MDEV-31887: wrong result with split optimization

Wrong result is produced when split-materialized optimization is used for
grouping with order by and limit.

The fix is to not let Split-Materialized optimization to happen
when the sub-query has an ORDER BY with LIMIT, by returning FALSE early
in the method opt_split.cc#check_for_splittable_materialized()

However, with just the above change, there is a side-effect of
NOT "using index for group by" in the scenario when
all the following conditions are met: -
  1. The query has derived table with GROUP BY and ORDER BY LIMIT
  2. joined in a way that would allow Split-Materialized
     if ORDER BY LIMIT wasn't present
  3. An index suitable for using "index for group-by"
  4. No where clause so that, "using for group by" is applicable,
     but the index is not included in "possible_keys".
The reason being, join_tab's "keys" field wasn't being set in
sql_select.cc#make_join_select(). So, made this change as well
as part of this PR.

Author: bsrikanth-mariadb

mysql-test/main/derived_split_innodb.result

@@ -1002,5 +1002,48 @@ set statement optimizer_switch='split_materialized=off' for $query;
 a	b	name	total_amt
 1	NULL	A	10
 DROP TABLE t1,t2;
+#
+# MDEV-37407 Wrong result with ORDER BY LIMIT
+# Both, with and without split_materialized should
+# produce the same results
+#
+SET @save_optimizer_switch= @@optimizer_switch;
+CREATE TABLE t1
+(a varchar(35), b varchar(4), KEY (a))
+ENGINE=InnoDB;
+INSERT INTO t1 VALUES
+('Albania','AXA'), ('Australia','AUS'), ('Myanmar','MMR'),
+('Bahamas','BS'), ('Brazil','BRA'), ('Barbados','BRB');
+CREATE TABLE t2
+(a varchar(4), b varchar(50), PRIMARY KEY (b,a), KEY (a))
+ENGINE=InnoDB;
+INSERT INTO t2 VALUES
+('AUS','Anglican'), ('MMR','Baptist'), ('BS','Anglican'),
+('BS','Baptist'), ('BS','Methodist'), ('BRB','Methodist'),
+('BRA','Baptist'), ('USA','Baptist');
+ANALYZE TABLE t1 PERSISTENT FOR ALL;
+Table	Op	Msg_type	Msg_text
+test.t1	analyze	status	Engine-independent statistics collected
+test.t1	analyze	status	OK
+ANALYZE TABLE t2 PERSISTENT FOR ALL;
+Table	Op	Msg_type	Msg_text
+test.t2	analyze	status	Engine-independent statistics collected
+test.t2	analyze	status	OK
+set optimizer_switch='split_materialized=off';
+SELECT t1.a
+FROM (SELECT a FROM t2 GROUP BY a ORDER BY a, COUNT(DISTINCT b) LIMIT 1) dt
+JOIN t1 ON
+dt.a=t1.b
+WHERE t1.a LIKE 'B%';
+a
+set optimizer_switch='split_materialized=on';
+SELECT t1.a
+FROM (SELECT a FROM t2 GROUP BY a ORDER BY a, COUNT(DISTINCT b) LIMIT 1) dt
+JOIN t1 ON
+dt.a=t1.b
+WHERE t1.a LIKE 'B%';
+a
+DROP TABLE t1,t2;
+SET optimizer_switch= @save_optimizer_switch;
 # End of 10.11 tests
 SET GLOBAL innodb_stats_persistent=@save_innodb_stats_persistent;

mysql-test/main/derived_split_innodb.test

@@ -620,5 +620,49 @@ evalp set statement optimizer_switch='split_materialized=off' for $query;
 
 DROP TABLE t1,t2;
 
+--echo #
+--echo # MDEV-37407 Wrong result with ORDER BY LIMIT
+--echo # Both, with and without split_materialized should
+--echo # produce the same results
+--echo #
+
+SET @save_optimizer_switch= @@optimizer_switch;
+
+CREATE TABLE t1
+  (a varchar(35), b varchar(4), KEY (a))
+ENGINE=InnoDB;
+
+INSERT INTO t1 VALUES
+('Albania','AXA'), ('Australia','AUS'), ('Myanmar','MMR'),
+('Bahamas','BS'), ('Brazil','BRA'), ('Barbados','BRB');
+ 
+CREATE TABLE t2
+  (a varchar(4), b varchar(50), PRIMARY KEY (b,a), KEY (a))
+ENGINE=InnoDB;
+
+INSERT INTO t2 VALUES
+('AUS','Anglican'), ('MMR','Baptist'), ('BS','Anglican'),
+('BS','Baptist'), ('BS','Methodist'), ('BRB','Methodist'),
+('BRA','Baptist'), ('USA','Baptist');
+
+ANALYZE TABLE t1 PERSISTENT FOR ALL;
+ANALYZE TABLE t2 PERSISTENT FOR ALL;
+
+let $q=
+SELECT t1.a
+FROM (SELECT a FROM t2 GROUP BY a ORDER BY a, COUNT(DISTINCT b) LIMIT 1) dt
+     JOIN t1 ON
+     dt.a=t1.b
+WHERE t1.a LIKE 'B%';
+
+set optimizer_switch='split_materialized=off';
+eval $q;
+
+set optimizer_switch='split_materialized=on';
+eval $q;
+
+DROP TABLE t1,t2;
+
+SET optimizer_switch= @save_optimizer_switch;
 --echo # End of 10.11 tests
 SET GLOBAL innodb_stats_persistent=@save_innodb_stats_persistent;

mysql-test/main/distinct.result

@@ -212,7 +212,7 @@ id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	SIMPLE	t1	index	NULL	PRIMARY	4	NULL	1	Using index
 explain SELECT distinct a from t3 order by a desc limit 2;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t3	index	NULL	a	5	NULL	40	Using index
+1	SIMPLE	t3	range	a	a	5	NULL	10	Using index for group-by; Using temporary; Using filesort
 explain SELECT distinct a,b from t3 order by a+1;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	SIMPLE	t3	ALL	NULL	NULL	NULL	NULL	204	Using temporary; Using filesort

mysql-test/main/group_min_max.result

@@ -2451,7 +2451,7 @@ EXPLAIN SELECT 1 FROM t1 AS t1_outer WHERE EXISTS
 (SELECT max(b) FROM t1 GROUP BY a HAVING a < 2);
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	PRIMARY	t1_outer	index	NULL	a	10	NULL	15	Using index
-2	SUBQUERY	t1	index	NULL	a	10	NULL	15	Using index
+2	SUBQUERY	t1	range	a	a	5	NULL	6	Using index for group-by
 EXPLAIN SELECT 1 FROM t1 AS t1_outer WHERE 
 (SELECT max(b) FROM t1 GROUP BY a HAVING a < 2) > 12;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra

sql/opt_split.cc

@@ -355,6 +355,7 @@ struct SplM_field_ext_info: public SplM_field_info
        with available statistics.
     10. The select doesn't use WITH ROLLUP (This limitation can probably be
        lifted)
+    11. The select doesn't have ORDER BY with LIMIT
 
   @retval
     true   if the answer is positive
@@ -388,6 +389,9 @@ bool JOIN::check_for_splittable_materialized()
   if (!partition_list)
     return false;
 
+  if (select_lex->order_list.elements > 0 && !unit->lim.is_unlimited()) //!(11)
+    return false;
+
   Json_writer_object trace_wrapper(thd);
   Json_writer_object trace_split(thd, "check_split_materialized");
 

sql/sql_select.cc

@@ -13657,7 +13657,21 @@ make_join_select(JOIN *join,SQL_SELECT *select,COND *cond)
 
             if (build_tmp_join_prefix_cond(join, tab, &sel->cond))
               return true;
-	    /*
+
+            /*
+              To be removed in 11.0+:
+              Caution: we can reach this point with quick=NULL. Below, we'll
+              use tab->keys and not tab->const_keys like
+              get_quick_record_count() did. If we have constructed a
+              group-min-max quick select, make sure we're able to construct it
+              again
+            */
+            if (sel->quick && sel->quick->get_type() ==
+                QUICK_SELECT_I::QS_TYPE_GROUP_MIN_MAX)
+            {
+              tab->keys.set_bit(sel->quick->index);
+            }
+      /*
               We can't call sel->cond->fix_fields,
               as it will break tab->on_expr if it's AND condition
               (fix_fields currently removes extra AND/OR levels).