We have a user friendly frontend providing PowerBI Embedded combined with a tabular cube coming from analysis services to enable superfast interaction with the data. This works flawlessly.
PowerBI Embedded comes with a number of limitations, e.g. custom expressions and limitations on formatting tables/matrix visualizations and so we want to provide some users with power user capabilities via PowerBI Paginated Reports.
However we run into a surprise. In PowerBI Paginated Reports we add the datasource connection to analysis services and the model is then loaded just fine. We can select all the fields but somehow we are expected to create a dataset; initially we expected this to be a virtual schema of the full model but in reality it retrieves the entire data from the cube. With millions of rows you can imagine the usability of this construction ....
Is there a way to simply live query the tabular cube with PowerBI Paginated Reports and avoid it from creating a dataset by copying the data?
In a paginated report a "dataset" is simply the result of a query that's available for binding to visuals in your report. It's a completely different concept from a Power BI "Data Set". Every time the report needs to be rendered each dataset is queried from its data source.
When retrieving data in a paginated report from AAS or Power BI be sure to use a DAX query that returns a reasonable number of rows, and don't use an MDX query (or a designer that generates an MDX query).