DP-700 cheat sheet
Last-minute reference for DP-700. The three skill areas carry equal weight, so revise all three.
Exam facts
| Duration | 100 minutes |
| Passing score | 700 / 1000 |
| Questions | Not published |
| Price | Set per country or region; shown at booking |
| Languages | 10 |
| Renewal | Annual, free |
Skill areas
| Area | Weight |
|---|---|
| Implement and manage an analytics solution | 30–35% |
| Ingest and transform data | 30–35% |
| Monitor and optimize an analytics solution | 30–35% |
Requirement to answer
| Requirement | Answer |
|---|---|
| Low-code transformation | Dataflow Gen2 |
| Code at scale, libraries | Notebook |
| Orchestrate steps, retries, alerts | Pipeline |
| Run when a file arrives | Event-based trigger |
| Fixed cadence | Schedule |
| Different lakehouse per stage | Deployment rule |
| Version control | Git integration |
| Shared libraries for notebooks | Environment |
| Filter rows per user | Row-level security policy |
| Hide values, keep rows | Dynamic data masking |
| Access to one folder | OneLake security role |
| Organisation-approved content | Certified endorsement |
| Who did what | Audit log in Purview |
| Data without copying | OneLake shortcut |
| Keep a database in sync | Mirroring |
| Only changed rows | Incremental load with a watermark |
| Latest row per key | row_number over a window, keep 1 |
| Sub-second time series | Eventstream, Eventhouse, KQL |
| Complex logic on streams | Spark structured streaming |
| Faster KQL on shortcut data | Query acceleration |
Window types
| Window | Remember |
|---|---|
| Tumbling | Fixed, no overlap; KQL bin() |
| Hopping | Fixed, overlapping by hop |
| Sliding | Output on enter and leave |
| Session | Ends after inactivity gap |
| Snapshot | Same timestamp |
Symptom and cure
| Symptom | Cure |
|---|---|
| Many small files | OPTIMIZE |
| Old unreferenced files | VACUUM (default retention 7 days) |
| Shuffle on small lookup | Broadcast join |
| Slow Copy of large table | Partition option, copy parallelism |
| Stale warehouse plans | Update statistics |
| Recent data slow in Eventhouse | Caching policy |
| Old data accumulating | Retention policy |
| Missing Eventhouse records | .show ingestion failures |
| Notify on failure | On fail path with Teams or email activity |
Traps
- OPTIMIZE and VACUUM are not the same.
- Masking is not blocking. Rows and access are unchanged.
- Workspace roles override narrow controls.
- Promoted is not Certified.
- A shortcut reads in place; a copy duplicates.
- Do not load facts before dimensions.
- collect() on big data causes driver out-of-memory.
Night-before checklist
Read the outline’s three headings, say the window types, and take one last practice test. Sleep. Details in the exam day guide.