We have a simple design with Dev, QA and Production workspaces.
A lakehouse in Dev has a default semantic model; we are also flexible on switching to a custom semantic model. We would like to deploy the model to higher environments.
The first design attempt was to use Deployment Pipelines. This does not work. Deployment pipelines do not support Direct Lake:
The next attempt involved exploring alternatives via the XMLA endpoint.
We've enabled the XMLA read/write endpoint per Model write support with XMLA endpoint. We then tried editing the default model via the Tabular Editor. This is unsupported according to Direct Lake Datasets: How to use them with Tabular Editor:
Ok. We've switched from a default semantic model to using a custom one. Tabular Editor could then edit and save it. However, this blocks the model from opening up or remaining editable in Power BI Service:
In addition to the substantial functionality loss (i.e. disabling the Service's model UI for both editing and viewing) there are additional complexities with the Tabular Editor approach. This includes the necessity of workarounds to create the initial model.
Finally, we've checked the Fabric REST API. It lacks operations for creating or modifying semantic models.
This seems like a major gap in the SDLC around the heavily-promoted Direct Lake feature. It makes it very challenging to use in production.
We've checked the Fabric Roadmap and saw no references to announcements around improving this workflow.
How does everyone deploy Direct Lake models?
