Forum Discussion
Composite Models in Fabric
- 2 years ago
From Power BI desktop, you can live connect to a semantic model created off of a lake house, and combine with other sources. However, queries to the semantic model (created off of the lakehouse), will fallback to DirectQuery (vs DirectLake).
Hi everyone,
I’m facing an issue with an embedded Power BI composite report and would appreciate any guidance.
Scenario:
The report uses two data sources:
A Fabric semantic model (DirectQuery)
A Lakehouse table (Import mode) created from an Excel file
The report works correctly in the Power BI/Fabric workspace, with data visible from both sources.
The report uploads successfully via our custom UI.
Problem:
When the report is opened via our embedded web application:
Only the Lakehouse (Import) visuals load
Visuals that rely on the Fabric semantic model (DirectQuery) fail
Error received: “Failed to move the data reader to the next row”
This happens even though the report, semantic model, and Lakehouse are in the same workspace
What we’ve already done:
Verified permissions on the workspace and semantic model
Confirmed dataset and report IDs are correctly referenced
Implemented a manual step to inject all upstream datasets into the embed token (not just the immediate dataset), using a configuration table
The issue still persists
Questions:
Are there known limitations when embedding composite models (DirectQuery + Import)?
Are there additional permissions or configuration steps required for embedded scenarios?
What typically causes DirectQuery semantic model visuals to fail only in embedded apps?
Any insights or similar experiences would be greatly appreciated.
Thanks in advance!