[CF 52/5482] v20250218 - handle batch explosion in hash joins

This commit was automatically generated by a robot at cfbot.cputube.org.
It is based on patches submitted to the PostgreSQL mailing lists and
registered in the PostgreSQL Commitfest application.

This branch will be overwritten each time a new patch version is posted to
the email thread, and also periodically to check for bitrot caused by changes
on the master branch.

Commitfest entry: https://commitfest.postgresql.org/52/5482
Patch(es): https://www.postgresql.org/message-id/310eeb10-1c75-4361-966a-33fa12970e6e@vondra.me
Author(s):

Author: Commitfest Bot

patch/batch-explosion.sql

@@ -0,0 +1,46 @@
+drop table if exists t;
+
+create table t (a int, b text);
+
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+copy t from '/tmp/hash-collisions.data';
+
+insert into t select i, md5(i::text) from generate_series(1,1000000) s(i);
+
+create index on t (b);
+cluster t USING t_b_idx;
+
+vacuum analyze;
+
+set hash_mem_multiplier = 1.0;
+
+set work_mem = '2MB';
+
+explain select * from t t1 join t t2 on (t1.a = t2.a);
+explain analyze select * from t t1 join t t2 on (t1.a = t2.a);
+
+
+set work_mem = '1MB';
+
+explain select * from t t1 join t t2 on (t1.a = t2.a);
+explain analyze select * from t t1 join t t2 on (t1.a = t2.a);
+
+
+set work_mem = '512kB';
+
+explain select * from t t1 join t t2 on (t1.a = t2.a);
+explain analyze select * from t t1 join t t2 on (t1.a = t2.a);
+
+
+set work_mem = '768kB';
+
+explain select * from t t1 join t t2 on (t1.a = t2.a);
+explain analyze select * from t t1 join t t2 on (t1.a = t2.a);

patch/disabled-growth.sql

@@ -0,0 +1,41 @@
+drop table if exists t;
+
+create table t (a int, b text);
+
+copy t from '/tmp/hash-collisions.data' ;
+
+-- make it larger than 1MB
+insert into t select * from t;
+insert into t select * from t;
+insert into t select * from t;
+
+-- add a couple redundant columns
+alter table t add column r1 double precision, add column r2 double precision, add column r3 double precision, add column r4 double precision;
+
+update t set r1 = 0, r2 = 0, r3 = 0, r4 = 0;
+
+vacuum full t;
+
+insert into t select i, md5(i::text), r, r, r, r from (select i, random() as r from generate_series(1,1000000) s(i) where hashint4(i) > 1000000);
+
+vacuum analyze t;
+
+set max_parallel_workers_per_gather = 0;
+
+set work_mem = '1MB';
+
+explain analyze
+select * from t t1 join t t2 on (t1.a = t2.a)
+ where t1.r1 + 0.0 < 1.0
+   and t1.r2 + 0.0 < 1.0
+   and t1.r3 + 0.0 < 1.0
+   and t1.r4 + 0.0 < 1.0;
+
+set work_mem = '512kB';
+
+explain analyze
+select * from t t1 join t t2 on (t1.a = t2.a)
+ where t1.r1 + 0.0 < 1.0
+   and t1.r2 + 0.0 < 1.0
+   and t1.r3 + 0.0 < 1.0
+   and t1.r4 + 0.0 < 1.0;

patch/disabled-growth2.sql

@@ -0,0 +1,44 @@
+drop table if exists t;
+
+create table t (a int, b text);
+
+-- make sure the hash has first 3 bits set to 0
+insert into t select i a, md5(i::text) from generate_series(1,1000000) s(i) where hashint4(i)::bit(32) & 0x0000E000::bit(32) = 0x00000000::bit(32);
+
+insert into t select * from t;
+insert into t select * from t;
+
+alter table t add column r1 double precision, add column r2 double precision, add column r3 double precision, add column r4 double precision;
+update t set r1 = 0, r2 = 0, r3 = 0, r4 = 0;
+
+vacuum full t;
+vacuum analyze t;
+
+set max_parallel_workers_per_gather = 0;
+
+set work_mem = '2MB';
+
+explain analyze
+select * from t t1 join t t2 on (t1.a = t2.a)
+ where t1.r1 + 0.0 < 1.0
+   and t1.r2 + 0.0 < 1.0
+   and t1.r3 + 0.0 < 1.0
+   and t1.r4 + 0.0 < 1.0;
+
+set work_mem = '1MB';
+
+explain analyze
+select * from t t1 join t t2 on (t1.a = t2.a)
+ where t1.r1 + 0.0 < 1.0
+   and t1.r2 + 0.0 < 1.0
+   and t1.r3 + 0.0 < 1.0
+   and t1.r4 + 0.0 < 1.0;
+
+set work_mem = '256kB';
+
+explain analyze
+select * from t t1 join t t2 on (t1.a = t2.a)
+ where t1.r1 + 0.0 < 1.0
+   and t1.r2 + 0.0 < 1.0
+   and t1.r3 + 0.0 < 1.0
+   and t1.r4 + 0.0 < 1.0;