Re: pg9.6 segfault using simple query (related to use fk for join estimates)

On 5 May 2016 at 16:04, David Rowley <david.rowley@2ndquadrant.com> wrote:
> I've started making some improvements to this, but need to talk to
> Tomas. It's currently in the middle of his night, but will try to
> catch him in his morning to discuss this with him.

Ok, so I spoke to Tomas about this briefly, and he's asked me to send
in this patch. He didn't get time to look over it due to some other
commitments he has today.

I do personally feel that if the attached is not good enough, or not
very close to good enough then probably the best course of action is
to revert the whole thing. I'd much rather have this out than to feel
some responsibility for someone's performance problems with this
feature. But I also do feel that the patch allows a solution to a big
problem that we currently have with PostgreSQL, which there's any easy
workarounds for... that's multi-column join estimates.

In the attached I've left the GUC remaining. The reason for the GUC is
for testing purposes and it should be removed before release. It
should likely be documented though, even if we're planning to remove
it later. I've not gotten around to that in this patch though.


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