When a Dataflow Gen2 refresh fails, the red status tells you something went wrong, but not always at which step. Since September 2026, downloading detailed diagnostic logs is generally available in Microsoft Fabric. The files record the operations of the Power Query Mashup engine and help you rebuild what happened during a run.
Where to download the logs
In the workspace, open the Dataflow Gen2 and go to its recent runs. Select the run you want to investigate and, on the details screen, choose the option to download detailed logs. Fabric generates a ZIP file with the records of that run.
The logs become available a few minutes after the refresh ends and can be downloaded for up to 28 days. You need at least the Viewer role in the workspace. For dataflows that use an on-premises gateway, the administrator also has to enable diagnostics consent on the gateway.
A path to the root cause
Start by keeping the original ZIP untouched. Then extract the files without changing the folder structure and gather the .log files from the mashuplogs folder, including the directories for each engine instance. Every line is an independent JSON record.
To rebuild the run, sort the records by start time (Start) and group the operations by ActivityId. Also check the correlation file (CorrelationId) and confirm that the refresh and session IDs match the run you selected. This sequence helps you tell a slow step apart from a one-off failure, even when several instances work in parallel.
Look at duration, the CPU fields and the exception details. A long operation can point to slow reads or writes; an exception can point to a type conversion, credentials, an unavailable source or another Power Query step. Compare the sequence with the run history and the loaded tables to find where the behavior changed.
Detailed logs versus workspace monitoring
Downloaded logs are best for examining one specific run. To follow many runs over time, workspace monitoring gives you another view: it records refresh and publish events in a table you can query with KQL. The team can compare duration, status and trends across dataflows and set alerts for recurring failures.
Use the run history and the monitoring hub for daily triage, detailed logs to investigate a single run, and workspace monitoring for trends and alerts. The tools complement each other: a diagnostic file explains one run; a series of events helps you spot regressions across the whole estate.
Treat the files as sensitive information
The records can contain query expressions, endpoint information and error details. Before sending a ZIP to anyone, review its contents and remove sensitive data. Keep an unmodified original copy to preserve the context in case you need to open a Microsoft support ticket.
Downloading diagnostics does not fix the query for you, but it shortens the path between "the refresh failed" and the operation that needs attention. For teams that maintain several Dataflows Gen2, combining per-run logs with workspace monitoring makes investigations traceable and helps catch problems before they become routine.