Memory hook: Choose the engine for the work, then decide whether to copy.
Must remember
A lakehouse suits Delta tables, files and Spark-based engineering with a SQL analytics endpoint for supported queries. A warehouse suits relational T-SQL warehousing and transactional data manipulation. An Eventhouse/KQL database suits time-oriented events, logs and interactive exploration. Select the store by query pattern, transformation tools and operational needs, not by the word lake alone.
Discover governed assets in OneLake catalog and streaming/event sources in Real-Time hub. Discovery does not grant access. Connections bind a source and authentication method; gateways may be needed for sources outside directly reachable cloud paths. Separate author credentials from the identity used for unattended execution.
Ingestion copies data into a destination on a schedule or continuously. A shortcut exposes supported existing storage without maintaining another full copy. Mirroring replicates supported operational changes with its own source/type/latency constraints. Shortcuts still require permissions, and their performance/availability can depend on the referenced source.
OneLake integration can expose supported Eventhouse data in Delta format for other analytical engines. Check enablement, availability latency and supported data types. Semantic-model OneLake integration exposes supported model-table data for downstream reuse; it does not mean exported rows include every interactive measure, report filter or security behavior.
Use the lakehouse SQL endpoint for its supported read/query surface; do not assume it permits warehouse-style table DML. Keep transformations in the appropriate engine, then verify the published schema and freshness before building dependent models.
Choose under exam pressure
| Requirement | Choice and reason |
|---|---|
| Large event stream queried with KQL | Eventhouse. |
| SQL-centric dimensional warehouse | Warehouse. |
| Reuse existing supported storage without a duplicate copy | A shortcut with validated permissions and source dependence. |
Traps
- A shortcut is not a backup of the source.
- A visible catalog item is not proof of data access.
Active recall
1. Lakehouse versus warehouse?
Files/Delta/Spark-oriented engineering versus relational SQL-centric warehousing.
2. What does Real-Time hub help discover?
Streaming and event data sources.
3. Why check gateway requirements?
The execution service must reach the source using a supported network/authentication path.
4. Does OneLake integration preserve every measure result?
No; table availability is different from all interactive semantic-model calculations.
5. Why verify freshness after integration?
A downstream representation may become available later than the original write.