Dataflow incremental refresh datetime power queries do not match gui paramaters

[goodbyetonto]

I am not certain yet whether this is an issue or my ignorance, however I just began to use incremental refresh on a new dataflow (SQL Server data source) and noticed the following: 

The 'RangeStart' query that is generated from activating incremental refresh is defaulting to a datetime that does not match what is set from the GUI. 

GUI params: 

However, when I look at the Power Query generated from this for 'RangeStart', it is:

#datetime(0001,01,01,00,00,00) meta [IsParameterQuery=true, List={}, Type="DateTime", IsParameterQueryRequired=true, Description="autogenerated_for_incremental_refresh"] (1/1/1, 12:00:00 AM)

 

 Seen just below, you can see there is no filtering happening on the fiscal_calendar_date, who's oldest year should be 2021, not 2017.

I have changed the power query for 'RangeStart' to be the correct date/time stamp, and will reload the table from within power query, and it filters just fine. The moment I save and close and refresh the dataflow, the 'RangeStart' query reverts back to the incorrect date/time stamp and I lose the filtering schema that I was wanting to apply through incremental refresh. Here is the table query for your reference. You can see that the join is done prior to the incremental refresh filtering, which part of me wonders if my applied steps were done in the wrong order perhaps?

Does anyone know whether or not this is a bug and if not, could you please provide your insights here? Thank you so much!

[lbendlin · Accepted solution]

The parameter values for RangeStart and RangeEnd in Power Query are not relevant for the actual partitions.  They only help you the developer to limit the development data.

In fact, with a dataflow you have no control over these parameters or where they are placed in the Power Query (for some unexplicable reason they are always placed at the end which makes no sense whatsoever - the filter should apply as soon as possible).

Ignore them and check the refresh details of the dataflow to make sure you see the generated partitions.

Ideally you would do an incremental refresh on a semantic model, not a dataflow.