Forum Discussion
DeltaLake.Table fails on Usage Metrics semantic model because _delta_log contains URL-encoded parque
- 3 months ago
I confirmed all above , as a work around I create this code
Latest _delta_log entry:"path":"Users-28f8e96d-15d0-4ef8-b_%28215434%29.00000000000000000170.parquet"Physical file:Users-28f8e96d-15d0-4ef8-b_(215434).00000000000000000170.parquetDifference:Delta log contains URL-encoded characters:%28 = (%29 = )The AzureStorage.DataLake() file listing returns decoded filenames while DeltaLake.Table() appears to expect the encoded names contained in the Delta log.Additional evidence:DeltaLake.Table(Source)fails with:Expression.Error: The key didn't match any rows in the table.DetailsReason = Expression.ErrorErrorCode = 10061Key = [Name = "Reports-045af53d-15a1-41ad-a8_%28268%29.00000000000000000174.parquet"]Table = #table({"Content", "Name", "Extension", "Date accessed", "Date modified", "Date created", "Attributes", "Folder Path"}, {})Applying:FixedNames =Table.TransformColumns(Source,{{"Name",each Uri.EscapeDataString(_),type text}})and then:DeltaLake.Table(FixedNames)successfully returns the table data.This strongly indicates that the failure is caused by a filename encoding mismatch between the Delta log metadata and the filenames returned by AzureStorage.DataLake().
This looks less like a query issue and more like a potential regression in how DeltaLake.Table is resolving file names.
Based on what you've described:
AzureStorage.DataLake can successfully enumerate the folders, parquet files, and _delta_log.
The Delta log references URL-encoded file names (%28 and %29), while the storage connector returns decoded names (( and )).
The failure occurs consistently across multiple Usage Metrics tables, suggesting it's not isolated to a single table.
If this started without any changes on your side and previously worked, it does point toward a recent change in either the Fabric runtime or the Delta connector.
A few things worth checking:
Confirm the issue reproduces in a new query and in a different workspace to rule out corruption.
Test another Delta table that doesn't contain parentheses or other special characters in the parquet file names.
Verify whether the same behavior occurs from a Notebook or Spark, which can help determine whether the issue is specific to Power Query's DeltaLake.Table implementation.
Since this appears to affect multiple Usage Metrics semantic model tables and the storage itself is accessible, I'd recommend opening a Microsoft support ticket with a minimal repro. The mismatch between URL-encoded paths in _delta_log and decoded paths returned by AzureStorage.DataLake could indicate a regression that Microsoft will need to investigate.