Forum Discussion
Gold Layer + Semantic Model + Reporting - Security for TDA question
- 6 months ago
Hi fabricpribeiro,
there is currently no single Microsoft document that describes this architecture end‑to‑end. However, the design pattern is possible when the relevant components are combined: Microsoft Fabric security, Direct Lake with a fixed (workspace) identity, semantic model RLS, OneLake security(Preview), and Microsoft Entra ID B2B. PFB referencesdue to access permission constraints and licensing limitations in our environment, we were unable reproduce at our end, could let us know if you could able achieve as below suggested? and also please refer some useful docsRecommended approach: Use Microsoft Entra ID security groups for user personas (All, Power, Advanced, Master). External users can be onboarded via Entra ID B2B and added to the same groups for consistent access management.
For Gold → Semantic Model access, use a single fixed identity (workspace identity / service principal) that holds permissions on Gold Lakehouses. End‑user access is enforced via RLS in the semantic model, avoiding direct Gold access and permission sprawl while enabling cross‑workspace reuse.
OneLake security governs storage‑level access (data at rest) and does not replace semantic model RLS, which handles user‑level filtering.
Persona model: All Users view reports; Power Users build reports on semantic models; Advanced Users extend models with read access to Gold; Master Users have direct Gold access.
This aligns with Fabric security best practices: encryption at rest (OneLake), TLS in transit, and Entra ID–based service‑to‑service authentication.
Thanks,Prashanth
each workspace representing a data domain.
What kind of company size is this? How many workspaces do you have across all capacities?
Hello, its a mid size company, probably around 200 users
We have one cacpacity for ingestion, one capacity for reporting and then separade capacities for the ones (domains) which are more consuming
But the question is about security not so much about performance
Thanks a lot,
Pedro