Direct Lake storage mode
1. Direct Lake storage mode
In the last lesson, you created a semantic model from a warehouse. Next, you'll see how that model reads the warehouse's data through Direct Lake, Fabric's storage mode built for large, changing tables.2. Direct Lake reads Fabric Delta tables
Semantic models traditionally face a tradeoff. Import mode copies data into the model: queries are fast, but the copy is only as fresh as its last scheduled refresh. DirectQuery sends every query to the source: data stays current, but reports can be slower. Direct Lake avoids that tradeoff for Fabric data. It reads Delta tables stored in OneLake directly, loading only the columns a query needs into memory. Lakehouses and warehouses both qualify.3. Tracing the model back to its source
Since a Direct Lake model reads straight from its source, you need to know which Fabric item that is. Suppose a Sales report is missing yesterday's transactions. Before investigating the data itself, you'd confirm the model reads from the expected Sales warehouse. Lineage shows exactly that: the Fabric item supplying the model's tables. Find the Lineage command toward the right end of the semantic model ribbon.4. How Direct Lake lineage works
In the lineage view, each item appears as a card, and arrows show how data flows. For a model created with New semantic model, the warehouse or lakehouse card sits upstream, with an arrow into the semantic model. That source item stores the Delta tables the model reads.5. Choose a Direct Lake route
Direct Lake models reach their tables in one of two ways. With Direct Lake on OneLake, the model reads Delta tables straight from OneLake, so it can combine tables from several items, such as a lakehouse and a warehouse. With Direct Lake on SQL, the model goes through a SQL analytics endpoint, the SQL interface every lakehouse and warehouse provides. The model then uses only that endpoint's tables, so it can't combine a warehouse with a separate lakehouse, but it suits teams already querying that endpoint. Either way, the data stays in OneLake.6. Refresh frames the latest table state
Whichever route you choose, the source tables keep changing. Each change to a Delta table creates a new version, and framing points the model at the latest one. It updates references rather than copying rows, so it's much lighter than an Import refresh. The Keep your Direct Lake data up to date setting frames automatically by default. You can also refresh manually, for example when automatic updates are turned off.7. Fallback changes the query path
Framing changes which data the model sees. Fallback changes how a query runs. On the SQL route, when a model's Direct Lake behavior setting is Automatic and a query can't be served by Direct Lake, it switches to DirectQuery. The SQL analytics endpoint then runs the query, so results still appear, but more slowly. Because the endpoint reads the tables directly, these queries return the latest data, even before the model is reframed. Direct Lake on OneLake has no fallback.8. Permissions and limits affect the query path
Access and capacity both affect whether queries stay in Direct Lake. Users need permission to the model, and the model's connection needs permission to the source. On the SQL route, security rules on the endpoint can trigger fallback, and so can exceeding a capacity guardrail, a limit on how much data loads into memory. On the OneLake route, exceeding a guardrail makes the refresh fail instead. So check how queries actually run rather than assuming they stay in Direct Lake.9. Let's practice!
Let's practice!Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.