Re: How to make partitioning scale better for larger numbers of partitions

On 13 July 2018 at 18:53, Tsunakawa, Takayuki
<tsunakawa(dot)takay(at)jp(dot)fujitsu(dot)com> wrote:
> By the way, what do you think is the "ideal and should-be-feasible" goal and the "realistic" goal we can reach in the near future (e.g. PG 12)? Say,

Depends. Patched don't move that fast without review and nothing gets
committed without a committer.

> * Planning and execution time is O(log n), where n is the number of partitions
> * Planning time is O(log n), execution time is O(1)
> * Planning and execution time is O(1), where n is the number of partitions

It's going to depend on how many partitions are pruned. We still need
to generate paths for all non-pruned partitions which is going to be
slow when there are many partitions.

I think we can get pretty close to the non-partitioned planning
performance with SELECT/UPDATE/DELETE when all but 1 partition
survives pruning. There are always going to be some additional
overhead we can't get rid of, but hopefully, those will be small.

Please feel free to review what I have in the July 'fest.

--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services