If we want to do anything much smarter than that like trying every
combination of the GROUP BY clause, then plan times are likely going
to explode. The join order search is done based on the chosen query
pathkeys, which in many queries is the pathkeys for the GROUP BY
clause (see standard_qp_callback()). This means throughout the join
search, planner will try and form paths that provide pre-sorted input
that allows the group by to be implemented efficiently with pre-sorted
data. You might see Merge Joins rather than Hash Joins, for example.
Are there guidelines or principles you could share about writing the group by clause such that it is more efficient?
