Re: Performance improvement for joins where outer side is unique

David Rowley <david(dot)rowley(at)2ndquadrant(dot)com> writes:
> I've attached an updated patch which introduces JOIN_INNER_UNIQUE and
> JOIN_LEFT_UNIQUE. So unique inner joins no longer borrow JOIN_SEMI.

OK.

> In EXPLAIN, I named these new join types "Unique Inner" and "Unique
> Left".

Hm. I'm back to being unhappy about the amount of churn introduced
into the regression test outputs by this patch. I wonder whether we
could get away with only mentioning the "unique" aspect in VERBOSE
mode.

I'm also a bit suspicious of the fact that some of the plans in
aggregates.out changed from merge to hash joins; with basically
no stats at hand in those tests, that seems dubious. A quick look
at what the patch touched in costsize.c suggests that this might
be because you've effectively allowed cost_hashjoin to give a cost
discount for inner unique, but provided no similar intelligence
in cost_mergejoin.

> I know we're close to the feature freeze, so I just want to say that
> I'll be AFK starting 5:30am Friday New Zealand time (13:30 on the 8th
> New York time), until Sunday ~4pm.

Understood. I don't know if we'll get this in or not, but I'll work
on it.

regards, tom lane