On Sat, 23 Mar 2019 at 04:56, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> David Rowley <david.rowley@2ndquadrant.com> writes:
> > Append has the additional
> > saving of not having to determine to perform a sort on the top row
> > from each subpath.
>
> Uh, what? sorted-Append and MergeAppend would need pre-sorts on
> exactly the same set of children.
I was talking about the binary heap code that MergeAppend uses to
decide which subplan to pull from next.
> In cases where, say, the first child requires no sort but also doesn't
> emit very many rows, while the second child requires an expensive sort,
> the planner will have a ridiculously optimistic opinion of the cost of
> fetching slightly more rows than are available from the first child.
> This might lead it to wrongly choose a merge join over a hash for example.
umm.. Yeah, that's a good point. I seemed to have failed to consider
that the fast startup plan could lower the cost of a merge join with a
limit. I agree with that concern. I also find it slightly annoying
since we already make other plan shapes that can suffer from similar
problems, e.g Index scan + filter + limit, but I agree we don't need
any more of those as they're pretty painful when they hit.
I'll change the patch around to pull out the code you've mentioned.
Thanks for spelling out your point to me.
--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services