Modeling Distinct as aggregation limits the opportunity of optimizations. From query processing perspective, distinct works closer to selection, but with dedup. More context and performance numbers can be found here: https://github.com/apache/incubator-pinot/pull/6285