Unexpected behavior for auto relationships in Microsoft Fabric

I have not written a blog post about on Microsoft Fabric before but this has really caught my eye.

When creating a semantic model in Microsoft Fabric on a Lakehouse or a Data Warehouse we have the model view similar to Power BI Desktop.

The relationships there behave completely different though and I want to show how this currently (as of 10th April 2024) behaves in comparison with Power BI Desktop.

Default relationships wrong in Microsoft Fabric

Here are the main differences I have found:

1. When you drag a relationship from DIM to FACT in Power BI Desktop it will check the cardinality. In Fabric it is not doing this currently.

2. The filter direction is inverted in the sense that it will always create the filter direction from FACT -> DIM if you drag it from DIM to FACT.

Microsoft support has confirmed that somehow this is by design even though I have to say I do not really understand why it makes sense like that.

Here is an excerpt from the documentation:Validation is limited for Direct Lake models. User selections are assumed correct and no queries will validate cardinality and cross filter selections for relationships, or for the selected date column in a date table.

Summary:

In short, be very careful when creating relationships in Fabric and always double check that the cardinality/filter direction is the way you intend it to and do not expect it to be correct by default.