Currently when count query has a filter and there is an index The oldest index that has the prefix of the filter is picked. Yet this is not always optimal as there can be a smaller index which would require less scanning steps within the index. Distinct already implements a similar approach. The task here is to implement a similar selection method for count.