Horton

Re: POC, WIP: OR-clause support for indexes

Alexander Korotkov <aekorotkov@gmail.com>

View thread
From: Alexander Korotkov <aekorotkov@gmail.com>
To: Andrei Lepikhov <lepihov@gmail.com>
Cc: Alena Rybakina <a.rybakina@postgrespro.ru>, Nikolay Shaplov <dhyan@nataraj.su>, pgsql-hackers@lists.postgresql.org, jian he <jian.universality@gmail.com>, Robert Haas <robertmhaas@gmail.com>, pgsql-hackers@postgresql.org, Peter Geoghegan <pg@bowt.ie>, Marcos Pegoraro <marcos@f10.com.br>, teodor@sigaev.ru, Peter Eisentraut <peter@eisentraut.org>, Ranier Vilela <ranier.vf@gmail.com>
Date: 2024-08-23T12:58:46Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Make group_similar_or_args() reorder clause list as little as possible

  2. Allow usage of match_orclause_to_indexcol() for joins

  3. Skip not SOAP-supported indexes while transforming an OR clause into SAOP

  4. Remove the wrong assertion from match_orclause_to_indexcol()

  5. Teach bitmap path generation about transforming OR-clauses to SAOP's

  6. Transform OR-clauses to SAOP's during index matching

  7. Fix the value of or_to_any_transform_limit in postgresql.conf.sample

  8. Transform OR clauses to ANY expression

  9. MergeAttributes code deduplication

  10. SEARCH and CYCLE clauses

  11. Improve estimation of OR clauses using extended statistics.

  12. Teach btree to handle ScalarArrayOpExpr quals natively.

  13. Revise collation derivation method and expression-tree representation.

  14. Instead of trying to force WHERE clauses into CNF or DNF normal form,

Attachments

Hi!

Thank you for your feedback.

On Fri, Aug 23, 2024 at 1:23 PM Andrei Lepikhov <lepihov@gmail.com> wrote:
>
> On 21/8/2024 16:52, Alexander Korotkov wrote:
> >> /* Only operator clauses scan match */
> >> Should it be:
> >> /* Only operator clauses can match */
> >> ?
> >
> > Corrected, thanks.
> I found one more: /* Only operator clauses scan match  */ - in the
> second patch.
> Also I propose:
> - “might match to the index as whole” -> “might match the index as a whole“
> - Group similar OR-arguments intro dedicated RestrictInfos -> ‘into’

Fixed.

> >> The second one:
> >> When creating IndexClause, we assign the original and derived clauses to
> >> the new, containing transformed array. But logically, we should set the
> >> clause with a list of ORs as the original. Why did you do so?
> >
> > I actually didn't notice that.  Corrected to set the OR clause as the
> > original.  That change turned recheck to use original OR clauses,
> > probably better this way.  Also, that change spotted misuse of
> > RestrictInfo.clause and RestrictInfo.orclause in the second patch.
> > Corrected this too.
> New findings:
> =============
>
> 1)
> if (list_length(clause->args) != 2)
>         return NULL;
> I guess, above we can 'continue' the process.
>
> 2) Calling the match_index_to_operand in three nested cycles you could
> break the search on first successful match, couldn't it? At least, the
> comment "just stop with first matching index key" say so.

Fixed.

> 3) I finally found the limit of this feature: the case of two partial
> indexes on the same column. Look at the example below:
>
> SET enable_indexscan = 'off';
> SET enable_seqscan = 'off';
> DROP TABLE IF EXISTS test CASCADE;
> CREATE TABLE test (x int);
> INSERT INTO test (x) SELECT * FROM generate_series(1,100);
> CREATE INDEX ON test (x) WHERE x < 80;
> CREATE INDEX ON test (x) WHERE x > 80;
> VACUUM ANALYZE test;
> EXPLAIN (ANALYZE, COSTS OFF, TIMING OFF)
> SELECT * FROM test WHERE x=1 OR x = 79;
> EXPLAIN (ANALYZE, COSTS OFF, TIMING OFF)
> SELECT * FROM test WHERE x=91 OR x = 81;
> EXPLAIN (ANALYZE, COSTS OFF, TIMING OFF)
> SELECT * FROM test WHERE x=1 OR x = 81 OR x = 83;
>
> The last query doesn't group clauses into two indexes. The reason is in
> match_index_to_operand which classifies all 'x=' to one class. I'm not
> sure because of overhead, but it may be resolved by using
> predicate_implied_by to partial indexes.

Yes, this is the conscious limitation of my patch: to consider similar
OR arguments altogether and one-by-one, not in arbitrary groups.  The
important thing here is that we still generating BitmapOR patch as we
do without the patch.  So, there is no regression.  I would leave this
as is to not make this feature too complicated. This could be improved
in future though.

------
Regards,
Alexander Korotkov
Supabase