All 35 lessons
Start here: cloud, data, Fabric and Power BI basics
- 1. Start here: the exam, the plan and what you already know
- 2. Cloud in plain words: SaaS, PaaS, IaaS and where Fabric fits
- 3. Data basics: files, tables, formats and the lakehouse idea
- 4. A tour of Microsoft Fabric: OneLake, workspaces and items
- 5. Power BI in plain words: semantic models, reports and DAX
Prepare data: get data in (exam area 2, 45-50%)
- 6. Connect to data and find it: connections, gateways, OneLake catalog and Real-Time hub
- 7. Choosing a data store: lakehouse, warehouse or eventhouse
- 8. Use data without copying: shortcuts, mirroring and OneLake integration
- 9. Ingest data: Dataflow Gen2, pipelines, Copy job and notebooks
- 10. Power Query in Dataflow Gen2: filter, data types, merge and new columns
- 11. Messy data: duplicates, missing values and nulls
Prepare data: transform and query (exam area 2)
Design and build semantic models (exam area 3, 25-30%)
- 18. Choose a storage mode: Import, DirectQuery, Direct Lake and Dual
- 19. Relationships: star schemas, bridge tables and many-to-many
- 20. DAX foundations: measures, filter context, CALCULATE and variables
- 21. DAX functions the exam tests: iterators, table filters, windows and information functions
- 22. Calculation groups, dynamic format strings and field parameters
- 23. Composite models and large semantic model storage format
Optimize models, then secure them (exam areas 3 and 1)
- 24. Direct Lake in depth: fallback, refresh and the two flavors
- 25. Incremental refresh for semantic models
- 26. Faster reports and DAX: finding and fixing slow queries
- 27. Who can see what: workspace roles and item permissions
- 28. Fine-grained security: rows, columns, objects and files
- 29. Governance: sensitivity labels and endorsement
Lifecycle, then exam practice (exam area 1)
24 min
Direct Lake in depth: fallback, refresh and the two flavors
Configure Direct Lake, control DirectQuery fallback and refresh, and choose between Direct Lake on OneLake and on SQL analytics endpoint.
By the end of this lesson you can
- Explain transcoding, framing and automatic updates, and when to turn updates off
- Predict when Direct Lake on SQL falls back to DirectQuery, and control it with DirectLakeBehavior
- Choose between Direct Lake on OneLake and Direct Lake on SQL analytics endpoint
- Spot guardrail and table-health problems and pick the fix
From earlier lessons · 1 of 2
What does a refresh of a Direct Lake semantic model copy?
Pick one answer.
From earlier lessons · 2 of 2
A relationship crosses two source groups in a composite model. A store with no matching row on the other side is missing from a table visual. Why?
Pick one answer.
Juniper Coffee's sales model runs on Direct Lake. At 6:10 one morning, a store manager opened it while the nightly Spark job was halfway through. Half of yesterday's sales were missing. A month later, IT added SQL row-level security to the warehouse, and the same report turned slow. Both stories come from the same few settings.
How Direct Lake answers a query
Exam depth: know it cold.
A Delta table is a set of Parquet files, which store data by column, plus a log of changes. When a DAX query arrives, the model works out which columns it needs. That includes columns used by relationships and measures.
Any needed column not yet in memory is loaded whole from OneLake. This on-demand load is transcoding. Loaded columns stay resident until they're evicted: after a reframe with changed data, after a long idle spell, or under memory pressure.
So a model warms up as people use it:
- Cold: nothing in memory. First queries are slower.
- Semiwarm: only changed data needs loading.
- Warm: the needed columns are in memory, as fast as Import.
- Hot: warm, and the query caches are filled too.
Framing: what refresh means here
Exam depth: know it cold.
Framing is what a refresh does in Direct Lake. The model reads the Delta log of each table and points at the latest Parquet files. It copies no data, so it usually takes seconds.
The moment of framing becomes the baseline. Queries see each Delta table as of the last successful framing, not necessarily its latest state. New files written afterward stay invisible until the next framing.
Framing is incremental: the model drops only the column segments whose files changed and keeps the rest in memory. That's a feature for consistency. During a long ETL load, the model can keep showing the last complete version.
Automatic updates and when to turn them off
Exam depth: know it cold.
Keep your Direct Lake data up to date is a model setting, on by default. Find it in the semantic model's settings, under Refresh. When it's on, the model frames itself whenever the Delta tables change, only for the tables that changed.
Leave it on for most models, especially with low-latency needs. Turn it off when users must not see data until an ETL run finishes. Then frame on your terms: refresh in the Fabric portal, on a schedule, or from the Power BI REST API or TOM as the ETL's last step.
After a non-recoverable refresh error, automatic updates pause. They resume after an on-demand refresh succeeds.
In Power BI Desktop live editing, Refresh also updates the schema. A scheduled refresh only reframes.
You are an analytics engineer at Juniper Coffee. A nightly pipeline rewrites the sales fact table, then three dimensions, over 40 minutes. Reports on the Direct Lake model must never show a half-finished load. What should you do?
Pick one answer.
DirectQuery fallback: who falls back, and when
Exam depth: know it cold.
Fallback means a query leaves Direct Lake and runs as DirectQuery against the SQL analytics endpoint. It returns the latest data, but it's slower.
Only Direct Lake on SQL falls back. Direct Lake on OneLake never does; it always runs Direct Lake only.
Direct Lake on SQL stays in Direct Lake only while all of these hold.
- No referenced table has SQL row-level security, masking or object-level security at the endpoint.
- No table is based on a non-materialized SQL view.
- No table exceeds a guardrail on Parquet files, row groups or rows. One table over a limit stops Direct Lake for the whole model.
- The model was framed after its Delta tables were created or changed.
A query that touches a table or column the user is denied by SQL security returns an error.
// Which tables fall back to DirectQuery, and why?EVALUATE TABLETRAITS()
Control fallback with DirectLakeBehavior
Exam depth: know it cold.
DirectLakeBehavior is a model property that decides what happens when Direct Lake conditions aren't met. It applies to Direct Lake on SQL only.
| Value | What happens | Use it for |
|---|---|---|
| Automatic (default). | Queries silently fall back to DirectQuery. | Production: reports keep working. |
| DirectLakeOnly. | Queries fail with an error. | Development: find and fix the cause. |
| DirectQueryOnly. | Every query uses DirectQuery. | Development: measure fallback speed. |
Set it in Model view: open the semantic model's Properties pane and change Direct Lake behavior. You can also set it with TOM or TMSL.
You are an analytics engineer at Juniper Coffee, building a Direct Lake on SQL model in a dev workspace. You suspect some tables silently fall back. You want any query that can't use Direct Lake to fail, so you can fix it. Which setting should you choose?
Pick one answer.
The two flavors, side by side
Exam depth: know it cold.
Both flavors load data from OneLake into VertiPaq. They differ in how they find tables and check permissions. Microsoft recommends Direct Lake on OneLake for new models.
| Feature | Direct Lake on OneLake | Direct Lake on SQL analytics endpoint |
|---|---|---|
| Sources. | One or more Fabric items with Delta tables, natively or through shortcuts. | One Fabric item with a SQL analytics endpoint. |
| Finds tables and checks access through. | OneLake APIs. | The SQL analytics endpoint. |
| Fallback to DirectQuery. | Never. | Yes, set by DirectLakeBehavior. |
| SQL views. | Not supported. | Allowed, but always fall back. |
| SQL endpoint RLS. | Not applied; queries succeed. | Queries fall back, or fail if fallback is off. |
| Data security. | OneLake security roles, or Read and ReadAll on the item. | SQL permissions: Read on the item plus SELECT. |
| Guardrail exceeded. | Refresh fails; model can't be queried. | Falls back; refresh succeeds with a warning. |
| Mix with Import tables. | Yes, in web modeling. | No. |
| Calculated columns on Direct Lake tables. | Yes (preview). | No. |
| Create from. | Power BI Desktop, OneLake catalog, SQL endpoint page. | SQL analytics endpoint page only. |
| Deployment pipeline data source rules. | Not supported; use M parameter rules. | Supported. |
You are an analytics engineer at Juniper Coffee. A new model needs tables from the sales Lakehouse, the loyalty Lakehouse and the finance Warehouse. Access must follow OneLake security roles, and queries must never switch to DirectQuery. Which storage mode should you use?
Pick one answer.
Guardrails by capacity SKU
Exam depth: recognize it.
Guardrails are per-SKU limits on each Direct Lake table, plus one limit on model size.
| SKU | Parquet files per table | Row groups per table | Rows per table | Model size on OneLake |
|---|---|---|---|---|
| F2 to F32. | 1,000. | 1,000. | 300 million. | 10 to 40 GB. |
| F64. | 5,000. | 5,000. | 1.5 billion. | Unlimited. |
| F512 and up. | 10,000. | 10,000. | 12 to 24 billion. | Unlimited. |
Model size is checked for the whole model. The others are checked per query. Max memory (25 GB on F64) isn't a guardrail. Past it, the model pages columns in and out, which slows queries.
Keep Delta tables Direct Lake friendly
Exam depth: recognize it.
Direct Lake is only as fast as the tables under it.
- V-Order: a Fabric write-time sort and compression for Parquet. V-Ordered files load faster and compress better. Direct Lake depends on it.
- Fewer, bigger files and row groups: run
OPTIMIZEto compact small files andVACUUMto remove old ones. This also clears guardrail breaches. If a table still exceeds limits, move to a bigger SKU. - Right data types and lower cardinality: integers instead of numbers stored as text.
With Direct Lake on SQL, disabling fallback also avoids slower hybrid query plans.
Try it in Fabric
A free Fabric trial is enough. Use a Lakehouse with a few tables.
- Open the Lakehouse's SQL analytics endpoint and select New semantic model.
- Name the model, pick Direct Lake on SQL under Storage, tick two tables and create it.
- Open the model. In Model view, open the model's Properties and set Direct Lake behavior to Direct Lake Only.
- In DAX query view, run
EVALUATE TABLETRAITS()and read DirectLakeFallbackInfo. - In the workspace, open the model's Settings. Under Refresh, find Keep your Direct Lake data up to date.
Exam-style practice
You are an analytics engineer at Juniper Coffee. A Direct Lake on SQL model's Sales table is based on the SQL view vw_SalesClean. Reports are slow, and TABLETRAITS shows the table falls back. You need it served in Direct Lake mode. What should you do?
Pick one answer.
You are a BI developer at Juniper Coffee. The Warehouse uses SQL row-level security so managers see only their region. The team moves the model from Direct Lake on SQL to Direct Lake on OneLake to stop fallback. What happens to managers' queries?
Pick one answer.
You are an analytics engineer at Juniper Coffee. On an F32 capacity, a Direct Lake on OneLake model's refresh fails. One loyalty table has 2,600 small Parquet files. You need the model working without buying a larger SKU. What should you do?
Pick one answer.
Worth keeping
- Direct Lake loads columns on demand and shows data as of the last framing.
- Automatic updates are on by default; turn them off and refresh after ETL for consistent loads.
- Only Direct Lake on SQL falls back; DirectLakeBehavior picks Automatic, DirectLakeOnly or DirectQueryOnly.
- Direct Lake on OneLake: many items, OneLake security, no fallback, and SQL RLS doesn't apply.
- Guardrails are per SKU; compact files with OPTIMIZE and keep V-Order on.
Direct Lake skips the big refresh. Import models still need one, and refreshing five years of till sales every night is a waste. Next lesson: incremental refresh, which reloads only the newest slice.
Leaderboard
Loading…
50 XP for each lesson you finish, 10 for each right answer. Names show only for learners who turn on Show me on leaderboards.
Sources for this lesson (8)
Each fact in the lesson is marked with the number of the page it comes from.
- [6]Microsoft Learn: Study guide for Exam DP-600: Implementing Analytics Solutions Using Microsoft Fabric
- [45]Microsoft Learn: Direct Lake overview
- [60]Microsoft Learn: Model relationships in Power BI Desktop
- [198]Microsoft Learn: How Direct Lake works
- [199]Microsoft Learn: Develop Direct Lake semantic models
- [247]Microsoft Learn: Integrate Direct Lake security
- [248]Microsoft Learn: Understand Direct Lake query performance
- [249]Microsoft Learn: Direct Lake in Power BI Desktop