ALL function seems weird - HELP

Hi ChandeepChhabra 

The short answer is that if:

• A column of type date is the primary key of a relationship ( 'Dates Dim'[Group 2] in your example ); and
• A filter is applied to that column within CALCULATE

then the DAX engine treats the table containing the column of type date as though it had been marked as a date table, and automatically removes filters on that table when applying the filter on the specific date column.

See this SQLBI article.

In your example, within the Rank 3 measure

1. RANKX iterates over the rows of ALL( 'Dates Dim'[Group 2] ).
2. For each of those rows, it evaluates the measure [Sales 3].
3. In the course of evaluating [Sales 3], context transition adds the current row's value of 'Dates Dim'[Group 2] as a filter.
4. Due to the "automatic date table" behaviour described above, the automatic ALL ( 'Dates Dim' ) removes all filters on 'Dates Dim'.

So you end up with a rank for each value of 'Dates Dim'[Group 2] which is determined ignoring any other filters on 'Dates Dim', namely the filter on 'Dates Dim'[Group 1] due to grouping in the visual.

A possible fix still using RANKX could be:

Rank 3 =
VAR DatePartition =
    CALCULATETABLE (
        VALUES ( 'Dates Dim'[Group 2] ),
        REMOVEFILTERS ( 'Dates Dim'[Group 2] )
    )
RETURN
    RANKX (
        DatePartition,
        CALCULATE (
            [Sales 3],
            KEEPFILTERS ( DatePartition )
        ),
        ,
        DESC
    )

or you could write a measure using RANK specifying partitioning explicity.