DP-750 cheat sheet
Last-minute reference for DP-750. Preparing data and deploying pipelines are the two largest areas, at 30–35% each.
Exam facts
| Duration | 120 minutes |
| Passing score | 700 / 1000 |
| Questions | Not published |
| Price | Set per country or region; shown at booking |
| Languages | 10 |
| Renewal | Free, every 12 months |
Skill areas
| Area | Weight |
|---|---|
| Prepare and process data | 30–35% |
| Deploy and maintain data pipelines and workloads | 30–35% |
| Set up and configure an Azure Databricks environment | 15–20% |
| Secure and govern Unity Catalog objects | 15–20% |
Requirement to answer
| Requirement | Answer |
|---|---|
| Scheduled unattended run, pay only for the run | Job compute or serverless jobs |
| Many concurrent SQL and BI users | SQL warehouse |
| Faster cluster start | Pool, or serverless |
| Stop paying for idle interactive compute | Auto-termination |
| Isolate prod from dev | Separate catalogs |
| Files must survive DROP TABLE | External table |
| Query a database in place | Connection plus foreign catalog |
| Read a table | SELECT plus USE CATALOG plus USE SCHEMA |
| Hide rows by user or group | Row filter |
| Mask values by user or group | Column mask |
| Rule driven by tags across many tables | ABAC |
| Secret without hard-coding | Key Vault-backed secret scope |
| Storage access without a stored key | Managed identity |
| Share data with an outside party | Delta Sharing (now OpenSharing) |
| Who uses this column? | Lineage in Catalog Explorer |
| Keep history of changes per row | SCD type 2 or temporal table |
| Upsert from a CDC feed | MERGE |
| High-cardinality or changing filters | Liquid clustering |
| New files only, scalable, schema evolution | Auto Loader |
| SQL-only incremental file load | COPY INTO |
| Managed ingestion from supported sources | Lakeflow Connect |
| Continuous events from Event Hubs | Structured Streaming with a checkpoint |
| Bad rows dropped, pipeline continues | Expectation with drop action |
| Stop when bad data appears | Expectation with fail action |
| Reject mismatched schema | Schema enforcement |
| Accept a new column | Schema evolution |
Jobs and deployment
| Need | Use |
|---|---|
| Order of tasks | Task dependencies |
| Start on new files | File arrival trigger |
| Transient failure | Automatic retries |
| Rerun only failed part | Repair run |
| Cancel a run | Stop |
| Package and deploy as code | Bundle, databricks bundle deploy -t <target> |
| Change history and review | Git folders, pull requests |
Tuning and monitoring
| Symptom or goal | Tool or command |
|---|---|
| Small files slow reads | OPTIMIZE |
| Old unreferenced files cost storage | VACUUM (default 7-day retention) |
| One task far slower than others | Skew; Spark UI and query profile |
| Disk spill | More memory per task |
| Central logs and alerts | Log Analytics and Azure Monitor |
Traps
- SELECT alone is not enough to read a table.
- OPTIMIZE compacts; VACUUM deletes. They are not interchangeable.
- Managed tables lose their data on DROP; external ones do not.
- Names change: Lakeflow Spark Declarative Pipelines, Declarative Automation Bundles.
Take the practice test next, or revisit the exam tips.