Like in cost_seqscan(), I'd expect the subpath cost to be divided amongparallel workers. The patch below shows what I mean. Am I right?
-- Antonin HouskaWeb: https://www.cybertec-postgresql.com