An Azure service for ingesting, preparing, and transforming data at scale.
Hello @豊田 明
Based on the tests you've already completed, I agree that this doesn't look like a basic Azure SQL permission or Managed Identity configuration problem.
Microsoft currently lists Azure SQL Database with MSI (Managed Identity) as a supported source/authentication combination for ADF Power Query/Data Wrangling.
The interesting part of your configuration is the network path:
- Azure SQL public access is disabled
- SQL is reached through a Managed Private Endpoint
- Copy Activity and dataset preview work
- Power Query Online evaluation fails
- Your Managed VNet IR isn't available under Run on Azure IR
Power Query Activity doesn't execute through exactly the same path as Copy Activity. Microsoft documents that ADF translates the Power Query M script into a data flow script and executes it using the ADF Data Flow Spark environment.
That means a successful linked-service test or Copy Activity doesn't necessarily prove that the Power Query authoring/runtime environment has the same private-network connectivity.
The Run on Azure IR behavior is particularly relevant. Microsoft documents this setting as selecting the Azure Integration Runtime that provides the Power Query compute environment. If your Managed VNet IR can't be selected and only AutoResolveIntegrationRuntime appears, I would investigate that before changing the SQL authentication method.
I would try one additional test: publish the Power Query and execute it from a pipeline, rather than relying only on interactive evaluation in the Power Query editor.
If:
Dataset Preview -> succeeds
Copy Activity -> succeeds
Power Query editor -> fails
Power Query pipeline -> succeeds
then this strongly points to an issue specifically with the Power Query interactive authoring/evaluation layer. The Microsoft moderator on this thread has suggested the same isolation test.
If the pipeline execution also fails, capture the Power Query Activity's complete error output and run ID. At that point, I'd raise a Microsoft support case and specifically ask them to verify connectivity from the Power Query/Data Flow Spark runtime to the Managed Private Endpoint, rather than simply validating the Azure SQL linked service.
I would not enable permanent public network access or switch to SQL authentication just to work around this. Azure SQL + MSI is documented as supported, and Microsoft has also advised against changing those settings at this stage.
Interestingly, similar cases have previously been reported where an Azure SQL linked service works correctly for Copy Activity but Power Query reports "invalid or missing credentials."
Sharing these references with you:
Microsoft – Data wrangling in Azure Data Factory
Microsoft – Power Query activity in Azure Data Factory
If you can share the pipeline execution result/error (with sensitive IDs removed), that should tell us whether we're dealing specifically with the Power Query editor or with the Power Query runtime/private networking path.
Please "Accept the Answer" if this information helped you. This will help us and others in the community.