Horton

Re: Properly pathify the union planner

Richard Guo <guofenglinux@gmail.com>

View thread View on postgr.es
From: Richard Guo <guofenglinux@gmail.com>
To: David Rowley <dgrowleyml@gmail.com>
Cc: PostgreSQL Developers <pgsql-hackers@lists.postgresql.org>
Date: 2024-03-25T06:05:07Z
Lists: pgsql-hackers 
On Mon, Mar 25, 2024 at 9:44 AM David Rowley <dgrowleyml@gmail.com> wrote:

> It seems ok that
> the ec_indexes are not set for the top-level set RelOptInfo as
> get_eclass_for_sort_expr() does not make use of ec_indexes while
> searching for an existing EquivalenceClass.  Maybe we should fix this
> varno == 0 hack and adjust get_eclass_for_sort_expr() so that it makes
> use of the ec_indexes.
>
> It's possible to see this happen with a query such as:
>
> SELECT oid FROM pg_class UNION SELECT oid FROM pg_class ORDER BY oid;


I see what you said.  Yeah, there might be some optimization
possibilities in this area.  And I agree that this should not be a
blocker in pushing this patch.


> I didn't see that as a reason not to push this patch as this occurs
> both with and without this change, so I've now pushed this patch.


Great to see this patch has been pushed!

Thanks
Richard

Commits

  1. Don't adjust ressortgroupref in generate_setop_child_grouplist()

  2. Don't zero tuple_fraction when planning UNIONs with ORDER BYs

  3. Fix assert failure when planning setop subqueries with CTEs

  4. Allow planner to use Merge Append to efficiently implement UNION

  5. Simplify PathKey checking code

  6. Clarify the 'rows' parameter in create_append_path