CDP-3002 Cloudera Data Engineer cover: Airflow, Iceberg and Deployment shown as three 10% blocks next to the 70% Spark and tuning block

 

In CDP-3002, the Cloudera Data Engineer exam, Airflow, Iceberg and Deployment are worth 10% each, so together they make up 30% of the exam, or roughly 15 of its 50 questions. Most study plans pour everything into Spark and treat these three areas as an afterthought. With a 55% pass mark, that is a risky bet.

 

This entry is about that 30%. For each of the three areas I set out what Cloudera's exam guide lists, what that means in practice, and a small hands-on task that covers it. The exam facts were read on Cloudera's own pages on 18 November 2026.

What is the 30% of CDP-3002 outside Spark and tuning?

It is three areas of equal weight: Airflow (10%), Deployment (10%) and Iceberg (10%). Spark (48%) and Performance Tuning (22%) make up the other 70%.

 

The official Cloudera Data Engineer exam guide gives the format and the topic list. The facts that matter for planning:

  • Exam: Cloudera Data Engineer (CDP-3002)
  • 50 questions in 90 minutes
  • Pass score: 55%
  • Delivery: online and proctored, through QuestionMark
  • Allowed resources: none, so no documentation open during the exam
  • Spark 48%, Performance Tuning 22%, Airflow 10%, Deployment 10%, Iceberg 10%
  • Price: $330 on the Cloudera Data Engineer exam store page

The topic lines under the three smaller areas are short:

  • Airflow: incremental extraction from a source system, scheduling ETL pipelines, scheduling quality checks, and working with DAGs
  • Deployment: using the API and CLI, and working in the Data Engineering Service
  • Iceberg: "Understand Iceberg"

That last line is the shortest on the page, yet the audience description at the top of the guide names Apache Iceberg next to formats, partitioning and schema design. Read it as a full topic, not a footnote.

Why can't you skip Airflow, Iceberg and Deployment?

Because the pass mark leaves little room. Here is the arithmetic, done on the assumption that each question counts the same (Cloudera does not say how items are weighted).

  • 55% of 50 questions is 27.5, so you need about 28 correct answers.
  • The 30% block is about 15 questions, roughly 5 per area.
  • If you got none of those 15 right, you would need 28 of the remaining 35, which is 80% on Spark and tuning.

80% on the hardest part of the exam is a lot to ask. Turn it around: if you get 10 of the 15 small-area questions right, you need only 18 of the 35 Spark and tuning questions, about 51%. A few evenings on Airflow, Iceberg and the CDE tools buy a lot of safety.

 

Time is the other reason. 90 minutes for 50 questions is 1.8 minutes, or 108 seconds, per question. A question on a topic you never practised eats that time quickly, because you end up reasoning from scratch.

 

The AnalyticsExam channel has a short walkthrough of how to prepare for CDP-3002 as a whole, which is a good overview before you go into the three areas below.

 

What does the Airflow 10% expect you to do?

It expects you to build and schedule a pipeline as Python DAG files, not just to define what a DAG is. All four topic lines are verbs: implement, schedule, schedule, work with.

 

Cloudera's page on Apache Airflow in Cloudera Data Engineering explains the setup. Each CDE virtual cluster includes an embedded Airflow instance, workflows are written as Python DAG files, and you can also connect CDE to your own Airflow deployment. CDE ships operators for running CDE jobs and for running SQL on Cloudera Data Warehouse, and it uses Airflow connections to hold the access details of the target system.

Incremental extraction

An incremental load reads only what changed since the last run. In Airflow the usual pattern is to use the time window that each scheduled run is given and to filter the source on it, so every run picks up one slice and nothing is read twice. Be clear on what happens when a run is retried or back-filled: a well-built incremental task reads the same slice again and writes the same result.

Scheduling ETL pipelines and quality checks

A pipeline is a set of tasks with dependencies: extract, transform with a Spark job, then load. A quality check is just another task placed after the load, which fails the run when a rule is broken (row count is zero, a key column has nulls, totals do not match). Know how a failed check stops the tasks that depend on it, and how retries work.

Working with DAGs

Know the basic parts of a DAG file: the schedule, the start date, the tasks, and the order between them. Know the difference between the DAG (the definition) and a DAG run (one execution of it). If Airflow is new to you, the core concepts section of the Apache Airflow documentation covers these parts in a short read.

 

Hands-on task: write one DAG with three tasks (extract, Spark transform, quality check), schedule it daily, then break the check on purpose and watch what the run does.

What does the Deployment 10% cover?

It covers getting Spark work into Cloudera Data Engineering and running it there: through the UI, the CLI and the API. The guide lists two lines: "Use the API and CLI" and "Work in the Data Engineering Service".

 

CDE is described in Cloudera's docs as a serverless service that lets you submit Spark jobs to an auto-scaling cluster. In practice you deal with a few objects:

  • Virtual cluster: the place where your jobs run, with its own embedded Airflow
  • Resource: a set of files (your PySpark script, a JAR, a config file) uploaded so jobs can use them
  • Job: the definition of what to run, which file, which settings
  • Job run: one execution of a job, with its logs and status

The CDE command line client can create and update jobs, show job details, manage resources and run jobs. Cloudera's CLI page adds two details that are easy to miss: the client is not forward compatible, so you download the client that matches your cluster version from that cluster's details page, and it can read a password file for non-interactive use, such as an automation script.

 

Hands-on task: take one PySpark script, upload it as a resource, create a job from it with the CLI, run it, and read the run's log. Then do the same job through the UI, so you can picture both paths in a question.

What does "Understand Iceberg" mean for the exam?

It means knowing what Apache Iceberg adds to a table compared with a plain Hive-style table, and how that changes the way you write, partition and query data with Spark.

 

The Apache Iceberg documentation lists the features worth knowing, in plain terms:

  • Schema evolution: add, drop, update or rename columns, with no side effects. Schema changes are metadata changes, so no data files are rewritten.
  • Hidden partitioning: Iceberg produces partition values from a column, such as the day of an event timestamp, and keeps track of that link. Queries filter on the column itself and Iceberg skips the files that cannot match, so users do not need to add partition filters by hand.
  • Partition evolution: the partition layout can change as data grows. Old data keeps its old layout and new data uses the new one.
  • Time travel: a query can read an exact earlier snapshot of the table, which makes results reproducible.
  • Version rollback: a table can be reset to an earlier good state.
  • Serializable isolation: table changes are atomic, so readers never see a half-written change.

Cloudera lists a free on-demand course for this area, DENG-152 Cloudera Iceberg Fundamentals, among the suggested training on the exam guide.

 

Hands-on task: create an Iceberg table partitioned by the day of a timestamp column, insert two batches, add a column, then query the table as of the first snapshot. You will have touched four of the six features above in about half an hour.

 

CDP-3002 infographic: hands-on checklist for the Airflow, Deployment and Iceberg areas of the Cloudera Data Engineer exam

How do the three areas connect in one pipeline?

They form one workflow, and studying them together is faster than studying them apart. A worked scenario, in my own words, shows how.

 

A team loads daily orders into an Iceberg table. A Spark job, deployed as a CDE job from an uploaded resource, reads one day of new orders. An Airflow DAG in the same virtual cluster runs that job every morning, then runs a quality check that fails if the day's row count is zero. One day the check fails because the source sent nothing. The team fixes the source, re-runs that day's DAG run, and the incremental task reads the same one-day window again. Later the business adds a discount column, which is an Iceberg schema change that needs no rewrite of old data, and an analyst who wants last week's figures exactly as they were reads an older snapshot.

 

If you can explain each step of that story, which tool does it and why, you have covered most of what the 30% asks.

What is a four-week plan for the 30%?

Run it beside your Spark study, at about two hours a week on top of it. Spark and tuning still get most of your time, since they are 70% of the exam.

  1. Week 1, Airflow basics: DAGs, tasks, dependencies, schedules, DAG runs. Write the three-task DAG from the Airflow section.
  2. Week 2, Airflow in practice: incremental extraction by time window, quality checks, retries and re-runs. Break the pipeline on purpose and fix it.
  3. Week 3, Deployment: virtual clusters, resources, jobs and job runs. Run the same job from the CLI and from the UI.
  4. Week 4, Iceberg: the six features above, then the full pipeline scenario, start to finish, in your own words.

At the end, write one page per area with the terms and the "why" behind each. Without any resources allowed in the exam, those pages are what you revise from in the last week.

Which mistakes cost points in the 30%?

Most come from studying the three areas as definitions instead of as tasks. These are the usual ones:

  • Treating "Understand Iceberg" as a single definition. The features behave differently, and questions can turn on which one solves a given problem.
  • Mixing up Iceberg's hidden partitioning with Hive-style partition columns that users must filter on themselves.
  • Writing an incremental load that reads "everything since the last success" instead of a fixed window, which breaks on re-runs.
  • Knowing Airflow but not how it sits inside CDE: the embedded instance per virtual cluster, the CDE operators and the connections.
  • Using only the UI for CDE and never touching the CLI, even though the guide names the API and CLI first.
  • Leaving the 30% for the last few days, when Spark revision will crowd it out.

What else should you know before booking CDP-3002?

For a checklist view of every CDP-3002 topic with its weight, the CDP-3002 syllabus and exam details page on AnalyticsExam is handy for the final week, when you tick off each area against your notes.

 

The exam is taken online with a proctor, through QuestionMark. Check the system requirements linked from the exam guide a few days before exam day, not on the day itself. Go in knowing your pace: about 108 seconds per question, so a question that is taking three minutes is a signal to move on.

 

If you also work on the platform side, our companion study blog has a post on the CDP-2001 Cloudera Administrator on-premises exam, which covers the cluster side of Cloudera that data engineers rely on.

Frequently asked questions

1. How much of CDP-3002 is Airflow, Iceberg and Deployment?

30% in total, 10% each. If the weights map onto question counts, that is about 15 of the 50 questions, roughly 5 per area.

2. What is the passing score for CDP-3002?

55%. With 50 questions, that is about 28 correct answers if every question counts the same. Cloudera does not say how items are weighted.

3. How long is the CDP-3002 exam and how much does it cost?

90 minutes for 50 questions, which is about 108 seconds per question. The exam costs $330 on Cloudera's store page.

4. Can I use documentation during the CDP-3002 exam?

No. The exam guide says no reference materials, white papers, user guides or other resources are allowed during the exam.

5. Is there free training for the Iceberg part of CDP-3002?

Yes. Cloudera lists DENG-152 Cloudera Iceberg Fundamentals as a free on-demand course among the suggested training on the exam guide.