Best Practice for Building and Sharing Semantic Models in Microsoft Fabric

Hi Manoshi 

 In Fabric, for best performance and cost efficiency, use directlake (anything else will require larger capacities to run).

 Add all of your calculated columns using notebooks before adding tables to the warehouse (bronze, silver, gold architecture), then build semantic models using star/snowflake schemas.

 With directlake, having multiple semantic models that serve similar purposes isn't such a problem, as the warehouse bares the brunt of the CU requirements, and the warehouse is the one source of truth so data isn't duplicated or manipulated.

 Your devs will not be able to add calculated columns in the model when using directlake, so you will need to add all columns for them, which will require extra effort on your part (and dropping tables if evolving schemas aren't enabled), but it will improve governance. Once your notebooks are templated, this isn't much work.

 On workspaces, it's going to be helpful to have them stored together with the warehouse for central administration, and in a capacity emergency, the workspace can be moved to a 'recovery' capacity. Security would need to be applied at the item level rather than workspace in this scenario.