How to deploy Direct Lake semantic models?

[vitaly]

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?

[vitaly]

Update: Something changed within the deployment pipelines recently around Direct Lake models.

Deployment pipelines now seem to be picking up the Direct Lake semantic models & deploying them successfull across stages. This seems to apply to both custom and default Direct Lake models. Yet documentation still just says "unsupported".

This is extremely confusing! Does anyone from Microsoft (or elsewhere) have any insights into this?

[cw88]

Hi, 

i can confirm that direct lake models are now supported in my environment. It also seems to work without any problems. 🙂

[KevinChant]

I came up with one method that you can potentially use to do this, but I can't gaurantee that it will be supported or work after future updates:

https://www.kevinrchant.com/2024/03/02/initial-tests-to-copy-a-direct-lake-semantic-model-to-another... 

[cw88]

@KevinChant : Thanks for the workaround, i will try it!

@v-yetao1-msft : I wonder if microsoft has any plans to implement direct lake semantic models in deployment pipelines and/or git?

[cw88]

Does anyone know whether there is now a solution to this problem?

[annwallinger]

It is not straightforward how the models can be deployed. How can we make a template and copy it to other warehouses for example?

[vitaly]

@v-yetao1-msft 

We are looking for advice from Microsoft Direct Lake SMEs and/or the Fabric community to resolve the specific use case detailed in this post. Resending us the same link we originally supplied (which has no solutions) is not very helpful.

[v-yetao1-msft]

Hi  @vitaly 

There is no documentation available that directly describes the deployment of Direct Lake semantic models . I hope this official document is helpful.

Learn about Direct Lake in Power BI and Microsoft Fabric - Power BI | Microsoft Learn

Best Regards,
Community Support Team _ Ailsa Tao
If this post helps, then please consider Accept it as the solution to help the other members find it more quickly.