WL#7339 Use improved records per key estimate interface in optimizer

      
This worklog implements the needed changes in the optimizer
and server to start using the new "records per key" interface that
was implmeneted in WL#7338. This interface allows storage engines
to provide "records per key" values using float values instead of
integer values. Using this new interface instead of using the
currently used rec_per_key values gives more correct record estimates
for ref access against an indexed column.

Author: Olav Sandstaa

mysql-test/include/explain_non_select.inc

@@ -188,6 +188,12 @@ CREATE TABLE t3 (a3 INT NOT NULL, b3 INT NOT NULL, PRIMARY KEY (a3,b3));
 INSERT INTO t1 VALUES (1,1), (2,1), (1,3);
 INSERT INTO t2 VALUES (1,1), (2,2), (3,3);
 INSERT INTO t3 VALUES (1,1), (2,1), (1,3);
+# Need to run ANALYZE to get correct index statistics
+--disable_query_log
+--disable_result_log
+ANALYZE TABLE t1, t2, t3;
+--enable_result_log
+--enable_query_log
 --let $query =  DELETE t1,t2,t3 FROM t1,t2,t3 WHERE a1=a2 AND b2=a3 AND b1=b3
 --let $select = SELECT *        FROM t1,t2,t3 WHERE a1=a2 AND b2=a3 AND b1=b3
 --source include/explain_utils.inc
@@ -591,6 +597,12 @@ CREATE TABLE t2(f3 INT, f4 INT);
 CREATE INDEX IDX ON t2(f3);
 INSERT INTO t1 VALUES(1,0),(2,0);
 INSERT INTO t2 VALUES(1,1),(2,2);
+# Need to run ANALYZE to get correct index statistics
+-- disable_query_log
+-- disable_result_log
+ANALYZE TABLE t1,t2;
+-- enable_result_log
+-- enable_query_log
 --let $query =  UPDATE t1 SET t1.f2=(SELECT MAX(t2.f4) FROM t2 WHERE t2.f3=t1.f1)
 --let $select = SELECT              (SELECT MAX(t2.f4) FROM t2 WHERE t2.f3=t1.f1) FROM t1
 --source include/explain_utils.inc

mysql-test/include/subquery_sj.inc

@@ -4173,7 +4173,8 @@ INSERT INTO t1 VALUES
 (15,2,'a','x'),
 (16,1,'d','d'),
 (17,8,'y','y'),
-(18,6,'t','u');
+(18,6,'t','u'),
+(19,6,'t','u');
 
 CREATE TABLE t2 (
   pk int NOT NULL,

mysql-test/include/subquery_sj_innodb.inc

@@ -9,7 +9,8 @@ SET GLOBAL innodb_stats_persistent=0;
 -- enable_query_log
 
 create table t0 (a int);
-insert into t0 values (0),(1),(2),(3),(4),(5),(6),(7),(8),(9);
+insert into t0 values (0),(1),(2),(3),(4),(5),(6),(7),(8),(9),
+                      (10),(12),(14),(16),(18);
 
 # First test simple cases: I20 order, no join buffering.
 
@@ -26,11 +27,21 @@ create table t2 (
 ) engine=innodb;
 insert into t2 select a, a/2 from t0;
 
+# Run analyze to ensure more correct index statistics
+-- disable_query_log
+-- disable_result_log
+ANALYZE TABLE t1,t2;
+-- enable_result_log
+-- enable_query_log
+
 select * from t1;
 select * from t2;
 explain select * from t2 where b in (select a from t1);
 select * from t2 where b in (select a from t1);
 
+truncate table t0;
+insert into t0 values (0),(1),(2),(3),(4),(5),(6),(7),(8),(9);
+
 # Try an InnoDB table with very long rowid
 create table t3 (
    a int, 

mysql-test/r/explain_for_connection_rqg_json.result

@@ -103,20 +103,13 @@ SYNC_POINT:before_reset_query_plan
 Diff: EXPLAIN NORMAL vs OTHER
 @@ -3,3 +3,3 @@
  1	PRIMARY	<derived2>	NULL	ALL	NULL	NULL	NULL	NULL	2	100.00	Using where; Using join buffer (Block Nested Loop)
--1	PRIMARY	table3	NULL	eq_ref	PRIMARY,col_int_key	PRIMARY	4	table2.pk	1	5.00	Using where
+-1	PRIMARY	table3	NULL	eq_ref	PRIMARY,col_int_key	PRIMARY	4	table2.pk	1	12.50	Using where
 +1	PRIMARY	table3	NULL	eq_ref	PRIMARY,col_int_key	PRIMARY	4	table2.pk	1	100.00	Using where
  3	DEPENDENT SUBQUERY	subquery2_t1	NULL	ALL	NULL	NULL	NULL	NULL	20	100.00	NULL
 QUERY: SELECT     (SELECT   MAX( DISTINCT sq1_alias1.col_int_key ) AS sq1_field1 FROM ( cc AS sq1_alias1 , bb AS sq1_alias2 ) WHERE sq1_alias2.pk <> alias1.pk AND sq1_alias2.pk <= alias1.col_int_key ) AS field1 , COUNT(  alias1.col_varchar_key ) AS field2 FROM ( bb AS alias1 LEFT OUTER JOIN ( ( cc AS alias2 INNER JOIN cc AS alias3 ON (alias3.col_int_key = alias2.col_int_key  ) ) ) ON (alias3.col_varchar_nokey = alias2.col_varchar_key  ) ) WHERE (   EXISTS ( (SELECT   sq2_alias1.col_varchar_nokey AS sq2_field1 FROM c AS sq2_alias1 WHERE  ( sq2_alias1.col_varchar_nokey , sq2_alias1.col_varchar_nokey )  IN (SELECT   c_sq1_alias1.col_varchar_key AS c_sq1_field1 , MAX(  c_sq1_alias1.col_varchar_nokey ) AS c_sq1_field2 FROM cc AS c_sq1_alias1    ) GROUP BY sq2_field1 ) ) ) OR ( alias1.col_int_key > 51 AND alias1.col_int_key < ( 51 + 209 ) OR alias1.col_int_key > 51 AND alias1.col_int_key < ( 51 + 159 ) ) OR  (SELECT  SQL_SMALL_RESULT MIN(  sq3_alias1.pk ) AS sq3_field1 FROM ( c AS sq3_alias1 RIGHT OUTER JOIN c AS sq3_alias2 ON (sq3_alias2.col_int_key = sq3_alias1.pk  ) )  ) IS NOT NULL  GROUP BY field1 HAVING field2 <> 5 ORDER BY alias1.col_time_key ASC , field2  , CONCAT( alias2.col_varchar_key, alias3.col_varchar_key )
 STATUS: EQUAL
 QUERY: SELECT     (SELECT 5 FROM dummy ) AS field1 FROM ( b AS table1 INNER JOIN ( ( ( SELECT   subquery2_t1.* FROM ( cc AS subquery2_t1 INNER JOIN cc AS subquery2_t2 ON (subquery2_t2.col_varchar_key = subquery2_t1.col_varchar_key  ) )  ) AS table2 LEFT  JOIN cc AS table3 ON (table3.col_int_key = table2.pk  ) ) ) ON (table3.col_varchar_key = table2.col_varchar_key  ) ) WHERE (   EXISTS ( (SELECT   subquery3_t2.col_varchar_nokey AS subquery3_field1 FROM ( d AS subquery3_t1 INNER JOIN bb AS subquery3_t2 ON (subquery3_t2.col_int_key = subquery3_t1.pk  ) )   ) ) ) OR ( table1.col_varchar_key = 'w' AND table1.col_varchar_key = table3.col_varchar_key ) ORDER BY table1.col_varchar_key DESC /*+JavaDB:Postgres: NULLS LAST */ , field1 /*+JavaDB:Postgres: NULLS FIRST */ LIMIT 10 OFFSET 6
-STATUS:NOT_EQUAL
-SYNC_POINT:before_reset_query_plan
-Diff: EXPLAIN NORMAL vs OTHER
-@@ -3,3 +3,3 @@
- 1	PRIMARY	<derived3>	NULL	ALL	NULL	NULL	NULL	NULL	20	100.00	Using join buffer (Block Nested Loop)
--1	PRIMARY	table3	NULL	ref	col_int_key,col_varchar_key	col_int_key	5	table2.pk	1	5.00	Using where
-+1	PRIMARY	table3	NULL	ref	col_int_key,col_varchar_key	col_int_key	5	table2.pk	1	100.00	Using where
- 4	SUBQUERY	subquery3_t2	NULL	ALL	col_int_key	NULL	NULL	NULL	1	100.00	Using where
+STATUS: EQUAL
 QUERY: SELECT    alias1.col_time_key AS field1 , alias2.col_varchar_key AS field2 FROM ( b AS alias1 , ( SELECT   sq1_alias1.* FROM cc AS sq1_alias1  ) AS alias2 , c AS alias3 ) WHERE (   EXISTS ( (SELECT 5 FROM DUAL ) ) ) OR ( alias1.col_varchar_key  IN ('e' , 'USA') OR alias1.col_varchar_key  IN ('USA') )  GROUP BY field1, field2 HAVING field2 > 'e' 
 STATUS: EQUAL
 QUERY: SELECT     (SELECT 'z' FROM dummy ) AS field1 , table1.pk AS field2 , table1.col_time_key AS field3 , table2.col_datetime_key AS field4 , table1.col_datetime_key AS field5 , table2.pk AS field6 FROM ( ( SELECT   subquery2_t1.* FROM ( c AS subquery2_t1 INNER JOIN cc AS subquery2_t2 ON (subquery2_t2.pk = subquery2_t1.pk  ) )  ) AS table1 INNER JOIN ( SELECT   subquery3_t1.* FROM ( b AS subquery3_t1 LEFT  JOIN c AS subquery3_t2 ON (subquery3_t2.pk = subquery3_t1.pk  ) )  ) AS table2 ON (table2.pk = table1.col_int_key  ) ) WHERE (   EXISTS ( (SELECT   subquery4_t1.col_varchar_key AS subquery4_field1 FROM ( cc AS subquery4_t1 LEFT OUTER JOIN c AS subquery4_t2 ON (subquery4_t2.pk = subquery4_t1.col_int_key  ) )   ) ) ) OR ( table1.pk = 141 AND table1.pk = 79 ) ORDER BY table1.col_datetime_key DESC /*+JavaDB:Postgres: NULLS LAST */

mysql-test/r/explain_for_connection_rqg_trad.result

@@ -11,7 +11,7 @@ MIN(c)
 0
 EXPLAIN SELECT MIN(c) FROM t1 GROUP BY b;
 id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
-1	SIMPLE	t1	NULL	range	b	b	263	NULL	3	100.00	Using index for group-by
+1	SIMPLE	t1	NULL	range	b	b	263	NULL	2	100.00	Using index for group-by
 Warnings:
 Note	1003	/* select#1 */ select min(`test`.`t1`.`c`) AS `MIN(c)` from `test`.`t1` group by `test`.`t1`.`b`
 DROP TABLE t1;