Alexander Kuzmenkov

Make cost-based decision about partial aggregation paths

1

Currently we just remove all existing paths, but after that we add a hash aggregation path (ts_plan_add_hashagg()) which we can prefer based on the cost, which leads to weird behavior.

Choose cheapest aggregated path as base for chunk-wise aggregation

1

Currently we use the cheapest path before final aggregation for this (i.e. Append over individual chunks), but it does not consider the cost of a Sort that might be required...