Re: Default setting for enable_hashagg_disk

On Fri, Jul 10, 2020 at 2:50 PM Stephen Frost <sfrost(at)snowman(dot)net> wrote:
> Nothing of what you've said thus far has shown me that there were
> material bits of the discussion that I've missed.

Maybe that's just because you missed those bits too?

> No, that other people
> feel differently or have made comments supporting one thing or another
> isn't what I would consider material- I'm as allowed my opinions as much
> as others, even when I disagree with the majority (or so claimed anyhow-
> I've not gone back to count, but I don't claim it to be otherwise
> either).

You are of course entitled to your opinion.

The problem we're trying to address here is paradoxical, in a certain
sense. The HashAggs-that-spill patch is somehow not at fault on the
one hand, but on the other hand has created this urgent need to
ameliorate what is for all intents and purposes a regression.
Everything is intertwined. Yes -- this *is* weird! And, I admit that
the hash_mem proposal is unorthodox, even ugly -- in fact, I've said
words to that effect on perhaps a dozen occasions at this point. This
is also weird.

I pointed out that my hash_mem proposal was popular because it seemed
like it might save time. When I see somebody I know proposing
something strange, my first thought is "why are they proposing that?".
I might only realize some time later that there are special
circumstances that make the proposal much more reasonable than it
seemed at first (maybe even completely reasonable). There is no
inherent reason why other people supporting the proposal makes it more
valid, but in general it does suggest that special circumstances might
apply. It guides me in the direction of looking for and understanding
what they might be sooner.

--
Peter Geoghegan