Message-ID: <26d8828b-a939-463d-9617-514d50f27a69@postgrespro.ru>
Date: Thu, 30 Nov 2023 12:00:41 +0300
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: POC, WIP: OR-clause support for indexes
Content-Language: en-US
To: Andrei Lepikhov <a.lepikhov@postgrespro.ru>,
 Robert Haas <robertmhaas@gmail.com>, pgsql-hackers@postgresql.org,
 Alexander Korotkov <aekorotkov@gmail.com>
Cc: Peter Geoghegan <pg@bowt.ie>, "Finnerty, Jim" <jfinnert@amazon.com>,
 Marcos Pegoraro <marcos@f10.com.br>, teodor@sigaev.ru,
 Ranier Vilela <ranier.vf@gmail.com>,
 Tomas Vondra <tomas.vondra@enterprisedb.com>,
 Peter Eisentraut <peter@eisentraut.org>
References: <567ED6CA.2040504@sigaev.ru>
 <CA+Tgmoa+ywixgrgV2x4QHJgLvre31wJDsx-dC1gXDmtJUgv=Ww@mail.gmail.com>
 <CAPpHfdt58O5jzKMjkScNbGecM5JqfaMfdGLjs4rk5XisfjTNBw@mail.gmail.com>
 <16a7e3aa-24c0-4986-8820-ea2857bd7b6b@postgrespro.ru>
 <421656a9-5df7-41bb-9d58-1c1edbd5dc3e@postgrespro.ru>
 <28c727ba-5260-42a8-88b0-4ecdeeca6ef8@postgrespro.ru>
 <4c1c2c0b-802f-45ae-a7cd-007bfd8957cf@postgrespro.ru>
 <52f0d302-7963-452f-8c5a-f66d261514eb@postgrespro.ru>
 <43ad8a48-b980-410d-a83c-5beebf82a4ed@postgrespro.ru>
 <b5a5b515-6f38-431a-98b8-d30abae62fde@postgrespro.ru>
 <be8ae03d-9eb3-4adb-beb8-3b356b3c68bb@postgrespro.ru>
 <CAPpHfdu-_j6=E9CHjSGu0=SidfTqdtj7qk78dCdW-=4RFDHBDw@mail.gmail.com>
 <04ef16eb-46c1-4ce7-9f68-d1c80ef0be81@postgrespro.ru>
 <CA+TgmoZCgP6FrBQEusn4yaWm02XU8OPeoEMk91q7PRBgwaAkFw@mail.gmail.com>
 <cfdc7fef-cf0b-437b-89fa-62ffd6677f7d@postgrespro.ru>
 <59e67a40-95a8-4d74-ae4e-027ea0f59084@postgrespro.ru>
 <8ff70ada-4ec7-436a-8bf8-d1c7d8551ff1@postgrespro.ru>
From: Alena Rybakina <a.rybakina@postgrespro.ru>
In-Reply-To: <8ff70ada-4ec7-436a-8bf8-d1c7d8551ff1@postgrespro.ru>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: <https://www.postgresql.org/message-id/26d8828b-a939-463d-9617-514d50f27a69%40postgrespro.ru>
Precedence: bulk

On 30.11.2023 11:30, Andrei Lepikhov wrote:
> On 30/11/2023 15:00, Alena Rybakina wrote:
>> 2. The second patch is my patch version when I moved the OR 
>> transformation in the s index formation stage:
>>
>> So, I got the best query plan despite the possible OR to ANY 
>> transformation:
>
> If the user uses a clause like "x IN (1,2) AND y=100", it will break 
> your 'good' solution.

No, unfortunately I still see the plan with Seq scan node:

postgres=# explain analyze select * from test where x in (1,2) and y = 100;

                                                      QUERY PLAN
--------------------------------------------------------------------------------------------------------------------
  Gather  (cost=1000.00..12690.10 rows=1 width=12) (actual 
time=72.985..74.832 rows=0 loops=1)
    Workers Planned: 2
    Workers Launched: 2
    ->  Parallel Seq Scan on test  (cost=0.00..11690.00 rows=1 width=12) 
(actual time=68.573..68.573 rows=0 loops=3)
          Filter: ((x = ANY ('{1,2}'::integer[])) AND (y = '100'::double 
precision))
          Rows Removed by Filter: 333333
  Planning Time: 0.264 ms
  Execution Time: 74.887 ms

(8 rows)

> In my opinion, the general approach here is to stay with OR->ANY 
> transformation at the parsing stage and invent one more way for 
> picking an index by looking into the array and attempting to find a 
> compound index.
> Having a shorter list of expressions, where uniform ORs are grouped 
> into arrays, the optimizer will do such work with less overhead. 

Looking at the current index generation code, implementing this approach 
will require a lot of refactoring so that functions starting with 
get_indexes do not rely on the current baserestrictinfo, but use only 
the indexrestrictinfo, which is a copy of baserestrictinfo. And I think, 
potentially, there may be complexity also with the equivalences that we 
can get from OR expressions. All interesting transformations are 
available only for OR expressions, not for ANY, that is, it makes sense 
to try the last chance to find a suitable plan with the available OR 
expressions and if that plan turns out to be better, use it.


-- 
Regards,
Alena Rybakina
Postgres Professional