Bug#36625968: Long query execution times due to hypergraph optimizer choosing un-indexed NLJ

Query plans with nested loop joins are risky if the inner side of the
join cannot use indexes to limit the number of rows to read. Even
though they are estimated to be slightly cheaper than the
corresponding hash join in some cases, if the row estimates are a
little off, the actual cost of executing the join can multiply many
times. The performance of hash joins degrades more gracefully when the
estimates are off.

In order to avoid these risky plans, the hypergraph optimizer is
changed so that it proposes nested loop joins only if it can push the
join condition to an index lookup on the right hand side of the join.
If it doesn't have an index to use for the join condition, only the
corresponding hash join is proposed.

Exceptions to this are made for cases in which a hash join cannot be
used (mainly cases involving full-text search, lateral derived tables
or table functions). Also, a nested loop join is allowed if either of
the sides is known to return at most one row, in which case a nested
loop join should be at least as good as a hash join, and there's no
risk of row estimates being far off.

Change-Id: Iab22e296a5042a7ac1cce80bacc265c7a2c0b78e

Author: kahatlen

mysql-test/common/binlog/mysqlbinlog_row.inc

@@ -470,12 +470,12 @@ SET sql_mode = default;
 # Check multi-table update
 #
 CREATE TABLE t1 (a int NOT NULL DEFAULT 0, b int NOT NULL DEFAULT 0);
-CREATE TABLE t2 (a int NOT NULL DEFAULT 0, b int NOT NULL DEFAULT 0);
+CREATE TABLE t2 (a int NOT NULL DEFAULT 0, b int NOT NULL DEFAULT 0, UNIQUE (b));
 INSERT INTO t1 SET a=1;
 INSERT INTO t1 SET b=1;
 INSERT INTO t2 SET a=1;
 INSERT INTO t2 SET b=1;
-UPDATE t1, t2 SET t1.a=10, t2.a=20;
+UPDATE t1, t2 SET t1.a=10, t2.a=20 WHERE t1.b = t2.b;
 DROP TABLE t1,t2;
 
 

mysql-test/common/rpl/rpl_multi_update3.test

@@ -36,11 +36,13 @@ UPDATE t2, (SELECT a FROM t1 ORDER BY a) AS t SET t2.b = t.a+5 ;
 SET @@optimizer_switch=@optimizer_switch_saved;
 
 SELECT * FROM t1 ORDER BY a;
+--skip_if_hypergraph  # Non-deterministic multi-table update of t2 above.
 SELECT * FROM t2 ORDER BY a;
 
 --source include/rpl/sync_to_replica.inc
 connection slave;
 SELECT * FROM t1 ORDER BY a;
+--skip_if_hypergraph  # Non-deterministic multi-table update of t2 above.
 SELECT * FROM t2 ORDER BY a;
 
 connection master;

mysql-test/include/ctype_utf8mb4.inc

@@ -931,6 +931,7 @@ eval CREATE TABLE t2 (
 
 INSERT INTO t2 VALUES ('1234567890',2,'2005-05-24 13:53:25');
 
+--sorted_result
 SELECT content, t2.msisdn FROM t1, t2 WHERE t1.msisdn = '1234567890';
 
 DROP TABLE t1,t2;

mysql-test/include/select.inc

@@ -4220,6 +4220,7 @@ SELECT table4 . `time_nokey` AS field1 FROM
             table1 . `int_nokey` != 'f')
   GROUP BY field1  ORDER BY field1 , field1;
 
+--sorted_result
 SELECT table1 .`time_key` field2  FROM B table1  LEFT JOIN  BB JOIN A table5 ON table5 .`date_nokey`  ON table5 .`int_nokey` GROUP  BY field2;
 --enable_warnings
 

mysql-test/include/subquery.inc

@@ -2151,6 +2151,7 @@ CREATE TABLE t3 ( c3 integer );
 INSERT INTO t3 VALUES ( 7 );
 INSERT INTO t3 VALUES ( 8 );
 
+--sorted_result
 SELECT c1,c2 FROM t1 LEFT JOIN t2 ON c1 = c2
   WHERE EXISTS (SELECT c3 FROM t3 WHERE c2 IS NULL );
 

mysql-test/include/subquery_sj.inc

@@ -661,6 +661,7 @@ INSERT INTO t3 VALUES (22,'come','h'),
  (24,'lpjdzvkp','ababa'),
  (25,'d','GGDD');
 
+--sorted_result
 SELECT * FROM t1 WHERE (field1, field2) IN (
   SELECT table1.col_varchar_255_utf8_key AS field1,
          table1.col_varchar_255_utf8_key AS field2
@@ -2905,6 +2906,7 @@ INSERT INTO t1 VALUES(10),(20);
 create table t2 select * from t1;
 create table t3 select * from t1;
 
+--sorted_result
 SELECT *
 FROM t1
 WHERE 1 IN(SELECT 1
@@ -4190,6 +4192,7 @@ SELECT * FROM t2
 WHERE f2 IN (SELECT t1.f1
              FROM t1 LEFT OUTER JOIN (t2 AS b1 JOIN t2 AS b2 ON TRUE) ON TRUE);
 eval EXPLAIN $query;
+--sorted_result
 eval $query;
 
 DROP TABLE t1, t2;
@@ -4336,7 +4339,14 @@ WHERE col_varchar_10_latin1_key IN (
     WHERE alias1.pk AND alias1.pk < 3 OR alias1.pk AND alias3.pk);
 ANALYZE TABLE t1, t2, t3;
 eval EXPLAIN $query;
+if ($HYPERGRAPH_TEST) {
+  # Different warnings with hypergraph than with old optimizer.
+  --disable_warnings
+}
 eval $query;
+if ($HYPERGRAPH_TEST) {
+  --enable_warnings
+}
 
 DROP TABLE t1, t2, t3;
 
@@ -5129,6 +5139,7 @@ eval $query;
 ALTER TABLE t1 ENGINE=Innodb;
 ALTER TABLE t2 ENGINE=Innodb;
 
+--sorted_result
 eval $query;
 
 DROP TABLE t1, t2;

mysql-test/include/with_recursive.inc

@@ -338,6 +338,14 @@ select 1 from t1 straight_join qn
 )
 select * from qn;
 
+WITH RECURSIVE qn(n) AS
+(
+SELECT 1 FROM t1
+UNION ALL
+SELECT n+1 FROM qn STRAIGHT_JOIN t1 WHERE n < 3
+)
+SELECT COUNT(*) FROM qn;
+
 --error ER_CTE_RECURSIVE_FORBIDDEN_JOIN_ORDER
 WITH RECURSIVE qn AS
 (

mysql-test/r/ctype_utf8mb4_heap.result

@@ -1138,8 +1138,8 @@ UNIQUE KEY `PK_user` (`msisdn`)
 INSERT INTO t2 VALUES ('1234567890',2,'2005-05-24 13:53:25');
 SELECT content, t2.msisdn FROM t1, t2 WHERE t1.msisdn = '1234567890';
 content	msisdn
-ERR Имри.Афимим.Аеимимримдмримрмрирор имримримримр имридм ирбднримрфмририримрфмфмим.Ад.Д имдимримрад.Адимримримрмдиримримримр м.Дадимфшьмримд им.Адимимрн имадми	1234567890
 11 g	1234567890
+ERR Имри.Афимим.Аеимимримдмримрмрирор имримримримр имридм ирбднримрфмририримрфмфмим.Ад.Д имдимримрад.Адимримримрмдиримримримр м.Дадимфшьмримд им.Адимимрн имадми	1234567890
 DROP TABLE t1,t2;
 SET sql_mode = default;
 SET sql_mode = 'NO_ENGINE_SUBSTITUTION';

mysql-test/r/ctype_utf8mb4_innodb.result

@@ -1205,8 +1205,8 @@ UNIQUE KEY `PK_user` (`msisdn`)
 INSERT INTO t2 VALUES ('1234567890',2,'2005-05-24 13:53:25');
 SELECT content, t2.msisdn FROM t1, t2 WHERE t1.msisdn = '1234567890';
 content	msisdn
-ERR Имри.Афимим.Аеимимримдмримрмрирор имримримримр имридм ирбднримрфмририримрфмфмим.Ад.Д имдимримрад.Адимримримрмдиримримримр м.Дадимфшьмримд им.Адимимрн имадми	1234567890
 11 g	1234567890
+ERR Имри.Афимим.Аеимимримдмримрмрирор имримримримр имридм ирбднримрфмририримрфмфмим.Ад.Д имдимримрад.Адимримримрмдиримримримр м.Дадимфшьмримд им.Адимимрн имадми	1234567890
 DROP TABLE t1,t2;
 SET sql_mode = default;
 SET sql_mode = 'NO_ENGINE_SUBSTITUTION';

mysql-test/r/ctype_utf8mb4_myisam.result

@@ -1205,8 +1205,8 @@ UNIQUE KEY `PK_user` (`msisdn`)
 INSERT INTO t2 VALUES ('1234567890',2,'2005-05-24 13:53:25');
 SELECT content, t2.msisdn FROM t1, t2 WHERE t1.msisdn = '1234567890';
 content	msisdn
-ERR Имри.Афимим.Аеимимримдмримрмрирор имримримримр имридм ирбднримрфмририримрфмфмим.Ад.Д имдимримрад.Адимримримрмдиримримримр м.Дадимфшьмримд им.Адимимрн имадми	1234567890
 11 g	1234567890
+ERR Имри.Афимим.Аеимимримдмримрмрирор имримримримр имридм ирбднримрфмририримрфмфмим.Ад.Д имдимримрад.Адимримримрмдиримримримр м.Дадимфшьмримд им.Адимимрн имадми	1234567890
 DROP TABLE t1,t2;
 SET sql_mode = default;
 SET sql_mode = 'NO_ENGINE_SUBSTITUTION';