Databricks Data Engineer Professional study plan

Six weeks at 8–10 hours a week. Nine sections is a lot of surface, so this plan groups them rather than treating each separately. It follows the outline that applies from 9 October 2026; the exam format page lists what changed.

You need a Databricks workspace with enough data to make performance visible. Several sections — optimisation, debugging, monitoring — cannot be learned on toy datasets where everything is instant.

Hold the Associate already? See the compressed plan at the end.

Weeks 1 and 2: developing code (23%)

The largest section, two weeks.

Week 1 — reading and reasoning about code:

• PySpark DataFrame operations: selects, filters, joins, aggregations, window functions.
• The same logic expressed in SQL, and when each is preferable.
• Lazy evaluation and what actually triggers work.
• Reading a snippet and predicting its output — the core exam skill.

Week 2 — writing it well:

• Structuring reusable, testable transformation code.
• Common inefficiencies: unnecessary shuffles, collecting to the driver, row-by-row operations.
• User-defined functions in Pandas, Python and SQL, Unity Catalog functions, and why Python UDFs are usually the slow option.
• Lakeflow Declarative Pipelines: streaming tables versus materialised views, AUTO CDC with SCD Type 1 and 2.
• Lakeflow Jobs with If/Else and For Each tasks; serverless and dependency settings.
• Stateful Structured Streaming: watermarks, output modes, foreachBatch, checkpoints.
• Tests with assertDataFrameEqual and assertSchemaEqual; project layout for Declarative Automation Bundles.

Week 3: cost and performance optimisation (15%)

The second largest section and the one that most distinguishes this exam.

• Reading the Spark UI: stages, tasks, shuffle volume, spill.
• Data skew — recognising it and redistributing.
• Small files — how they arise and how compaction fixes them.
• Liquid Clustering, CLUSTER BY AUTO, deletion vectors and Predictive Optimization, compared with partitioning and Z-ordering.
• Change Data Feed and the Delta cache.
• Join strategies, including broadcasting small tables.
• Caching — when it helps and when it wastes memory.
• Cluster sizing, autoscaling, and terminating idle compute.

Week 4: ingestion (12%), data manipulation (12%), modelling (5%)

Three related sections, 29% together.

• Ingestion and acquisition: formats including Iceberg, message buses, CDC pipelines into Delta or Iceberg, Lakeflow Connect for SQL Server, MySQL and PostgreSQL, plus OpenSharing (formerly Delta Sharing), Clean Rooms and Lakehouse Federation, which moved here from their old section.
• Data manipulation: window functions, joins and aggregations at scale, VARIANT for semi-structured data, ai_query, and expectations that keep, drop or fail on bad records.
• Data modelling: table layout for large tables, dimensional models and slowly changing dimensions, materialised views and metric views.

Week 5: monitoring (10%), debugging and deploying (10%), security (8%), governance (5%)

Four sections, 33% together. Cover each properly but briskly.

• Monitoring and alerting: system tables, event logs, the Jobs API and alerts; detecting a pipeline that succeeds while producing wrong data.
• Debugging and deploying: diagnosing failures from logs and the Spark UI, job repairs with parameter overrides, bundles and Git folders.
• Security and compliance: least privilege, ABAC with governed tags, masking and pseudonymisation, PII controls, data purging.
• Governance: tags and comments for discovery, and the Unity Catalog permission inheritance model.

Week 6: practice and repair

• Full practice exam under real conditions: 120 minutes, 60 questions.
• Sort mistakes by section. With nine sections, look for clusters rather than individual misses.
• Rebuild the weakest two, then take a second practice exam.

Where the hours go

Compressed plan for Associate holders

Three to four weeks:

1. Weeks 1–2: developing code. Entirely new emphasis.
2. Week 3: cost and performance optimisation, plus data modelling, CDC with Lakeflow Connect, sharing and federation.
3. Week 4: practice exam, then repair whatever the score report flags.

Do not restudy basic ingestion, governance and transformation — verify with practice questions and move on. Do learn VARIANT and ai_query, which the Associate outline does not name, and go deeper on ABAC.

The small sections are the trap. Security, governance and modelling are 8%, 5% and 5%, which feels negligible until you notice they total 18% of the exam. Give each its hours; give none of them zero.