Horton

Re: Add proper planner support for ORDER BY / DISTINCT aggregates

David Rowley <dgrowleyml@gmail.com>

View thread View on postgr.es
From: David Rowley <dgrowleyml@gmail.com>
To: Richard Guo <guofenglinux@gmail.com>
Cc: Dean Rasheed <dean.a.rasheed@gmail.com>, Ronan Dunklau <ronan.dunklau@aiven.io>, Justin Pryzby <pryzby@telsasoft.com>, Pavel Luzanov <p.luzanov@postgrespro.ru>, Tom Lane <tgl@sss.pgh.pa.us>, pgsql-hackers@lists.postgresql.org, Ranier Vilela <ranier.vf@gmail.com>
Date: 2023-01-11T04:12:45Z
Lists: pgsql-hackers 
On Wed, 11 Jan 2023 at 15:46, Richard Guo <guofenglinux@gmail.com> wrote:
> However the scan/join plan's
> tlist does not contain random(), which I think we need to fix.

I was wondering if that's true and considered that we don't want to
evaluate random() for the sort then again when doing the aggregate
transitions, but I see that does not really work before 1349d279, per:

postgres=# set enable_presorted_aggregate=0;
SET
postgres=# select string_agg(random()::text, ',' order by random())
from generate_series(1,3);
                        string_agg
-----------------------------------------------------------
 0.8659110018246505,0.15612649559563474,0.2022878955613403
(1 row)

I'd have expected those random numbers to be concatenated in ascending order.

Running: select random() from generate_Series(1,3) order by random();
gives me the results in the order I'd have expected.

I think whatever the fix is here, we should likely ensure that the
results are consistent regardless of which Aggrefs are the presorted
ones.  Perhaps the easiest way to do that, and to ensure we call the
volatile functions are called the same number of times would just be
to never choose Aggrefs with volatile functions when doing
make_pathkeys_for_groupagg().

David

Commits

  1. Don't presort ORDER BY/DISTINCT Aggrefs with volatile functions

  2. Add enable_presorted_aggregate GUC

  3. Remove pessimistic cost penalization from Incremental Sort

  4. Fix hypothetical problem passing the wrong GROUP BY pathkeys

  5. Remove unused fields from ExprEvalStep

  6. Improve performance of ORDER BY / DISTINCT aggregates

  7. Refactor function parse_subscription_options.