Microsoft Fabric One and Databricks Unity Catalog — Bidirectional Access Without Data Movement

[amitsharma1707]

Hi everyone,

We're working in a hybrid environment with both Microsoft Fabric and Databricks, and I want to understand the options for virtualizing data between the two platforms without physically moving or duplicating it — similar to how shortcuts work in Fabric OneLake.

Specifically, I'm looking for guidance on both directions:

1. Accessing Fabric OneLake data from Databricks

• What's the recommended way to read Delta tables stored in OneLake directly from Databricks (e.g., registering OneLake paths as external tables in Unity Catalog via ABFS, using external locations/storage credentials)?
• Are there limitations with Unity Catalog external locations pointing to OneLake (authentication, service principals, workspace-level access)?
• Can OneLake data be governed under Unity Catalog while the data physically stays in OneLake?

2. Accessing Databricks Unity Catalog data from Fabric (like a shortcut)

• What's the current experience with OneLake shortcuts to ADLS Gen2 where Databricks Delta tables reside?
• Has anyone used the Fabric mirroring for Azure Databricks Unity Catalog feature? How well does it reflect UC metadata and keep tables in sync without copying data?
• Any gotchas with Delta table features (deletion vectors, column mapping, liquid clustering) breaking readability on the Fabric side?

3. General considerations

• Performance implications of cross-platform virtualization vs. native storage
• How governance/permissions behave when the same data is visible in both Unity Catalog and OneLake
• Real-world experiences: does this dual-visibility setup work reliably in production?
• @Genpact 

[kartikchoudhary]

One thing I'd evaluate before choosing between virtualization and replication is the ownership model.

If Fabric and Databricks both need to operate on the same data, I'd look at four areas: authentication, governance/permissions, Delta feature compatibility, and workload performance.

The biggest risk isn't necessarily whether the platforms can technically read the data — it's whether governance and metadata remain consistent when the same underlying data is being exposed through two different control planes.

For production, I'd also test the specific Delta features being used rather than assuming all tables will behave identically across both environments.