Power BI dashboards using Direct Lake

[dbernstein_tp]

Hi Everyone, We are experimenting with using Direct Lake for our Power BI dashboards and I wanted to check in with this group to see if people have tried or are currently using this to connect PBI to Databricks tables. If so what is your experience with performance and in particular data governance?

[balajij8]

@dbernstein_tp 

Direct Lake is a Fabric feature designed to read Delta/Parquet files directly from One Lake storage.  For standard Power BI to Databricks cases (Power BI reports directly from Databricks tables), you can use below

• Direct Query on Databricks SQL Warehouses
• Import mode for scheduled batch extracts
• OpenSharing / Delta Sharing connector for cross-platform data exchange.

If data governance is the primary requirement, Direct Query is the recommended path. Because Direct Query sends native SQL queries back to the warehouse at runtime, all Unity Catalog security primitives including table permissions, row-level filters and column masks are strictly evaluated and enforced per user session. You can set large fact tables in Direct Query while configuring dimension tables to Dual mode if feasible that allows Power BI to optimize cross-table join execution plans while preserving access controls.

To hit sub 5 second interactive dashboard SLAs with Direct Query while maintaining live Unity Catalog governance, consider below

• Composite Models and Aggregations - Implement user-defined or automatic aggregation tables to serve high-level summarized metrics instantly while allowing drill-through to detail rows via Direct Query.
• Transformations at Source - Push complex business logic and transformations upstream into Databricks SQL views or materialized tables rather than relying on heavy DAX calculations.
• Warehouse Sizing and Concurrency - Configure query parallelization and scale out cluster endpoints on your SQL Warehouse to prevent visual concurrency bottlenecks during peak dashboard usage.

More details here

[NandanHegde15]

Your Power Bi report needs to be in Fabric capacity workspace in order for you leverage Direct Lake mode functionality.

So you can try the path of Databricks mirroring :

Fabric mirrors Unity catalog metadata and creates OneLake shortcuts to the Delta files in ADLS; no data movement, and no Databricks cluster needs to be running because Fabric reads storage directly using credential vending from Unity catalog. You get an auto-generated read-only SQL analytics endpoint and can build a Direct Lake model over it :

Zero-ETL Integration: Mirroring Databricks Unity Catalog in Microsoft Fabric | by Yasar Kocyigit | M...

Based on what I remember Unity catalog permissions do not carry across. Mirroring replicates metadata and file pointers, not ACLs, so Fabric becomes a second, independently-administered security perimeter. Column masks, row filters and dynamic views in UC all have to be re-authored.