Add tests for special hash values (#5431)

We fixed some crashes a while back that would only occur in cases where
the value of a distribution column would have result in a high or a very
low hash value. This adds a regression test for those crashes.

src/test/regress/expected/distributed_planning.out

@@ -2,12 +2,36 @@ SET search_path TO "distributed planning";
 -- Confirm the basics work
 INSERT INTO test VALUES (1, 2), (3, 4), (5, 6), (2, 7), (4, 5);
 INSERT INTO test VALUES (6, 7);
+-- Insert two edge case values, the first value hashes to MAX_INT32 and the
+-- second hashes to MIN_INT32. This should not break anything, but these cases
+-- caused some crashes in the past.
+-- See c6c31e0f1fe5b8cc955b0da42264578dcdae16cc and
+-- 683279cc366069db9c2ccaca85dfaf8572113cda for details.
+--
+-- These specific values were found by using the following two queries:
+-- select q.i from (select generate_series(0, 10000000000000) i) q where hashint8(q.i) = 2147483647 limit 1;
+-- select q.i from (select generate_series(0, 10000000000000) i) q where hashint8(q.i) = -2147483648 limit 1;
+INSERT INTO test VALUES (2608474032, 2608474032), (963809240, 963809240);
 SELECT * FROM test WHERE x = 1 ORDER BY y, x;
  x | y
 ---------------------------------------------------------------------
  1 | 2
 (1 row)
 
+-- Confirm that hash values are as expected
+SELECT hashint8(x) FROM test ORDER BY 1;
+  hashint8
+---------------------------------------------------------------------
+ -2147483648
+ -1905060026
+ -1330264708
+ -1011077333
+   -28094569
+   566031088
+  1134484726
+  2147483647
+(8 rows)
+
 SELECT t1.x, t2.y FROM test t1 JOIN test t2 USING(x) WHERE t1.x = 1 AND t2.x = 1 ORDER BY t2.y, t1.x;
  x | y
 ---------------------------------------------------------------------
@@ -24,7 +48,7 @@ SELECT * FROM test WHERE x = 1 OR x = 2 ORDER BY y, x;
 SELECT count(*) FROM test;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 SELECT * FROM test ORDER BY x;
@@ -36,7 +60,9 @@ SELECT * FROM test ORDER BY x;
           4 |          5
           5 |          6
           6 |          7
-(6 rows)
+  963809240 |  963809240
+ 2608474032 | 2608474032
+(8 rows)
 
 WITH cte_1 AS (UPDATE test SET y = y - 1 RETURNING *) SELECT * FROM cte_1 ORDER BY 1,2;
      x      |     y
@@ -47,7 +73,9 @@ WITH cte_1 AS (UPDATE test SET y = y - 1 RETURNING *) SELECT * FROM cte_1 ORDER
           4 |          4
           5 |          5
           6 |          6
-(6 rows)
+  963809240 |  963809239
+ 2608474032 | 2608474031
+(8 rows)
 
 -- observe that there is a conflict and the following query does nothing
 INSERT INTO upsert_test (part_key, other_col) VALUES (1, 1) ON CONFLICT DO NOTHING RETURNING *;
@@ -117,7 +145,7 @@ BEGIN;
 	SELECT count(*) FROM cte_1, cte_2;
  count
 ---------------------------------------------------------------------
-  1007
+  1009
 (1 row)
 
 ROLLBACK;
@@ -135,7 +163,7 @@ CREATE VIEW simple_view AS
 SELECT * FROM simple_view;
  cnt
 ---------------------------------------------------------------------
-   8
+  10
 (1 row)
 
 SELECT * FROM simple_view, test WHERE test.x = simple_view.cnt;
@@ -148,11 +176,11 @@ BEGIN;
 	SELECT count(*) FROM test;
  count
 ---------------------------------------------------------------------
-    16
+    18
 (1 row)
 
 	COPY (SELECT count(DISTINCT x) FROM test) TO STDOUT;
-10
+12
 	INSERT INTO test SELECT i,i FROM generate_series(0,100)i;
 ROLLBACK;
 -- prepared statements with custom types
@@ -258,13 +286,13 @@ WITH cte_1 AS (SELECT * FROM test_2) SELECT * FROM cte_1, test_2 WHERE  test_2.x
 SELECT count(DISTINCT x) FROM test;
  count
 ---------------------------------------------------------------------
-     5
+     7
 (1 row)
 
 SELECT count(DISTINCT y) FROM test;
  count
 ---------------------------------------------------------------------
-     5
+     7
 (1 row)
 
 -- query pushdown should work
@@ -300,7 +328,7 @@ INSERT INTO test(x, y) SELECT x, y FROM test WHERE x = 1;
 SELECT count(*) from test;
  count
 ---------------------------------------------------------------------
-     7
+     9
 (1 row)
 
 ROLLBACK;
@@ -310,7 +338,7 @@ INSERT INTO test(x, y) SELECT x, y FROM test;
 SELECT count(*) from test;
  count
 ---------------------------------------------------------------------
-    12
+    16
 (1 row)
 
 ROLLBACK;
@@ -320,7 +348,7 @@ INSERT INTO test(x, y) SELECT count(x), max(y) FROM test;
 SELECT count(*) from test;
  count
 ---------------------------------------------------------------------
-     7
+     9
 (1 row)
 
 ROLLBACK;
@@ -330,7 +358,7 @@ INSERT INTO test(x, y) SELECT y, x FROM test;
 SELECT count(*) from test;
  count
 ---------------------------------------------------------------------
-    12
+    16
 (1 row)
 
 ROLLBACK;
@@ -340,7 +368,7 @@ INSERT INTO test(x, y) SELECT a, b FROM ref;
 SELECT count(*) from test;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 ROLLBACK;
@@ -350,7 +378,7 @@ INSERT INTO test(x, y) SELECT c, d FROM local;
 SELECT count(*) from test;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 ROLLBACK;
@@ -360,7 +388,7 @@ INSERT INTO ref(a, b) SELECT x, y FROM test;
 SELECT count(*) from ref;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 ROLLBACK;
@@ -380,7 +408,7 @@ INSERT INTO local(c, d) SELECT x, y FROM test;
 SELECT count(*) from local;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 ROLLBACK;
@@ -409,7 +437,7 @@ SELECT count(*) FROM test WHERE false GROUP BY GROUPING SETS (x,y);
 SELECT count(*) FROM test;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 -- INSERT SELECT from distributed table to local table
@@ -418,7 +446,7 @@ INSERT INTO ref(a, b) SELECT x, y FROM test;
 SELECT count(*) from ref;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 ROLLBACK;
@@ -438,7 +466,7 @@ INSERT INTO local(c, d) SELECT x, y FROM test;
 SELECT count(*) from local;
  count
 ---------------------------------------------------------------------
-     6
+     8
 (1 row)
 
 ROLLBACK;

src/test/regress/expected/distributed_planning_create_load.out

@@ -24,7 +24,7 @@ SELECT create_time_partitions(table_name:='date_part_table',
 
 INSERT INTO date_part_table
 	SELECT '2020-01-01'::timestamp + '3 hours'::interval * i, i, i % 20 FROM generate_series(0,100)i;
-CREATE TABLE test(x int, y int);
+CREATE TABLE test(x bigint, y bigint);
 SELECT create_distributed_table('test','x');
  create_distributed_table
 ---------------------------------------------------------------------
@@ -32,28 +32,28 @@ SELECT create_distributed_table('test','x');
 (1 row)
 
 CREATE TYPE new_type AS (n int, m text);
-CREATE TABLE test_2(x int, y int, z new_type);
+CREATE TABLE test_2(x bigint, y bigint, z new_type);
 SELECT create_distributed_table('test_2','x');
  create_distributed_table
 ---------------------------------------------------------------------
 
 (1 row)
 
-CREATE TABLE ref(a int, b int);
+CREATE TABLE ref(a bigint, b bigint);
 SELECT create_reference_table('ref');
  create_reference_table
 ---------------------------------------------------------------------
 
 (1 row)
 
-CREATE TABLE ref2(a int, b int);
+CREATE TABLE ref2(a bigint, b bigint);
 SELECT create_reference_table('ref2');
  create_reference_table
 ---------------------------------------------------------------------
 
 (1 row)
 
-CREATE TABLE local(c int, d int);
+CREATE TABLE local(c bigint, d bigint);
 select citus_add_local_table_to_metadata('local');
  citus_add_local_table_to_metadata
 ---------------------------------------------------------------------

src/test/regress/sql/distributed_planning.sql

@@ -3,7 +3,19 @@ SET search_path TO "distributed planning";
 -- Confirm the basics work
 INSERT INTO test VALUES (1, 2), (3, 4), (5, 6), (2, 7), (4, 5);
 INSERT INTO test VALUES (6, 7);
+-- Insert two edge case values, the first value hashes to MAX_INT32 and the
+-- second hashes to MIN_INT32. This should not break anything, but these cases
+-- caused some crashes in the past.
+-- See c6c31e0f1fe5b8cc955b0da42264578dcdae16cc and
+-- 683279cc366069db9c2ccaca85dfaf8572113cda for details.
+--
+-- These specific values were found by using the following two queries:
+-- select q.i from (select generate_series(0, 10000000000000) i) q where hashint8(q.i) = 2147483647 limit 1;
+-- select q.i from (select generate_series(0, 10000000000000) i) q where hashint8(q.i) = -2147483648 limit 1;
+INSERT INTO test VALUES (2608474032, 2608474032), (963809240, 963809240);
 SELECT * FROM test WHERE x = 1 ORDER BY y, x;
+-- Confirm that hash values are as expected
+SELECT hashint8(x) FROM test ORDER BY 1;
 SELECT t1.x, t2.y FROM test t1 JOIN test t2 USING(x) WHERE t1.x = 1 AND t2.x = 1 ORDER BY t2.y, t1.x;
 SELECT * FROM test WHERE x = 1 OR x = 2 ORDER BY y, x;
 SELECT count(*) FROM test;

src/test/regress/sql/distributed_planning_create_load.sql

@@ -20,20 +20,20 @@ SELECT create_time_partitions(table_name:='date_part_table',
 INSERT INTO date_part_table
 	SELECT '2020-01-01'::timestamp + '3 hours'::interval * i, i, i % 20 FROM generate_series(0,100)i;
 
-CREATE TABLE test(x int, y int);
+CREATE TABLE test(x bigint, y bigint);
 SELECT create_distributed_table('test','x');
 
 CREATE TYPE new_type AS (n int, m text);
-CREATE TABLE test_2(x int, y int, z new_type);
+CREATE TABLE test_2(x bigint, y bigint, z new_type);
 SELECT create_distributed_table('test_2','x');
 
-CREATE TABLE ref(a int, b int);
+CREATE TABLE ref(a bigint, b bigint);
 SELECT create_reference_table('ref');
 
-CREATE TABLE ref2(a int, b int);
+CREATE TABLE ref2(a bigint, b bigint);
 SELECT create_reference_table('ref2');
 
-CREATE TABLE local(c int, d int);
+CREATE TABLE local(c bigint, d bigint);
 select citus_add_local_table_to_metadata('local');
 
 CREATE TABLE non_binary_copy_test (key int PRIMARY KEY, value new_type);