Power BI & Mirrored Databricks Catalog — Table Access Forbidden errors despite Unity Catalog access

Hi Anjan,

Thanks for the follow-up — this is still not resolved, and I don't think it's non of the listed catalogue types listed in the limitation section, its a simple catalogue with xlsx file in it for testing purposes. Tried with other catalogues,same error.

We also reviewed  with Microsoft colleguew the HAR file and backend telemetry for the failing Fabric preview/query operations.

The Databricks-side validation showed that the source catalog, schema, and tables are visible, and a direct query against one of the source tables completed successfully from Databricks. This confirms that the source object exists and can be read directly from the Databricks side.

However, the Fabric-side requests are failing after the mirrored catalog has been created, during the preview/query flow.

Key observations

• The mirrored catalog, schema, and tables are visible in Fabric.
• The Databricks catalog, schema, and tables are visible directly in Databricks.
• A direct Databricks SQL query against the source table completed successfully.
• External Data Access is enabled on the Databricks metastore.
• The required Unity Catalog access was reviewed during the session.
• The HAR and telemetry show that the Fabric UI error is generic, but the backend errors are more specific.
• One failing Fabric preview request returned:

Invalid object name 'test'

• The telemetry shows a SQL exception while Fabric is executing the preview query through the Databricks mirrored catalog path:

Microsoft.Data.SqlClient.SqlException

Number: 208

Message: Invalid object name

• A separate HAR request through the OneLake query path also failed with

QueryFeatureMwcTokenAuthFailure

Status: 403

ErrorCode: AuthorizationFailure

Troubleshooting steps performed

• Reviewed the Fabric connection used by the mirrored Databricks catalog.
• Confirmed that the connection is configured with OAuth authentication.
• Reviewed Databricks Unity Catalog permissions.
• Reviewed the Databricks metastore configuration and confirmed External Data Access is enabled.
• Ran a direct Databricks SQL query against one of the source tables, which returned data successfully.
• Reviewed the HAR file for the failing Fabric preview and query requests.
• Reviewed backend telemetry using the Root Activity IDs from the failed operations.
• Confirmed that the failure occurs in the Fabric preview/query path after the catalog and table metadata are already visible.

Current findings / hypothesis

Based on the current evidence, this does not appear to be a straightforward Databricks-side table permission issue.

The main reason is that the source table can be queried successfully from Databricks, while the failure occurs when Fabric attempts to preview or query the mirrored object through the Fabric Databricks mirrored catalog and OneLake shortcut path.

At this stage, the issue appears more likely related to one of the following areas:

• Fabric mirrored catalog metadata/object resolution.
• Fabric SQL endpoint metadata availability for the mirrored catalog.
• OneLake shortcut authorization or shortcut state.
• Fabric-side access/execute permissions for the user or item.

The two backend errors are important for the next investigation phase:

Invalid object name 'test'

and

QueryFeatureMwcTokenAuthFailure / 403 AuthorizationFailure

Together, these suggest that Fabric can discover the mirrored structure, but the preview/query layer is not able to successfully resolve or authorize access to the object through the mirrored catalog/shortcut path.

i ws suggested Action Plan

• Please test the same preview operation with a user that has Fabric Workspace Admin or Member permissions, to help rule out a Fabric-side permission/execute access issue.
• If possible, create a new mirrored catalog using the same Databricks connection and the same source catalog/schema. This will help confirm whether the issue is isolated to the existing mirrored artifact.
• If available, please validate from the Fabric SQL endpoint whether the impacted table is present in the SQL endpoint metadata by running:

SELECT TABLE_SCHEMA, TABLE_NAME

FROM INFORMATION_SCHEMA.TABLES

WHERE TABLE_SCHEMA = 'test_usr'

AND TABLE_NAME = 'test';

Output: 

Admin/Member permissions: I am a Fabric Admin and have admin rights on the tested workspace, so this rules out a workspace-level permission gap on my account.

New mirrored catalog: Created a fresh mirrored catalog using the same Databricks connection and source catalog/schema. Same result — preview/query still fails.

SQL endpoint metadata check: Ran the suggested query against INFORMATION_SCHEMA.TABLES for test — it returned zero rows. 

So not resolved yet, Will appreciate any ideas/support.

Best,

Roman