1.4 — Workspace-Level and Item-Level Access Controls
DP-700 Domain 1 · Skill 1.4
What You Need to Know
Fabric security has two planes. The control plane decides what you can do with items (workspace roles and item permissions). The data plane decides which data you can see (OneLake security roles and SQL permissions, covered in lesson 1.5). Least-privilege questions usually come down to picking the smallest grant that meets the requirement.
Workspace roles
| Capability | Admin | Member | Contributor | Viewer |
|---|---|---|---|---|
| Update or delete the workspace; add or remove admins | ✅ | |||
| Connect the workspace to Git; create a workspace identity | ✅ | |||
| Add Members and lower roles; let others reshare | ✅ | ✅ | ||
| Create and edit items; write data to lakehouses, warehouses and KQL databases | ✅ | ✅ | ✅ | |
| Run or cancel notebooks and pipelines | ✅ | ✅ | ✅ | |
| Read lakehouse data through Spark, OneLake APIs and Lakehouse explorer (ReadAll) | ✅ | ✅ | ✅ | |
| Connect to the SQL analytics endpoint and read with T-SQL (ReadData) | ✅ | ✅ | ✅ | ✅ |
| View pipeline, notebook and KQL item content and run output | ✅ | ✅ | ✅ | ✅ |
- A workspace role covers every item in the workspace. If a user should see only one item, share that item instead of adding them to the workspace.
- Assign roles to security groups where you can. A user in several groups gets the highest role among them.
- Viewers can query SQL endpoints, but they can't read OneLake files through Spark unless a OneLake security role grants it.
Item permissions (sharing)
Sharing gives access to one item without workspace membership. Sharing a lakehouse always grants Read, plus whichever of these you tick:
| Lakehouse permission | Allows |
|---|---|
| Read (always) | Open the item and see its metadata; connect to its SQL analytics endpoint |
| Read all with SQL analytics endpoint (ReadData) | Query all tables with T-SQL (subject to SQL security) |
| Read all with Apache Spark (ReadAll) | Read tables and files through Spark, notebooks and OneLake APIs |
| Execute Apache Spark | Run Spark jobs through the lakehouse Livy endpoint |
A warehouse share works the same way, and only workspace Admins and Members can share a warehouse:
- Read alone: the recipient can connect to the SQL endpoint, the equivalent of SQL CONNECT. To query anything, they still need a T-SQL
GRANTon the specific objects. - Read all data using SQL (ReadData): read-only access to every table and view, the equivalent of
db_datareader. - Read all data using Apache Spark (ReadAll): read the warehouse's files in OneLake through Spark.
- Monitor: query DMVs and Query Insights.
- Audit: configure and query SQL audit logs.
ReadData and ReadAll are separate permissions that don't overlap.
Choosing the smallest grant
| Requirement | Grant |
|---|---|
| Analysts query a lakehouse's tables with SQL, see no other workspace items, and can't use Spark | Share the lakehouse only, with Read all with SQL analytics endpoint |
| Data scientists read lakehouse files in notebooks from another workspace | Share with Read all with Apache Spark |
| An engineer builds and runs pipelines and notebooks in the workspace | Workspace Contributor |
| Someone manages access for others, without deleting the workspace | Member |
| Read-only access to one item's tables, but only some folders or tables | Share the item (Read), then add a OneLake security role (lesson 1.5) |
Separating sensitive data
Workspace roles can't hide one table from a Contributor, because Contributors can read and write all data in the workspace. If an engineer must write to some tables but must not read a sensitive table, put the sensitive table in another workspace (or another item the engineer has no access to), and keep the engineer as Contributor in the original workspace.
Key Concept: Workspace roles apply to all items. Sharing applies to one item. Permissions are additive: a grant never takes away access that someone already has from another grant.
Exam Traps
- Making someone a workspace Viewer when they should see only one lakehouse (Viewer exposes every item's metadata)
- Expecting Read all with SQL analytics endpoint to allow Spark reads, or the reverse
- Trying to block a Contributor from one table with a OneLake security role (roles don't restrict Admins, Members or Contributors)
- Granting Member when Contributor already covers creating and running items
Check your understanding
Practice Scenario
Data scientists who work in another workspace need to load the Parquet files and Delta tables of the lakehouse LH_Marketing into Spark DataFrames in their own notebooks. They must not get a role in the LH_Marketing workspace or see any of its other items. What should you do?
Log in to save your progress.
Sources
- Roles in workspaces in Microsoft Fabric, Microsoft Learn
- How lakehouse sharing works, Microsoft Learn
- Share your warehouse and manage permissions, Microsoft Learn
- Data security in OneLake, Microsoft Learn