From 1cbf72ab58706dd194eaee2a620b27c8961f2001 Mon Sep 17 00:00:00 2001 From: Tomas Vondra Date: Mon, 27 Jul 2026 15:27:39 +0200 Subject: [PATCH v8 16/20] Add basic regression tests for filter pushdown Adds three test suites specifically for Bloom filter pushdown, for basic schemas - two relations, starjoin and snowflake. The filter pushdown affects plans in a number of existing test suites, in an accidental way. These tests are focused on filter pushdown, generating plans with known plan. --- src/test/regress/expected/hashjoin_bloom.out | 63 +++ .../expected/hashjoin_bloom_snowflake.out | 386 +++++++++++++++++ .../regress/expected/hashjoin_bloom_star.out | 408 ++++++++++++++++++ src/test/regress/parallel_schedule | 3 + src/test/regress/sql/hashjoin_bloom.sql | 40 ++ .../regress/sql/hashjoin_bloom_snowflake.sql | 139 ++++++ src/test/regress/sql/hashjoin_bloom_star.sql | 147 +++++++ 7 files changed, 1186 insertions(+) create mode 100644 src/test/regress/expected/hashjoin_bloom.out create mode 100644 src/test/regress/expected/hashjoin_bloom_snowflake.out create mode 100644 src/test/regress/expected/hashjoin_bloom_star.out create mode 100644 src/test/regress/sql/hashjoin_bloom.sql create mode 100644 src/test/regress/sql/hashjoin_bloom_snowflake.sql create mode 100644 src/test/regress/sql/hashjoin_bloom_star.sql diff --git a/src/test/regress/expected/hashjoin_bloom.out b/src/test/regress/expected/hashjoin_bloom.out new file mode 100644 index 00000000000..fd6d895f4cc --- /dev/null +++ b/src/test/regress/expected/hashjoin_bloom.out @@ -0,0 +1,63 @@ +-- tests to validate bloom filter pushdown +-- no parallel query support for now +SET max_parallel_workers_per_gather = 0; +-- a couple very simple join queries +CREATE TABLE bloom_simple_dim (id int, r real); +CREATE TABLE bloom_simple_fact (id int, padding text); +INSERT INTO bloom_simple_dim SELECT i, random() FROM generate_series(1,1000) s(i); +INSERT INTO bloom_simple_fact +SELECT + 1 + mod(i, 1000), + md5(i::text) +FROM generate_series(1,100000) s(i); +VACUUM ANALYZE; +-- simple query, no join is selective enough for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_simple_fact f +JOIN bloom_simple_dim d ON (f.id = d.id); + QUERY PLAN +-------------------------------------------- + Hash Join + Hash Cond: (f.id = d.id) + -> Seq Scan on bloom_simple_fact f + -> Hash + -> Seq Scan on bloom_simple_dim d +(5 rows) + +-- join is 75% selective (not enough for a filter to be created) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_simple_fact f +JOIN bloom_simple_dim d ON (f.id = d.id) +WHERE d.r < 0.75; + QUERY PLAN +------------------------------------------------------ + Hash Join + Hash Cond: (f.id = d.id) + -> Seq Scan on bloom_simple_fact f + -> Hash + -> Seq Scan on bloom_simple_dim d + Filter: (r < '0.75'::double precision) +(6 rows) + +-- join is 50% selective (enough for a filter to be pushed down) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_simple_fact f +JOIN bloom_simple_dim d ON (f.id = d.id) +WHERE d.r < 0.5; + QUERY PLAN +----------------------------------------------------- + Hash Join + Hash Cond: (f.id = d.id) + -> Seq Scan on bloom_simple_fact f + Bloom Filter 1: keys=(id) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_simple_dim d + Filter: (r < '0.5'::double precision) +(8 rows) + +DROP TABLE bloom_simple_dim; +DROP TABLE bloom_simple_fact; diff --git a/src/test/regress/expected/hashjoin_bloom_snowflake.out b/src/test/regress/expected/hashjoin_bloom_snowflake.out new file mode 100644 index 00000000000..dbe7663614f --- /dev/null +++ b/src/test/regress/expected/hashjoin_bloom_snowflake.out @@ -0,0 +1,386 @@ +-- tests to validate bloom filter pushdown +-- no parallel query support for now +SET max_parallel_workers_per_gather = 0; +-- simple snowflake queries (fact + up to 7 dimensions) +CREATE TABLE bloom_snowflake_dim_1_1 (id int, r real); +CREATE TABLE bloom_snowflake_dim_1_2 (id int, r real); +CREATE TABLE bloom_snowflake_dim_1 (id int, id11 int, id12 int, r real); +CREATE TABLE bloom_snowflake_dim_2_1 (id int, r real); +CREATE TABLE bloom_snowflake_dim_2_2 (id int, r real); +CREATE TABLE bloom_snowflake_dim_2 (id int, id21 int, id22 int, r real); +CREATE TABLE bloom_snowflake_fact (id1 int, id2 int, padding text); +INSERT INTO bloom_snowflake_dim_1_1 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_1_2 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_1 SELECT i, 1 + mod(i, 10000), 1 + mod(i, 10000), random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_2_1 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_2_2 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_2 SELECT i, 1 + mod(i, 10000), 1 + mod(i, 10000), random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_fact +SELECT + 1 + mod(i, 10000), + 1 + mod(i, 10000), + md5(i::text) +FROM generate_series(1,1000000) s(i); +VACUUM ANALYZE; +-- simple query, no join is selective enough for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id); + QUERY PLAN +----------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id1 = d1.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Seq Scan on bloom_snowflake_fact f + -> Hash + -> Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 + -> Hash + -> Hash Join + Hash Cond: (d1.id12 = d12.id) + -> Hash Join + Hash Cond: (d1.id11 = d11.id) + -> Seq Scan on bloom_snowflake_dim_1 d1 + -> Hash + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + -> Hash + -> Seq Scan on bloom_snowflake_dim_1_2 d12 +(25 rows) + +-- join is 75% selective (not enough for a filter to be created) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d1.r < 0.75; + QUERY PLAN +------------------------------------------------------------------------------------ + Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_snowflake_fact f + -> Hash + -> Hash Join + Hash Cond: (d12.id = d1.id12) + -> Seq Scan on bloom_snowflake_dim_1_2 d12 + -> Hash + -> Hash Join + Hash Cond: (d11.id = d1.id11) + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + -> Hash + -> Seq Scan on bloom_snowflake_dim_1 d1 + Filter: (r < '0.75'::double precision) + -> Hash + -> Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 +(26 rows) + +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.75; + QUERY PLAN +------------------------------------------------------------------------------------ + Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_snowflake_fact f + -> Hash + -> Hash Join + Hash Cond: (d12.id = d1.id12) + -> Seq Scan on bloom_snowflake_dim_1_2 d12 + -> Hash + -> Hash Join + Hash Cond: (d1.id11 = d11.id) + -> Seq Scan on bloom_snowflake_dim_1 d1 + -> Hash + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + Filter: (r < '0.75'::double precision) + -> Hash + -> Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 +(26 rows) + +-- join is 50% selective (good enough for pushdown) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d1.r < 0.5; + QUERY PLAN +----------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id1 = d1.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Seq Scan on bloom_snowflake_fact f + Bloom Filter 3: keys=(id1) + -> Hash + -> Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 + -> Hash + Bloom Filter 3 + -> Hash Join + Hash Cond: (d12.id = d1.id12) + -> Seq Scan on bloom_snowflake_dim_1_2 d12 + Bloom Filter 2: keys=(id) + -> Hash + Bloom Filter 2 + -> Hash Join + Hash Cond: (d11.id = d1.id11) + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + Bloom Filter 1: keys=(id) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_snowflake_dim_1 d1 + Filter: (r < '0.5'::double precision) +(32 rows) + +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.5; + QUERY PLAN +----------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id1 = d1.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Seq Scan on bloom_snowflake_fact f + Bloom Filter 3: keys=(id1) + -> Hash + -> Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 + -> Hash + Bloom Filter 3 + -> Hash Join + Hash Cond: (d12.id = d1.id12) + -> Seq Scan on bloom_snowflake_dim_1_2 d12 + Bloom Filter 2: keys=(id) + -> Hash + Bloom Filter 2 + -> Hash Join + Hash Cond: (d1.id11 = d11.id) + -> Seq Scan on bloom_snowflake_dim_1 d1 + Bloom Filter 1: keys=(id11) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + Filter: (r < '0.5'::double precision) +(32 rows) + +-- increase the accepted build size (includes the fact) +SET bloom_filter_pushdown_max_build_relids = 4; +-- join is 50% selective (good enough for pushdown) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d1.r < 0.5; + QUERY PLAN +----------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id1 = d1.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Seq Scan on bloom_snowflake_fact f + Bloom Filter 3: keys=(id1) + -> Hash + -> Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 + -> Hash + Bloom Filter 3 + -> Hash Join + Hash Cond: (d12.id = d1.id12) + -> Seq Scan on bloom_snowflake_dim_1_2 d12 + Bloom Filter 2: keys=(id) + -> Hash + Bloom Filter 2 + -> Hash Join + Hash Cond: (d11.id = d1.id11) + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + Bloom Filter 1: keys=(id) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_snowflake_dim_1 d1 + Filter: (r < '0.5'::double precision) +(32 rows) + +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.5; + QUERY PLAN +----------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id1 = d1.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Seq Scan on bloom_snowflake_fact f + Bloom Filter 3: keys=(id1) + -> Hash + -> Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 + -> Hash + Bloom Filter 3 + -> Hash Join + Hash Cond: (d12.id = d1.id12) + -> Seq Scan on bloom_snowflake_dim_1_2 d12 + Bloom Filter 2: keys=(id) + -> Hash + Bloom Filter 2 + -> Hash Join + Hash Cond: (d1.id11 = d11.id) + -> Seq Scan on bloom_snowflake_dim_1 d1 + Bloom Filter 1: keys=(id11) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + Filter: (r < '0.5'::double precision) +(32 rows) + +-- needed to stabilize the join order +SET join_collapse_limit = 1; +-- two joins with 75% selectivity (not enough for pushdown individually, +-- but good enough when combined) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.75 AND d12.r < 0.75; + QUERY PLAN +-------------------------------------------------------------------------------- + Hash Join + Hash Cond: (d2.id22 = d22.id) + -> Hash Join + Hash Cond: (d2.id21 = d21.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (d1.id12 = d12.id) + -> Hash Join + Hash Cond: (d1.id11 = d11.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_snowflake_fact f + -> Hash + -> Seq Scan on bloom_snowflake_dim_1 d1 + -> Hash + -> Seq Scan on bloom_snowflake_dim_1_1 d11 + Filter: (r < '0.75'::double precision) + -> Hash + -> Seq Scan on bloom_snowflake_dim_1_2 d12 + Filter: (r < '0.75'::double precision) + -> Hash + -> Seq Scan on bloom_snowflake_dim_2 d2 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_1 d21 + -> Hash + -> Seq Scan on bloom_snowflake_dim_2_2 d22 +(27 rows) + +RESET join_collapse_limit; +RESET bloom_filter_pushdown_max_build_relids; +DROP TABLE bloom_snowflake_dim_1_1; +DROP TABLE bloom_snowflake_dim_1_2; +DROP TABLE bloom_snowflake_dim_1; +DROP TABLE bloom_snowflake_dim_2_1; +DROP TABLE bloom_snowflake_dim_2_2; +DROP TABLE bloom_snowflake_dim_2; +DROP TABLE bloom_snowflake_fact; diff --git a/src/test/regress/expected/hashjoin_bloom_star.out b/src/test/regress/expected/hashjoin_bloom_star.out new file mode 100644 index 00000000000..7e6a60d3266 --- /dev/null +++ b/src/test/regress/expected/hashjoin_bloom_star.out @@ -0,0 +1,408 @@ +-- tests to validate bloom filter pushdown +-- no parallel query support for now +SET max_parallel_workers_per_gather = 0; +-- needed, because the filter pushdown eliminates one of the inputs for +-- choosing join order (reduction of result size) +SET join_collapse_limit = 1; +-- simple starjoin queries (fact + up to 7 dimensions) +CREATE TABLE bloom_star_dim_1 (id int, r real); +CREATE TABLE bloom_star_dim_2 (id int, r real); +CREATE TABLE bloom_star_dim_3 (id int, r real); +CREATE TABLE bloom_star_dim_4 (id int, r real); +CREATE TABLE bloom_star_dim_5 (id int, r real); +CREATE TABLE bloom_star_dim_6 (id int, r real); +CREATE TABLE bloom_star_dim_7 (id int, r real); +CREATE TABLE bloom_star_fact (id1 int, id2 int, id3 int, id4 int, id5 int, id6 int, id7 int, padding text); +INSERT INTO bloom_star_dim_1 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_2 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_3 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_4 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_5 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_6 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_7 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_fact +SELECT + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + md5(i::text) +FROM generate_series(1,1000000) s(i); +VACUUM ANALYZE; +-- simple query, no join is selective enough for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id); + QUERY PLAN +--------------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id7 = d7.id) + -> Hash Join + Hash Cond: (f.id6 = d6.id) + -> Hash Join + Hash Cond: (f.id5 = d5.id) + -> Hash Join + Hash Cond: (f.id4 = d4.id) + -> Hash Join + Hash Cond: (f.id3 = d3.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_star_fact f + -> Hash + -> Seq Scan on bloom_star_dim_1 d1 + -> Hash + -> Seq Scan on bloom_star_dim_2 d2 + -> Hash + -> Seq Scan on bloom_star_dim_3 d3 + -> Hash + -> Seq Scan on bloom_star_dim_4 d4 + -> Hash + -> Seq Scan on bloom_star_dim_5 d5 + -> Hash + -> Seq Scan on bloom_star_dim_6 d6 + -> Hash + -> Seq Scan on bloom_star_dim_7 d7 +(29 rows) + +-- last join is 75% selective (not enough for a filter to be created) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d7.r < 0.75; + QUERY PLAN +--------------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id7 = d7.id) + -> Hash Join + Hash Cond: (f.id6 = d6.id) + -> Hash Join + Hash Cond: (f.id5 = d5.id) + -> Hash Join + Hash Cond: (f.id4 = d4.id) + -> Hash Join + Hash Cond: (f.id3 = d3.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_star_fact f + -> Hash + -> Seq Scan on bloom_star_dim_1 d1 + -> Hash + -> Seq Scan on bloom_star_dim_2 d2 + -> Hash + -> Seq Scan on bloom_star_dim_3 d3 + -> Hash + -> Seq Scan on bloom_star_dim_4 d4 + -> Hash + -> Seq Scan on bloom_star_dim_5 d5 + -> Hash + -> Seq Scan on bloom_star_dim_6 d6 + -> Hash + -> Seq Scan on bloom_star_dim_7 d7 + Filter: (r < '0.75'::double precision) +(30 rows) + +-- last join is 50% selective (good enough for a filter push down) +-- the join will be executed done last +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d7.r < 0.5; + QUERY PLAN +--------------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id7 = d7.id) + -> Hash Join + Hash Cond: (f.id6 = d6.id) + -> Hash Join + Hash Cond: (f.id5 = d5.id) + -> Hash Join + Hash Cond: (f.id4 = d4.id) + -> Hash Join + Hash Cond: (f.id3 = d3.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_star_fact f + Bloom Filter 1: keys=(id7) + -> Hash + -> Seq Scan on bloom_star_dim_1 d1 + -> Hash + -> Seq Scan on bloom_star_dim_2 d2 + -> Hash + -> Seq Scan on bloom_star_dim_3 d3 + -> Hash + -> Seq Scan on bloom_star_dim_4 d4 + -> Hash + -> Seq Scan on bloom_star_dim_5 d5 + -> Hash + -> Seq Scan on bloom_star_dim_6 d6 + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_star_dim_7 d7 + Filter: (r < '0.5'::double precision) +(32 rows) + +-- first join is 50% selective (good enough for a filter push down) +-- the join will be executed done last +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.5; + QUERY PLAN +----------------------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id7 = d7.id) + -> Hash Join + Hash Cond: (f.id6 = d6.id) + -> Hash Join + Hash Cond: (f.id5 = d5.id) + -> Hash Join + Hash Cond: (f.id4 = d4.id) + -> Hash Join + Hash Cond: (f.id3 = d3.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_star_fact f + Bloom Filter 1: keys=(id1) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_star_dim_1 d1 + Filter: (r < '0.5'::double precision) + -> Hash + -> Seq Scan on bloom_star_dim_2 d2 + -> Hash + -> Seq Scan on bloom_star_dim_3 d3 + -> Hash + -> Seq Scan on bloom_star_dim_4 d4 + -> Hash + -> Seq Scan on bloom_star_dim_5 d5 + -> Hash + -> Seq Scan on bloom_star_dim_6 d6 + -> Hash + -> Seq Scan on bloom_star_dim_7 d7 +(32 rows) + +-- two joins 50% selective +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.4 AND d7.r < 0.5; + QUERY PLAN +----------------------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id7 = d7.id) + -> Hash Join + Hash Cond: (f.id6 = d6.id) + -> Hash Join + Hash Cond: (f.id5 = d5.id) + -> Hash Join + Hash Cond: (f.id4 = d4.id) + -> Hash Join + Hash Cond: (f.id3 = d3.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_star_fact f + Bloom Filter 1: keys=(id1) + Bloom Filter 2: keys=(id7) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_star_dim_1 d1 + Filter: (r < '0.4'::double precision) + -> Hash + -> Seq Scan on bloom_star_dim_2 d2 + -> Hash + -> Seq Scan on bloom_star_dim_3 d3 + -> Hash + -> Seq Scan on bloom_star_dim_4 d4 + -> Hash + -> Seq Scan on bloom_star_dim_5 d5 + -> Hash + -> Seq Scan on bloom_star_dim_6 d6 + -> Hash + Bloom Filter 2 + -> Seq Scan on bloom_star_dim_7 d7 + Filter: (r < '0.5'::double precision) +(35 rows) + +-- all joins selective for a filter (more than how many we allow) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.3 AND d2.r < 0.35 AND d3.r < 0.4 AND d4.r < 0.45 AND d5.r < 0.5 AND d6.r < 0.55 AND d7.r < 0.6; + QUERY PLAN +----------------------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id7 = d7.id) + -> Hash Join + Hash Cond: (f.id6 = d6.id) + -> Hash Join + Hash Cond: (f.id5 = d5.id) + -> Hash Join + Hash Cond: (f.id4 = d4.id) + -> Hash Join + Hash Cond: (f.id3 = d3.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_star_fact f + Bloom Filter 1: keys=(id1) + Bloom Filter 2: keys=(id2) + Bloom Filter 3: keys=(id3) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_star_dim_1 d1 + Filter: (r < '0.3'::double precision) + -> Hash + Bloom Filter 2 + -> Seq Scan on bloom_star_dim_2 d2 + Filter: (r < '0.35'::double precision) + -> Hash + Bloom Filter 3 + -> Seq Scan on bloom_star_dim_3 d3 + Filter: (r < '0.4'::double precision) + -> Hash + -> Seq Scan on bloom_star_dim_4 d4 + Filter: (r < '0.45'::double precision) + -> Hash + -> Seq Scan on bloom_star_dim_5 d5 + Filter: (r < '0.5'::double precision) + -> Hash + -> Seq Scan on bloom_star_dim_6 d6 + Filter: (r < '0.55'::double precision) + -> Hash + -> Seq Scan on bloom_star_dim_7 d7 + Filter: (r < '0.6'::double precision) +(42 rows) + +SET bloom_filter_pushdown_max = 10; +-- all joins selective for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.3 AND d2.r < 0.35 AND d3.r < 0.4 AND d4.r < 0.45 AND d5.r < 0.5 AND d6.r < 0.55 AND d7.r < 0.6; + QUERY PLAN +----------------------------------------------------------------------------------------- + Hash Join + Hash Cond: (f.id7 = d7.id) + -> Hash Join + Hash Cond: (f.id6 = d6.id) + -> Hash Join + Hash Cond: (f.id5 = d5.id) + -> Hash Join + Hash Cond: (f.id4 = d4.id) + -> Hash Join + Hash Cond: (f.id3 = d3.id) + -> Hash Join + Hash Cond: (f.id2 = d2.id) + -> Hash Join + Hash Cond: (f.id1 = d1.id) + -> Seq Scan on bloom_star_fact f + Bloom Filter 1: keys=(id1) + Bloom Filter 2: keys=(id2) + Bloom Filter 3: keys=(id3) + Bloom Filter 4: keys=(id4) + Bloom Filter 5: keys=(id5) + Bloom Filter 6: keys=(id6) + Bloom Filter 7: keys=(id7) + -> Hash + Bloom Filter 1 + -> Seq Scan on bloom_star_dim_1 d1 + Filter: (r < '0.3'::double precision) + -> Hash + Bloom Filter 2 + -> Seq Scan on bloom_star_dim_2 d2 + Filter: (r < '0.35'::double precision) + -> Hash + Bloom Filter 3 + -> Seq Scan on bloom_star_dim_3 d3 + Filter: (r < '0.4'::double precision) + -> Hash + Bloom Filter 4 + -> Seq Scan on bloom_star_dim_4 d4 + Filter: (r < '0.45'::double precision) + -> Hash + Bloom Filter 5 + -> Seq Scan on bloom_star_dim_5 d5 + Filter: (r < '0.5'::double precision) + -> Hash + Bloom Filter 6 + -> Seq Scan on bloom_star_dim_6 d6 + Filter: (r < '0.55'::double precision) + -> Hash + Bloom Filter 7 + -> Seq Scan on bloom_star_dim_7 d7 + Filter: (r < '0.6'::double precision) +(50 rows) + +RESET bloom_filter_pushdown_max; +RESET join_collapse_limit; +DROP TABLE bloom_star_dim_1; +DROP TABLE bloom_star_dim_2; +DROP TABLE bloom_star_dim_3; +DROP TABLE bloom_star_dim_4; +DROP TABLE bloom_star_dim_5; +DROP TABLE bloom_star_dim_6; +DROP TABLE bloom_star_dim_7; +DROP TABLE bloom_star_fact; diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule index 8fa0a6c47fb..0de7bdc3c56 100644 --- a/src/test/regress/parallel_schedule +++ b/src/test/regress/parallel_schedule @@ -146,3 +146,6 @@ test: fast_default # run tablespace test at the end because it drops the tablespace created during # setup that other tests may use. test: tablespace + +# tests for hashjoin bloom filter pushdown +test: hashjoin_bloom hashjoin_bloom_star hashjoin_bloom_snowflake diff --git a/src/test/regress/sql/hashjoin_bloom.sql b/src/test/regress/sql/hashjoin_bloom.sql new file mode 100644 index 00000000000..f0a2a9cc496 --- /dev/null +++ b/src/test/regress/sql/hashjoin_bloom.sql @@ -0,0 +1,40 @@ +-- tests to validate bloom filter pushdown +-- no parallel query support for now +SET max_parallel_workers_per_gather = 0; + +-- a couple very simple join queries +CREATE TABLE bloom_simple_dim (id int, r real); +CREATE TABLE bloom_simple_fact (id int, padding text); + +INSERT INTO bloom_simple_dim SELECT i, random() FROM generate_series(1,1000) s(i); + +INSERT INTO bloom_simple_fact +SELECT + 1 + mod(i, 1000), + md5(i::text) +FROM generate_series(1,100000) s(i); + +VACUUM ANALYZE; + +-- simple query, no join is selective enough for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_simple_fact f +JOIN bloom_simple_dim d ON (f.id = d.id); + +-- join is 75% selective (not enough for a filter to be created) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_simple_fact f +JOIN bloom_simple_dim d ON (f.id = d.id) +WHERE d.r < 0.75; + +-- join is 50% selective (enough for a filter to be pushed down) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_simple_fact f +JOIN bloom_simple_dim d ON (f.id = d.id) +WHERE d.r < 0.5; + +DROP TABLE bloom_simple_dim; +DROP TABLE bloom_simple_fact; diff --git a/src/test/regress/sql/hashjoin_bloom_snowflake.sql b/src/test/regress/sql/hashjoin_bloom_snowflake.sql new file mode 100644 index 00000000000..e5dcb2dea05 --- /dev/null +++ b/src/test/regress/sql/hashjoin_bloom_snowflake.sql @@ -0,0 +1,139 @@ +-- tests to validate bloom filter pushdown +-- no parallel query support for now +SET max_parallel_workers_per_gather = 0; + +-- simple snowflake queries (fact + up to 7 dimensions) +CREATE TABLE bloom_snowflake_dim_1_1 (id int, r real); +CREATE TABLE bloom_snowflake_dim_1_2 (id int, r real); +CREATE TABLE bloom_snowflake_dim_1 (id int, id11 int, id12 int, r real); +CREATE TABLE bloom_snowflake_dim_2_1 (id int, r real); +CREATE TABLE bloom_snowflake_dim_2_2 (id int, r real); +CREATE TABLE bloom_snowflake_dim_2 (id int, id21 int, id22 int, r real); + +CREATE TABLE bloom_snowflake_fact (id1 int, id2 int, padding text); + +INSERT INTO bloom_snowflake_dim_1_1 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_1_2 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_1 SELECT i, 1 + mod(i, 10000), 1 + mod(i, 10000), random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_2_1 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_2_2 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_snowflake_dim_2 SELECT i, 1 + mod(i, 10000), 1 + mod(i, 10000), random() FROM generate_series(1,10000) s(i); + +INSERT INTO bloom_snowflake_fact +SELECT + 1 + mod(i, 10000), + 1 + mod(i, 10000), + md5(i::text) +FROM generate_series(1,1000000) s(i); + +VACUUM ANALYZE; + +-- simple query, no join is selective enough for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id); + +-- join is 75% selective (not enough for a filter to be created) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d1.r < 0.75; + +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.75; + +-- join is 50% selective (good enough for pushdown) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d1.r < 0.5; + +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.5; + +-- increase the accepted build size (includes the fact) +SET bloom_filter_pushdown_max_build_relids = 4; + +-- join is 50% selective (good enough for pushdown) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d1.r < 0.5; + +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.5; + +-- needed to stabilize the join order +SET join_collapse_limit = 1; + +-- two joins with 75% selectivity (not enough for pushdown individually, +-- but good enough when combined) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_snowflake_fact f +JOIN bloom_snowflake_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_snowflake_dim_1_1 d11 ON (d1.id11 = d11.id) +JOIN bloom_snowflake_dim_1_2 d12 ON (d1.id12 = d12.id) +JOIN bloom_snowflake_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_snowflake_dim_2_1 d21 ON (d2.id21 = d21.id) +JOIN bloom_snowflake_dim_2_2 d22 ON (d2.id22 = d22.id) +WHERE d11.r < 0.75 AND d12.r < 0.75; + +RESET join_collapse_limit; +RESET bloom_filter_pushdown_max_build_relids; + +DROP TABLE bloom_snowflake_dim_1_1; +DROP TABLE bloom_snowflake_dim_1_2; +DROP TABLE bloom_snowflake_dim_1; +DROP TABLE bloom_snowflake_dim_2_1; +DROP TABLE bloom_snowflake_dim_2_2; +DROP TABLE bloom_snowflake_dim_2; +DROP TABLE bloom_snowflake_fact; diff --git a/src/test/regress/sql/hashjoin_bloom_star.sql b/src/test/regress/sql/hashjoin_bloom_star.sql new file mode 100644 index 00000000000..7ee3069ccd6 --- /dev/null +++ b/src/test/regress/sql/hashjoin_bloom_star.sql @@ -0,0 +1,147 @@ +-- tests to validate bloom filter pushdown +-- no parallel query support for now +SET max_parallel_workers_per_gather = 0; + +-- needed, because the filter pushdown eliminates one of the inputs for +-- choosing join order (reduction of result size) +SET join_collapse_limit = 1; + +-- simple starjoin queries (fact + up to 7 dimensions) +CREATE TABLE bloom_star_dim_1 (id int, r real); +CREATE TABLE bloom_star_dim_2 (id int, r real); +CREATE TABLE bloom_star_dim_3 (id int, r real); +CREATE TABLE bloom_star_dim_4 (id int, r real); +CREATE TABLE bloom_star_dim_5 (id int, r real); +CREATE TABLE bloom_star_dim_6 (id int, r real); +CREATE TABLE bloom_star_dim_7 (id int, r real); + +CREATE TABLE bloom_star_fact (id1 int, id2 int, id3 int, id4 int, id5 int, id6 int, id7 int, padding text); + +INSERT INTO bloom_star_dim_1 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_2 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_3 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_4 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_5 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_6 SELECT i, random() FROM generate_series(1,10000) s(i); +INSERT INTO bloom_star_dim_7 SELECT i, random() FROM generate_series(1,10000) s(i); + +INSERT INTO bloom_star_fact +SELECT + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + 1 + mod(i, 10000), + md5(i::text) +FROM generate_series(1,1000000) s(i); + +VACUUM ANALYZE; + +-- simple query, no join is selective enough for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id); + +-- last join is 75% selective (not enough for a filter to be created) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d7.r < 0.75; + +-- last join is 50% selective (good enough for a filter push down) +-- the join will be executed done last +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d7.r < 0.5; + +-- first join is 50% selective (good enough for a filter push down) +-- the join will be executed done last +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.5; + +-- two joins 50% selective +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.4 AND d7.r < 0.5; + +-- all joins selective for a filter (more than how many we allow) +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.3 AND d2.r < 0.35 AND d3.r < 0.4 AND d4.r < 0.45 AND d5.r < 0.5 AND d6.r < 0.55 AND d7.r < 0.6; + + +SET bloom_filter_pushdown_max = 10; + +-- all joins selective for a filter +EXPLAIN (COSTS OFF) +SELECT * +FROM bloom_star_fact f +JOIN bloom_star_dim_1 d1 ON (f.id1 = d1.id) +JOIN bloom_star_dim_2 d2 ON (f.id2 = d2.id) +JOIN bloom_star_dim_3 d3 ON (f.id3 = d3.id) +JOIN bloom_star_dim_4 d4 ON (f.id4 = d4.id) +JOIN bloom_star_dim_5 d5 ON (f.id5 = d5.id) +JOIN bloom_star_dim_6 d6 ON (f.id6 = d6.id) +JOIN bloom_star_dim_7 d7 ON (f.id7 = d7.id) +WHERE d1.r < 0.3 AND d2.r < 0.35 AND d3.r < 0.4 AND d4.r < 0.45 AND d5.r < 0.5 AND d6.r < 0.55 AND d7.r < 0.6; + +RESET bloom_filter_pushdown_max; +RESET join_collapse_limit; + +DROP TABLE bloom_star_dim_1; +DROP TABLE bloom_star_dim_2; +DROP TABLE bloom_star_dim_3; +DROP TABLE bloom_star_dim_4; +DROP TABLE bloom_star_dim_5; +DROP TABLE bloom_star_dim_6; +DROP TABLE bloom_star_dim_7; +DROP TABLE bloom_star_fact; -- 2.55.0