Bug#24423143 - WRONG RESULTS FOR AGGREGATE QUERY

Optimizer chooses loose index scan (QUICK_GROUP_MIN_MAX_SELECT)
for a group by even when there exists a predicate with disjunction.
This is due to that either of the below two cases are encountered
and not handled properly.

1. A range tree created for index merge scan is not taken
into account while checking for the presence of disjuncted
conditions in get_best_group_min_max().

2. A disjunction condition could lead to null tree if an index
merge scan is not possible. A NULL tree will cause
get_best_group_min_max() to skip many relevent checks
like WA2.

There are two scenarios here,
  (a) The WHERE clause is a disjunction of conditions on
     MIN/MAX column only.
    => We can use min_max optimization considering all
    other criteria (eg:SA2) are met.

 (b) MIN/MAX column is participating in a disjunctive WHERE
     clause along with other columns.
     => min_max optimization is not applicable in this scenario.

Fix:
Skip loose index scan in below scenarios involving disjoint conditions
in WHERE clause.

1. When the condition in WHERE clause results in more than
one disjoint range trees (when index merge scan is possible).

2. The range tree is null and MIN/MAX column participates
in the WHERE clause along with other columns.
[There is no range tree if WHERE condition can't be represented
in a single range tree and index merge is not possible]

Change-Id: I2ec8ef815d9dfa64e0bf60e97d3b6129c070e24b

Author: Ajo Robert

mysql-test/r/group_min_max.result

@@ -3794,3 +3794,280 @@ Handler_read_rnd	0
 Handler_read_rnd_next	0
 DROP TABLE t1;
 # End of test for Bug#18109609
+#
+# Bug#24423143 - WRONG RESULTS FOR AGGREGATE QUERY
+#
+# Test index merge tree scenario
+CREATE TABLE A (
+aggr_col int,
+group_by_col int,
+KEY aggr_col_key (aggr_col),
+KEY group_by_col_key (group_by_col, aggr_col)
+) ENGINE=InnoDB;
+set optimizer_trace_max_mem_size=1048576;
+set @@session.optimizer_trace='enabled=on';
+set end_markers_in_json=on;
+INSERT INTO A VALUES (2,3),(5,6),(6,3),(7,NULL),(9,NULL),(10,6);
+ANALYZE TABLE A;
+Table	Op	Msg_type	Msg_text
+test.A	analyze	status	OK
+SELECT group_by_col, MIN(aggr_col) FROM A
+WHERE (group_by_col IN (70, 9)) OR (aggr_col > 2) GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)
+NULL	7
+3	6
+6	5
+EXPLAIN SELECT group_by_col, MIN(aggr_col) FROM A
+WHERE (group_by_col IN (70 ,9)) OR (aggr_col > 2) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	aggr_col_key,group_by_col_key	group_by_col_key	10	NULL	6	55.55	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)` from `test`.`A` where ((`test`.`A`.`group_by_col` in (70,9)) or (`test`.`A`.`aggr_col` > 2)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'disjuntive_predicate_present' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+SELECT group_by_col, MAX(aggr_col) FROM A
+WHERE (group_by_col IN (70, 9)) OR (aggr_col < 9) GROUP BY group_by_col;
+group_by_col	MAX(aggr_col)
+NULL	7
+3	6
+6	5
+EXPLAIN SELECT group_by_col, MAX(aggr_col) FROM A
+WHERE (group_by_col IN (70 , 9)) OR (aggr_col < 9) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	aggr_col_key,group_by_col_key	group_by_col_key	10	NULL	6	55.55	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where ((`test`.`A`.`group_by_col` in (70,9)) or (`test`.`A`.`aggr_col` < 9)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'disjuntive_predicate_present' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Test IMPOSSIBLE TREE scenario
+ALTER TABLE A DROP KEY aggr_col_key;
+SELECT group_by_col, MIN(aggr_col) FROM A
+WHERE (group_by_col IN (70 ,9)) OR (aggr_col > 2) GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)
+NULL	7
+3	6
+6	5
+EXPLAIN SELECT group_by_col, MIN(aggr_col) FROM A
+WHERE (group_by_col IN (70, 9)) OR (aggr_col > 2) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	6	55.55	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)` from `test`.`A` where ((`test`.`A`.`group_by_col` in (70,9)) or (`test`.`A`.`aggr_col` > 2)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+SELECT group_by_col, MAX(aggr_col) FROM A
+WHERE (group_by_col IN (70, 9)) OR (aggr_col < 9) GROUP BY group_by_col;
+group_by_col	MAX(aggr_col)
+NULL	7
+3	6
+6	5
+EXPLAIN SELECT group_by_col, MAX(aggr_col) FROM A
+WHERE (group_by_col IN (70, 9)) OR (aggr_col < 9) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	6	55.55	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where ((`test`.`A`.`group_by_col` in (70,9)) or (`test`.`A`.`aggr_col` < 9)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 3: aggregate field used as equal expression.
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE (group_by_col IN (3, 9)) OR (aggr_col = 9) GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+NULL	9	9
+3	2	6
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE (group_by_col IN (3, 9)) OR (aggr_col = 9) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	6	44.44	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where ((`test`.`A`.`group_by_col` in (3,9)) or (`test`.`A`.`aggr_col` = 9)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 4: non aggregate field used as equal expression.
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE (group_by_col = 3) OR (aggr_col > 8) GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+NULL	9	9
+3	2	6
+6	10	10
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE (group_by_col = 3) OR (aggr_col > 8) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	6	55.55	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where ((`test`.`A`.`group_by_col` = 3) or (`test`.`A`.`aggr_col` > 8)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 5: aggregate field used as non-zero expression.
+INSERT INTO A VALUES(0, 3);
+INSERT INTO A VALUES(0, 9);
+INSERT INTO A VALUES(8, 0);
+ANALYZE TABLE A;
+Table	Op	Msg_type	Msg_text
+test.A	analyze	status	OK
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE (group_by_col = 9) OR aggr_col GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+NULL	7	9
+0	8	8
+3	2	6
+6	5	10
+9	0	0
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE group_by_col = 9 OR aggr_col GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	9	91.11	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where ((`test`.`A`.`group_by_col` = 9) or `test`.`A`.`aggr_col`) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 6: non aggregate field used as non-zero expression.
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE group_by_col OR (aggr_col < 9) GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+NULL	7	7
+0	8	8
+3	0	6
+6	5	10
+9	0	0
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE group_by_col OR (aggr_col < 9) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	9	92.59	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where (`test`.`A`.`group_by_col` or (`test`.`A`.`aggr_col` < 9)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 7: aggregate field used in equal exp without a CONST
+INSERT INTO A VALUES(1,1),(1,2),(2,1);
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col = group_by_col GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+1	1	1
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col = group_by_col GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	12	100.00	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where (`test`.`A`.`group_by_col` = `test`.`A`.`aggr_col`) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 8: aggregate field used in a non-eq exp without a CONST
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col < group_by_col GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+2	1	1
+3	0	2
+6	5	5
+9	0	0
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col < group_by_col GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	12	33.33	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where (`test`.`A`.`aggr_col` < `test`.`A`.`group_by_col`) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 8
+INSERT INTO A VALUES(0,1),(1,0),(0,0);
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col OR group_by_col GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+NULL	7	9
+0	1	8
+1	0	2
+2	1	1
+3	0	6
+6	5	10
+9	0	0
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col OR group_by_col GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	15	99.00	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where (`test`.`A`.`aggr_col` or `test`.`A`.`group_by_col`) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 9
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col AND group_by_col GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+1	1	2
+2	1	1
+3	2	6
+6	5	10
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col AND group_by_col GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	15	81.00	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where (`test`.`A`.`aggr_col` and `test`.`A`.`group_by_col`) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+# Scenario 10: Added for completion. This fix does not have an impact.
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col<>0 AND group_by_col<>0 GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+1	1	2
+2	1	1
+3	2	6
+6	5	10
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE aggr_col<>0 AND group_by_col<>0 GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	range	group_by_col_key	group_by_col_key	10	NULL	7	100.00	Using where; Using index for group-by
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where ((`test`.`A`.`aggr_col` <> 0) and (`test`.`A`.`group_by_col` <> 0)) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+0
+# Scenario 11: ITEM_FUNC as an argument of ITEM_FUNC
+SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE group_by_col OR (group_by_col < (aggr_col = 1)) GROUP BY group_by_col;
+group_by_col	MIN(aggr_col)	MAX(aggr_col)
+0	1	1
+1	0	2
+2	1	1
+3	0	6
+6	5	10
+9	0	0
+EXPLAIN SELECT group_by_col, MIN(aggr_col), MAX(aggr_col) FROM A
+WHERE group_by_col OR (group_by_col < (aggr_col = 1)) GROUP BY group_by_col;
+id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	A	NULL	index	group_by_col_key	group_by_col_key	10	NULL	15	93.33	Using where; Using index
+Warnings:
+Note	1003	/* select#1 */ select `test`.`A`.`group_by_col` AS `group_by_col`,min(`test`.`A`.`aggr_col`) AS `MIN(aggr_col)`,max(`test`.`A`.`aggr_col`) AS `MAX(aggr_col)` from `test`.`A` where (`test`.`A`.`group_by_col` or (`test`.`A`.`group_by_col` < (`test`.`A`.`aggr_col` = 1))) group by `test`.`A`.`group_by_col`
+SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query' AS OK
+FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
+OK
+1
+SET optimizer_trace_max_mem_size=DEFAULT;
+SET optimizer_trace=DEFAULT;
+SET end_markers_in_json=DEFAULT;
+DROP TABLE A;
+# End of test for Bug#24423143

mysql-test/r/group_min_max_innodb.result

@@ -338,7 +338,7 @@ F	30	30
 SELECT TRACE RLIKE 'minmax_keypart_in_disjunctive_query'
 AS OK FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
 OK
-0
+1
 SET optimizer_trace_max_mem_size=DEFAULT;
 SET optimizer_trace=DEFAULT;
 SET end_markers_in_json=DEFAULT;