Re: partitioned tables referenced by FKs

On 2018-Nov-05, Corey Huinker wrote:

> This is an important and much needed feature!

I agree :-)

> Based on my extremely naive reading of this code, I have two perhaps
> equally naive questions:
>
> 1. it seems that we will continue to to per-row RI checks for inserts and
> updates. However, there already exists a bulk check in RI_Initial_Check().
> Could we modify this bulk check to do RI checks on a per-statement basis
> rather than a per-row basis?

One of the goals when implementing trigger transition tables was to
supplant the current per-row implementation of RI triggers with
per-statement. I haven't done that, but AFAIK it remains possible :-)

Changing that is definitely not a goal of this patch.

> 2. If #1 is possible, is the overhead of transitions tables too great for
> the single-row case?

Without an implementation, I can't say, but if I had to guess, I would
assume so. Or maybe there are clever optimizations for that particular
case.

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services