Forum Discussion
User data functions : How to access Lakehouse Fabric SQL Endpoint
Hi Team,
I’m trying to access a Lakehouse SQL analytics endpoint from a Fabric User Data Function (UDF) using the documented approach.
- Fabric lakehouses for read/write operations for Lakehouse files and for read-only operations for the SQL Endpoint.
I configured a Lakehouse connection and used the following pattern:
@udf.connection(argName="lakehouse", alias="Lakehousename")
def get_lakehouse_data(lakehouse: fn.FabricLakehouseClient):
conn = lakehouse.connectToSql()
cursor = conn.cursor()
cursor.execute("SELECT TOP 10 * FROM job")Observations:
- Connection to SQL endpoint is successful
But queries fail with:
Invalid object name 'job'- The table exists under Lakehouse → Tables
- It is also visible in the SQL endpoint query editor
Questions:
- Is accessing Lakehouse SQL endpoint from UDF fully supported?
- Are there any limitations around table visibility or schema resolution?
- Is additional configuration required (e.g., metadata sync, permissions, or connection binding)?
- Are there any working end-to-end samples for this scenario?
The documentation mentions SQL endpoint read support, but I’m unable to get it working in practice.
- Accessing a Lakehouse SQL analytics endpoint from a Fabric User Data Function is supported in read-only mode, but the UDF execution context does not set the Lakehouse as the default database; as a result, it might fail.Why don't you try with a three-part names such as lakehousename.dbo.tablenameThe following worked for me:@udf.connection(argName="lakehouse", alias="directdacpacdep")@udf.function()def get_data(lakehouse: fn.FabricLakehouseClient) -> list:conn = lakehouse.connectToSql()cursor = conn.cursor()cursor.execute("""SELECT TOP 10 *FROM direct_dacpac_deploy.dbo.dimension_city""")return cursor.fetchall()Additionally to answer your questionsQn1: Yes, this scenario is supported in read‑only mode. You can connect by connectToSql() but they cannot perform inserts, updates, deletes, or execute Spark operations only select.
Qn2: Yes. In a UDF, the SQL endpoint connection does not set the Lakehouse as the default database, so 3 part naming system is required.
Hope this helps!
Thanks
Aparnaa
Thank you everyone for your responses and guidance. I was able to get this working and wanted to share a complete working sample along with a few key observations that might help others.
Sample Code (Working)
import fabric.functions as fn
udf = fn.UserDataFunctions()
@udf.connection(argName="lakehouse", alias="LHNAME_Test")
@udf.function()
def get_lakehouse_data(lakehouse: fn.FabricLakehouseClient) -> list:
conn = lakehouse.connectToSql()
cursor = conn.cursor()# Note: Use schema.table (avoid database.schema.table)
cursor.execute("SELECT TOP 10 * FROM LHNAME_Test.dbo.cdm_job")
columns = [desc[0] for desc in cursor.description]
rows = cursor.fetchall()data = [dict(zip(columns, row)) for row in rows]
cursor.close()
conn.close()return data
Key Observations / Learnings
1. Library Management Issues
I encountered the following error:
ModuleNotFoundError: No module named 'fabric' ERROR: Cannot install fabric-user-data-functions ... because of conflicting dependencies (pyarrow etc.)👉 Root cause:
- Multiple versions of fabric-user-data-functions were added
- Conflict with pre-installed dependencies in Fabric runtime
👉 Resolution:
- Remove all duplicate versions from Library Management
Add only a single entry:
fabric-user-data-functions- Do not manually add pyarrow
3 Replies
- Accessing a Lakehouse SQL analytics endpoint from a Fabric User Data Function is supported in read-only mode, but the UDF execution context does not set the Lakehouse as the default database; as a result, it might fail.Why don't you try with a three-part names such as lakehousename.dbo.tablenameThe following worked for me:@udf.connection(argName="lakehouse", alias="directdacpacdep")@udf.function()def get_data(lakehouse: fn.FabricLakehouseClient) -> list:conn = lakehouse.connectToSql()cursor = conn.cursor()cursor.execute("""SELECT TOP 10 *FROM direct_dacpac_deploy.dbo.dimension_city""")return cursor.fetchall()Additionally to answer your questionsQn1: Yes, this scenario is supported in read‑only mode. You can connect by connectToSql() but they cannot perform inserts, updates, deletes, or execute Spark operations only select.
Qn2: Yes. In a UDF, the SQL endpoint connection does not set the Lakehouse as the default database, so 3 part naming system is required.
Hope this helps!
Thanks
Aparnaa
The whole point of Fabric UDFs is data writeback, so you will want to use the lakehouse directly, or a Fabric SQL database.
Thank you everyone for your responses and guidance. I was able to get this working and wanted to share a complete working sample along with a few key observations that might help others.
Sample Code (Working)
import fabric.functions as fn
udf = fn.UserDataFunctions()
@udf.connection(argName="lakehouse", alias="LHNAME_Test")
@udf.function()
def get_lakehouse_data(lakehouse: fn.FabricLakehouseClient) -> list:
conn = lakehouse.connectToSql()
cursor = conn.cursor()# Note: Use schema.table (avoid database.schema.table)
cursor.execute("SELECT TOP 10 * FROM LHNAME_Test.dbo.cdm_job")
columns = [desc[0] for desc in cursor.description]
rows = cursor.fetchall()data = [dict(zip(columns, row)) for row in rows]
cursor.close()
conn.close()return data
Key Observations / Learnings
1. Library Management Issues
I encountered the following error:
ModuleNotFoundError: No module named 'fabric' ERROR: Cannot install fabric-user-data-functions ... because of conflicting dependencies (pyarrow etc.)👉 Root cause:
- Multiple versions of fabric-user-data-functions were added
- Conflict with pre-installed dependencies in Fabric runtime
👉 Resolution:
- Remove all duplicate versions from Library Management
Add only a single entry:
fabric-user-data-functions- Do not manually add pyarrow
