1.1.1 PostgreSQL
1.1.2.1 S1语句
查看查询执行计划，子查询被优化（采用半连接）。

postgres=# EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT b1 FROM t1 WHERE a3=a1);

QUERY PLAN

------------------------------------------------------------------

Hash Semi Join (cost=55.90..114.35 rows=2040 width=12)

Hash Cond: (t3.a3 = t1.a1)

-> Seq Scan on t3 (cost=0.00..30.40 rows=2040 width=12)

-> Hash (cost=30.40..30.40 rows=2040 width=4)

-> Seq Scan on t1 (cost=0.00..30.40 rows=2040 width=4)

(5 行记录)

1.1.2.2 S2语句
查看查询执行计划，子查询被优化（采用半连接）。

postgres=# EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT a1 FROM t1 WHERE a3=a1);

QUERY PLAN

------------------------------------------------------------------

Hash Semi Join (cost=55.90..114.35 rows=2040 width=12)

Hash Cond: (t3.a3 = t1.a1)

-> Seq Scan on t3 (cost=0.00..30.40 rows=2040 width=12)

-> Hash (cost=30.40..30.40 rows=2040 width=4)

-> Seq Scan on t1 (cost=0.00..30.40 rows=2040 width=4)

(5 行记录)

1.1.2.3 S3语句
查看查询执行计划，子查询被优化（采用半连接）。

postgres=# EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT id1 FROM t1 WHERE a3=a1)

QUERY PLAN

------------------------------------------------------------------

Hash Semi Join (cost=55.90..114.35 rows=2040 width=12)

Hash Cond: (t3.a3 = t1.a1)

-> Seq Scan on t3 (cost=0.00..30.40 rows=2040 width=12)

-> Hash (cost=30.40..30.40 rows=2040 width=4)

-> Seq Scan on t1 (cost=0.00..30.40 rows=2040 width=4)

(5 行记录)

1.1.2.4 S4语句
查看查询执行计划，子查询被优化。

postgres=# EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT b1 FROM t1 WHERE a1>2);

QUERY PLAN

-------------------------------------------------------------

Result (cost=0.05..30.45 rows=2040 width=12)

One-Time Filter: $0

InitPlan 1 (returns $0) //子查询被优化，只被执行一次，类似其他数据库如MySQL对子查询的物化优化

-> Seq Scan on t1 (cost=0.00..35.50 rows=680 width=0)

Filter: (a1 > 2)

-> Seq Scan on t3 (cost=0.00..30.40 rows=2040 width=12)

(6 行记录)

1.1.2.5 S5语句
查看查询执行计划，子查询被优化。

postgres=# EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT a1 FROM t1 WHERE b3=b1);

QUERY PLAN

------------------------------------------------------------------------

Hash Join (cost=40.00..98.45 rows=1020 width=12)

Hash Cond: (t3.b3 = t1.b1)

-> Seq Scan on t3 (cost=0.00..30.40 rows=2040 width=12)

-> Hash (cost=37.50..37.50 rows=200 width=4)

-> HashAggregate (cost=35.50..37.50 rows=200 width=4) //在t1表上执行了一个hash聚集操作

Group Key: t1.b1

-> Seq Scan on t1 (cost=0.00..30.40 rows=2040 width=4)

(7 行记录)

1.1.2 MySQL
1.1.3.1 S1语句
查看查询执行计划，子查询没有被优化。

mysql> EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT b1 FROM t1 WHERE a3=a1);

+----+--------------------+-------+------+------+-------------+

| id | select_type | table | type | key | Extra |

+----+--------------------+-------+------+------+-------------+

| 1 | PRIMARY | t3 | ALL | NULL | Using where |

| 2 | DEPENDENT SUBQUERY | t1 | ref | a1 | NULL |

+----+--------------------+-------+------+------+-------------+

2 rows in set (0.02 sec)

1.1.3.2 S2语句
查看查询执行计划，子查询没有被优化。

mysql> EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT a1 FROM t1 WHERE a3=a1);

+----+--------------------+-------+------+------+-------------+

| id | select_type | table | type | key | Extra |

+----+--------------------+-------+------+------+-------------+

| 1 | PRIMARY | t3 | ALL | NULL | Using where |

| 2 | DEPENDENT SUBQUERY | t1 | ref | a1 | Using index |

+----+--------------------+-------+------+------+-------------+

2 rows in set (0.00 sec)

1.1.3.3 S3语句
查看查询执行计划，子查询没有被优化。

mysql> EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT id1 FROM t1 WHERE a3=a1);

+----+--------------------+-------+------+------+-------------+

| id | select_type | table | type | key | Extra |

+----+--------------------+-------+------+------+-------------+

| 1 | PRIMARY | t3 | ALL | NULL | Using where |

| 2 | DEPENDENT SUBQUERY | t1 | ref | a1 | Using index |

+----+--------------------+-------+------+------+-------------+

2 rows in set (0.00 sec)

1.1.3.4 S4语句
查看查询执行计划，子查询没有被优化。

mysql> EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT b1 FROM t1 WHERE a1>2);

+----+-------------+-------+-------+------+-----------------------+

| id | select_type | table | type | key | Extra |

+----+-------------+-------+-------+------+-----------------------+

| 1 | PRIMARY | t3 | ALL | NULL | NULL |

| 2 | SUBQUERY | t1 | range | a1 | Using index condition |

+----+-------------+-------+-------+------+-----------------------+

2 rows in set (0.01 sec)

1.1.3.5 S5语句
查看查询执行计划，子查询没有被优化。

mysql> EXPLAIN SELECT * FROM t3 WHERE EXISTS (SELECT a1 FROM t1 WHERE b3=b1);

+----+--------------------+-------+------+------+-------------+

| id | select_type | table | type | key | Extra |

+----+--------------------+-------+------+------+-------------+

| 1 | PRIMARY | t3 | ALL | NULL | Using where |

| 2 | DEPENDENT SUBQUERY | t1 | ALL | NULL | Using where |

+----+--------------------+-------+------+------+-------------+

2 rows in set (0.00 sec)

1.1.4 对比
我们把原始的SQL语句复制一份，目的是查阅方便。



S1: SELECT * FROM t3 WHERE b3 EXISTS (SELECT b1 FROM t1 WHERE a3=a1);

S2: SELECT * FROM t3 WHERE b3 EXISTS (SELECT a1 FROM t1 WHERE a3=a1);

S3: SELECT * FROM t3 WHERE b3 EXISTS (SELECT id1 FROM t1 WHERE a3=a1);

S4: SELECT * FROM t3 WHERE id3 EXISTS (SELECT b1 FROM t1 WHERE a1>2);

S5: SELECT * FROM t3 WHERE EXISTS (SELECT a1 FROM t1 WHERE b3=b1);



然后对比如下:

SQL语句

语句特点

ToprowDB

PostgreSQL

MySQL

S1

相关子查询，目标列为普通列，条件为索引键

Join优化

Semi Join优化

没有被优化

S2

相关子查询，目标列为键，条件为索引键

Join优化

Semi Join优化

没有被优化

S3

相关子查询，目标列为主键列，条件为索引键

Join优化

Semi Join优化

没有被优化

S4

非相关子查询，目标列为普通列，条件为索引键

没有被优化

类似物化优化

没有被优化

S5

相关子查询，目标列为键，条件为普通列

Semi Join优化

Join优化

没有被优化

分析：

q 从整体上看，对于EXISTS类型的子查询，PostgreSQL的优化能力强，MySQL差，ToprowDB接近PostgreSQL

q 对于EXISTS类型的子查询，PostgreSQL基本上是利用Semi Join来进行优化的；多数数据库的优化器都会有类似的方式来对EXISTS类型的子查询进行优化

q 但对于ToprowDB却采取的是普通Join，这是为什么？其实，ToprowDB的优化器也是先进行了Semi Join式优化，然后根据性（索引如S1中的a3=a1）进一步判断，在明知不会产生重复元组的情况下，把Semi Join进一步优化为Join。这点体现了ToprowDB优化器的先进性

q 对比S1和S5（关键在于WHERE条件选元组），PostgreSQL把S5优化为Join操作，这在PostgreSQL中属于特例，PostgreSQL在经过逻辑优化（转为Semi Join属于逻辑优化）之后，利用了代价估算模型来进一步判断是使用排序还是Hash聚集操作来优化EXISTS语义，经过选择，认定Hash聚集代价小于排序，因此选定了HashAggregate 作为其查询执行计划
————————————————
版权声明：本文为CSDN博主「那海蓝蓝」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/fly2nn/article/details/61924793