Power BI
Pipeline execution
RLS with empty roles and simulation steps
In Open Report and Simulation Journey steps, support for dataset chaining was introduced in version 1.18.0. This support requires us to perform additional checks on the dataset backing the report.
Since February 2025, Microsoft has prevented us from computing the roles to use when the report uses RLS. In version 1.37.5, we removed this computation, and now every report can be opened or simulated. However, no RLS is applied to those reports because we cannot compute the effective identity. As a result, we see what a report designer can see, not what a consumer can see. We are planning to review how we simulate effective identities in these cases in the future.
Support for specific object types
This applies to the following steps:
- Regression Testing steps
- Open Report and Simulation steps
Power BI exposes four object types to users:
- Reports
- Paginated Reports
- Dashboards (Standard)
- Dashboards (Metrics), previously named Scorecards
The Microsoft API does not allow us to filter only Reports when listing objects. As a result, these objects may appear in the pipeline designer, but when the steps execute, they may end with a FAILED status. Only classic Dashboards will not appear.
| Object type | Step type | Error in pipeline execution logs |
|---|---|---|
| Paginated report | Regression testing | No longer proposed to user |
| Paginated report | Simulation steps | No longer proposed to user |
| Scorecard | Simulation steps | “Report opening failed” {} |
| Scorecard | Regression testing | It will work |
Dataset Values and large datasets
When you configure a Dataset Values step, if the table you are analyzing contains a large number of records, only the first records are tested.
More precisely, if you are testing a set of more than 100,000 distinct values, we will test only the first 100,000. We are working on a way to retrieve the full set.
Duplicated Rows and large numbers of duplicates
When using the Duplicated Rows step, if the step finds more than 10,000 duplicated tuples (that is, each of those 10,000 tuples appears at least twice in the semantic model), the step ends in error. We are working on a way to support unlimited duplicate detection.
Tableau
Pipeline execution
Promoting a Workbook with unpublished views
Promoting a workbook that has unpublished views causes the promoted workbook in the target project to have all its views published, greatly increasing the number of available views in the workbook.
This is a known limitation of the Tableau APIs. The issue also affects Tableau revision history and Tableau web upload: if you download a workbook revision that had unpublished views and then publish this .twb/.twbx file, all views in the workbook will be published.
Version management
BI Platform deletion
When a BI Platform is deleted, contexts are not deleted in cascade. The impact on the Version Browser is:
- BI Contexts with a deleted BI Platform are still visible.
- Choosing one of these redirects the user to the site tree with an error indicating that we are unable to connect to Tableau.
This behavior is intended to prevent pipeline errors: a context linked to a deleted BI Platform can have its BI Platform updated afterward.
Restoring a workbook version with unpublished views
Restoring a workbook version that had unpublished views when the version was created causes the restored workbook to have all its views published, greatly increasing the number of available views in the workbook.
This is a known limitation of the Tableau APIs. The issue also affects Tableau revision history and Tableau web upload: if you download a workbook revision that had unpublished views and then publish this .twb/.twbx file, all views in the workbook will be published.