Should dataflows live in the same workspace as semantic models?

[lava_]

We're in the process of doing implementation planning for an enterprise self-service BI initiative. We've read that it's best practice to use a dataflow inside of a semantic model, but I'm having difficulty finding an answer about where the two ojects should reside in PBI service. Should they be in the same workspace or should all dataflows be in one workspace and semantic models in another?

Also, I know dataflows that reference other dataflows should be in the same workspace as it creates a downstream orchestration of refreshes. But what about semantic models that use dataflows? If they should be in the same workspace, if the dataflow is refreshed (on a schedule), does that automatically refresh the semantic model that's using that dataflow?

Any articles/videos/insight that could be shared would be appreciated!

[aj1973]

Hi @lava_ 

Dataflows are generally speaking sources of truth and meant for reusable purposes. So it's better to keep them in a secure and separate workspace. Know also that one Dataflow can be the source for one or multiple semantic models in different Workspaces.

The refresh process is not related, for semantic models the refresh should trigger after the related Dataflow has completed its refresh.

[vanessafvg]

my personal recommendation is that things should be grouped logically to do the the refreshing efficiently but also only giving access to what is required, so that it can be managed and governed easily.  Each situation could be different.

It just depends on how far upstream the data is and if your workspace is operating like a company wide shared data store, then yes probably best to  separate out the layers logically.   

But if you have a smaller data project it might make sense to keep them all grouped together.

[lava_]

Thank you. Yes, our workspace will essentially be operating like a company wide shared data store. When you say grouped logically, do you mean, for example, grouping the dataflows that reference one another in one workspace, grouping semantic models that fall under domain A in a workspace A, and grouping semantic models that domain B in workspace B? Something like that?

I was also thinking we could use the OneLake data hub to act as the "data store" as a way to have everything in one place and promote certain dataflows and semantic models.