Lesson 1: The Exam, Its Guide and Best-Answer Questions

Why start with the exam itself?

On most projects you would read the requirements before you open an editor. This exam works the same way. Google publishes a free exam guide that lists every topic it can test, and its questions are written as short scenarios that ask for the best answer, not just a correct one. Learn how both work and the next 47 levels will go more smoothly.

First, what this course is and isn't:

• The exam guide belongs to Google. You can download it free from Google's Professional Data Engineer certification page. The same page links to official sample questions. This course does not reproduce the guide. It teaches to it, so download your own copy and keep it open.
• This course is independent of Google. Google does not endorse it, and no course can promise you a pass. What it can do is cover every item in the guide and train the habits that win marks.

The exam facts

Google recommends at least 3 years of industry experience, including at least 1 year designing and managing solutions on Google Cloud. There are no formal prerequisites.

Gotcha: 2 hours for up to 50 questions gives you about 2.4 minutes each. Scenario questions are long, so spend about 2 minutes on a first pass, flag any question you're unsure of, and come back to it.

How the guide maps onto this course

The guide's Professional Data Engineer description sums up the job: someone who empowers data-driven decisions by collecting, transforming, storing, and delivering data. The guide splits that job into five weighted sections, and this course's phases follow the same order:

# Section weights from Google's exam guide, used to plan study time
sections = {
    "1 Designing data processing systems":     22,
    "2 Ingesting and processing the data":     25,  # heaviest section
    "3 Storing the data":                      20,
    "4 Preparing and using data for analysis": 15,
    "5 Maintaining and automating workloads":  18,
}

study_hours = 60
for name, pct in sections.items():
    print(f"{name:42} {pct:>3}%  -> {study_hours * pct / 100:4.1f} h")

# 1 Designing data processing systems         22%  -> 13.2 h
# 2 Ingesting and processing the data         25%  -> 15.0 h
# ...

The course gives each section time in proportion to its weight, and you should do the same. Section 2 (Pub/Sub, Dataflow and Apache Beam, orchestration with Cloud Composer) earns more of your hours than Section 4.

Reading a best-answer scenario

Often every option would work in practice. Only one fits all the constraints in the question. Treat the question like a requirements spec: find the constraint words first, then rule out any option that breaks one of them.

Common constraint words and what they usually point to:

• lowest cost: fewer always-on resources, cheaper storage classes, job-based rather than persistent clusters
• least operational overhead / fully managed: serverless, managed services instead of self-managed clusters on VMs
• real-time / within seconds: streaming (Pub/Sub, Dataflow), not nightly batch
• global consistency / strongly consistent across regions: Spanner, not a single-region database

Worked example

A retailer wants clickstream events from its website on a dashboard within seconds. The two-person data team wants the least operational overhead. What should they build?

A. Self-managed Apache Kafka on Compute Engine, then Spark Streaming on a persistent cluster B. Pub/Sub, then a Dataflow streaming pipeline, then BigQuery C. Hourly CSV exports to Cloud Storage, loaded into BigQuery on a schedule D. A persistent Managed Service for Apache Spark cluster polling the web servers' logs

Working through it:

Constraints:  [real-time: seconds]  [least operational overhead]
A  streaming yes, but you run Kafka VMs yourself      -> fails overhead
B  streaming, serverless end to end                   -> fits both  ✔
C  managed, but hourly                                -> fails real-time
D  always-on cluster you size, patch and pay for      -> fails overhead

A, C and D are realistic designs, and that's what makes them good distractors. Each one breaks exactly one constraint.

Best practice: Read the last sentence first. It tells you what is being asked ("Which storage system…", "What should you do first?"). For multiple-select questions the stem says how many answers to pick. Pick exactly that many.

Products are being renamed

Google has renamed several data products recently. The certification page itself warns that the exam will be updated to match the new branding, and tells you to check the exam guide for the names used on the exam. So you may see either name. This course uses the current name and notes the former one:

In most of these renames the brand changed and the tooling did not. For the Spark service, Google states that the API, client libraries, gcloud CLI and IAM role names are unchanged. Your scripts keep working:

# Still the command after the rename to Managed Service for Apache Spark
gcloud dataproc clusters create etl-cluster \
    --region=europe-west2 \
    --num-workers=2
# IAM roles keep their names too, e.g. roles/dataproc.editor

The DLP API is also still called the Cloud Data Loss Prevention API. The guide's own wording shows the overlap: it lists "Cloud Data Loss Prevention (Cloud DLP)" and "BigQuery sharing (Analytics Hub)". So treat both names as the same product.

What the exam asks about this

The exam won't ask you about its own fee. It tests the habits from this level on every question. Expect long scenarios built around constraint words, with distractors that are sound designs but break one requirement, and product names that may be old or new. Map each question to a guide section so you know which part of your knowledge it draws on.

Key takeaways

• The exam guide and sample questions are Google's and free on its certification page. This course is independent, not endorsed by Google, and cannot promise a pass.
• Standard exam: 2 hours, 40–50 multiple-choice and multiple-select questions, $200.
• Five sections weighted ~22/25/20/15/18%. Ingesting and processing is the heaviest.
• Find the constraint words, then rule out any option that breaks one of them.
• Learn both names for renamed products: Cloud DLP → Sensitive Data Protection, Dataproc → Managed Service for Apache Spark. The gcloud dataproc commands and API names are unchanged.