDE Certification? or the shorter primer at What Is CDE?.
Practicing Domain 2 Effectively Before Exam Day
Reading documentation is necessary but not sufficient for this domain - the exam rewards candidates who have actually clicked through Cloud Build triggers, written a delivery pipeline YAML, and watched a canary rollout in action. If hands-on lab time is limited, lean heavily on scenario-based practice questions that mirror the diagnose-the-gap and tool-selection styles described earlier. Working through timed sets on our Professional Cloud DevOps Engineer practice test platform is one of the fastest ways to expose weak spots in your Domain 2 knowledge before they cost you points on exam day.
It's also worth revisiting your weaker sub-area a second time after a short break - most candidates find ML pipeline testing gates the least intuitive on first pass, simply because it's the newest addition to traditional DevOps thinking. Running a second round of targeted questions from the full practice test library a few days before your exam date helps confirm the concepts have actually stuck, not just been recognized once.
Frequently Asked Questions
Domain 2 is approximately 25% of the exam. With 50-60 total questions, that works out to roughly 12-15 questions covering application, infrastructure, and ML pipeline topics.
You need working conceptual knowledge of how ML continuous training and delivery pipelines are structured, including data validation, evaluation gates, and model promotion. Deep data science expertise is not required.
Both. The exam expects familiarity with Cloud Deploy's delivery pipeline structure specifically, but it also tests general CI/CD reasoning that applies regardless of the underlying tool.
Domain 2 focuses on building the pipelines that deliver changes; Domain 3, covered in CDE Domain 3: Applying site reliability engineering practices, focuses on operating the resulting systems reliably. Rollback and canary concepts appear in both but from different angles.
Either order works, but many candidates study Domain 1's organizational foundation first since it's covered in CDE Domain 1: Bootstrapping and maintaining a Google Cloud organization, then move into Domain 2's pipeline mechanics once the environment context is clear.
- CDE Domain 1: Bootstrapping and maintaining a Google Cloud organization (20%) - Complete Study Guide 2026
- CDE Domain 3: Applying site reliability engineering practices (18%) - Complete Study Guide 2026
- CDE Domain 4: Implementing observability practices and troubleshooting issues (25%) - Complete Study Guide 2026
- CDE Exam Domains 2026: Complete Guide to All 5 Content Areas