Mode for Fabric semantic models that loads Delta or Parquet files from OneLake straight into VertiPaq, giving almost import-level performance without duplicating data.
Read more: Microsoft Learn
In the Ultra Transcenders books
AZ-305DP-900DP-600DP-700PL-300
Each book explains Direct Lake in context, with comparison tables and the common traps.
Terms in this definition
- Parquet
Columnar file format, open source and aimed at analytics. Event Hubs Capture cannot produce it natively.
- OneLake
Built on Azure Data Lake Storage Gen2, it is the one logical data lake for an entire Microsoft Fabric tenant, provisioned automatically, and the place where every Fabric workload keeps its data.
- VertiPaq
The in-memory engine behind Power BI; Direct Lake pulls Delta and Parquet data into it.
- Performance
Routing method in Traffic Manager that directs users to whichever endpoint offers the lowest latency.
Related terms
- Assume referential integrity
A relationship setting in DirectQuery and Direct Lake models that lets the engine send faster INNER JOIN queries to the source rather than OUTER JOINs. Only switch it on when every key on the many side has a match, otherwise unmatched fact rows quietly vanish from results.
- Direct Lake on OneLake
Recommended choice for new Direct Lake models: it pulls Delta tables from one or several Fabric items via OneLake APIs, with OneLake security applied, allows composite models and never switches to DirectQuery.
- Direct Lake on SQL
A Direct Lake variant that relies on the SQL analytics endpoint of a single item to discover tables and check permissions. Queries switch to DirectQuery when they hit SQL views, row-level security defined on the endpoint or exceeded guardrails, subject to the DirectLakeBehavior setting.
- DirectQuery
Power BI connection mode in which reports fetch data from the source in real time; Direct Lake outperforms it.
- Fixed identity
With single sign-on off, a Direct Lake model can be bound to one explicit cloud credential, such as a workspace identity or service principal. Permissions, RLS and CLS are then evaluated for that credential, not per viewer, so readers need no rights on the underlying item.
- Formula engine
Plans how a DAX query will be answered inside a semantic model, then asks the storage engine, VertiPaq for Import and Direct Lake, to fetch data and do the first aggregations.
- Framing
Direct Lake's way of refreshing: it reads the Delta log and repoints the model to the newest Parquet files, a metadata-only step, so queries reflect data as at the most recent successful run. A default-on option, Keep your Direct Lake data up to date, repeats it automatically as tables change.
- Live editing
When Power BI Desktop opens a Direct Lake semantic model this way, every change is written straight back to the copy in Fabric. Nothing is saved locally or published, Report view is unavailable, and each session starts a fresh version history point.