Deployment pipeline: Deployment rules for Direct Lake on OneLake semantic models

[frithjof_v]

Currently, it is not possible to use deployment rules with Direct Lake on OneLake semantic models.

The option is greyed out.

Please enable this, so we can automatically change the data sources when deploying from dev to prod.

[dom99]

Not sure how Microsoft expect anyone to use this with it being almost impossible to change the data source. Please provide a way to do it, ideally deployment rules or variable libraries, or even manually! Problem is its almost impossible to use at the moment without using horrible hacky workarounds with source control.

[rabbyn]

a colleague of mine shared this link with me as an other type of workaround to implement : https://www.datazoe.blog/post/deployment-pipelines-and-power-bi-semantic-models-with-direct-lake-on-... The blog describes a limitation/bug in Power BI semantic models using Direct Lake on OneLake when combined with Deployment Pipelines. The core problem When using Direct Lake on OneLake tables in a semantic model, you can use Deployment Pipeline rules… …but the connector does NOT support data source bindings (unlike Direct Lake on SQL). The workaround is to use parameter rules, but this introduces a bug: ❗ Bug in Power BI Desktop Parameters cannot be configured in the TMDL view of Power BI Desktop After setting a parameter: Live Edit in Power BI Desktop breaks Edit Table in web modeling is disabled You get an error when trying to edit the semantic model So once a parameter is used to externalize the OneLake URI, Power BI Desktop becomes unusable for editing that model. 🔧 Solution Proposed in the Blog Since Power BI Desktop cannot handle parameterized Direct Lake connections (due to the bug), the workaround is to edit the semantic model using external tools. ✔ Recommended workaround Use VS Code + Fabric Studio extension to manually modify the model's TMDL expression file (expressions.tmdl). Result: You regain the ability to change the data source per stage in a Deployment Pipeline — even though Power BI Desktop cannot handle this scenario.

[tsunloan]

This is a Bug pure and simple, not a feature gap. It's noting short of complete and utter incompetence that MS continue to release these features in completely unusable states and announce them as wins. Product and engineering both need to do better.

[dom99]

We had to implement a deployment pipeline from azure devops, using the python cicd library (now fully supported by microsoft). To change the data sources on release we used a powershell script to do a string replace on the expressions.tmdl file and any other sources in the tables folder. This actually works and doesnt come with any limitations (unlike the parameters method). I believe this is the only way to deploy direct lake on onelake models.

I had about 5 phone calls and an email chain about 30 messages back and forth with microsoft support. It took a number of attempts them to to understand what I was even talking about. In the end the conclusion was they didnt have any solution, and anythign suggested by them was just pointing to blog posts already linked to here, so I wouldnt recommend trying to get support for this.

Feel free to ask questions on how we managed to achieve a working pipeline. When I first began it seemed like a minefield and lots of unclear information on what to do.

[lis12345]

I also achieved that using terraform https://registry.terraform.io/providers/microsoft/fabric/latest/docs/resources/semantic_model natively without any issues.