Thomas Kellerer <spam_eater(at)gmx(dot)net> writes:
> while playing around with the parallel aggregates and seq scan in 9.6beta I noticed that Postgres will stop using parallel plans when cpu_tuple_cost is set to a very small number.
If you don't reduce the parallel-plan cost factors proportionally,
it's not very surprising that reducing that would tend to bias the
planner away from using parallel plans. See parallel_setup_cost and
parallel_tuple_cost.
regards, tom lane