1、嵌套循环-nested loop
当两个结果集做合并，需要找出两个结果集中都满足同一个或多个条件的数据时，就需要使用合适的方法来找出满足条件的行。
嵌套循环是集合连接中常见的一种连接方法。通常分为外部循环和内部循环，外部循环的集合称为驱动集合，内部循环的集合称为被驱动集合。
嵌套循环伪代码:
for each row in t1 matching range {
for each row in t2 matching reference key {
if row satisfies join conditions, send to client
}
}
工作过程:
- 优化器确定驱动集合并将其指定为外部循环。
外部循环生成一组用于驱动连接条件的行。集合可以是使用索引扫描、全表扫描或任何其他生成行的操作访问的表。
内部循环的迭代次数取决于外部循环中检索到的行数。例如，如果从外部表中检索了10行，则数据库必须在内部表中执行10次查找。如果从外部表检索了10,000,000行，那么数据库必须在内部表中执行10,000,000次查找。 - 优化器将另一个集合指定为内部循环。
- 对于每一个来自客户端的取回请求，基本流程如下:
1.从外层行源获取一行
2.探测内部行源以查找与谓词条件匹配的行
3.重复上述步骤，直到获取请求获得所有行
示例：
mysql> show indexes from emps;
+-------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+---------+------------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment | Visible | Expression |
+-------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+---------+------------+
| emps | 0 | PRIMARY | 1 | id | A | 299246 | NULL | NULL | | BTREE | | | YES | NULL |
+-------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+---------+------------+
1 row in set (0.00 sec)
mysql> show indexes from salaries;
+----------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+---------+------------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment | Visible | Expression |
+----------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+---------+------------+
| salaries | 0 | PRIMARY | 1 | emp_no | A | 293263 | NULL | NULL | | BTREE | | | YES | NULL |
| salaries | 0 | PRIMARY | 2 | from_date | A | 2838426 | NULL | NULL | | BTREE | | | YES | NULL |
+----------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+---------+------------+
2 rows in set (0.02 sec)
mysql> desc format=tree select * from emps e join salaries s where e.id'2001-06-22';
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| EXPLAIN |
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| -> Nested loop inner join (cost=22.8 rows=32.3)
-> Filter: (e.id Index range scan on e using PRIMARY over (id Filter: (s.to_date > DATE'2001-06-22') (cost=1.04 rows=3.23)
-> Index lookup on s using PRIMARY (emp_no=e.emp_no) (cost=1.04 rows=9.68)
|
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.01 sec)
#上述查询的工作过程：
#1.表emps通过索引范围扫描获取满足条件id desc analyze select * from emps e join salaries2 s where e.id'2001-06-22';
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| EXPLAIN |
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| -> Nested loop inner join (cost=4.75e+6 rows=945810) (actual time=0.215..5971 rows=18 loops=1)
-> Filter: (e.id Index range scan on e using PRIMARY over (id Filter: ((s.emp_no = e.emp_no) and (s.to_date > DATE'2001-06-22')) (cost=191778 rows=94581) (actual time=60.8..597 rows=1.8 loops=10)
-> Table scan on s (cost=191778 rows=2.84e+6) (actual time=0.0233..419 rows=2.84e+6 loops=10)
|
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (5.97 sec)
#可以看到如果被驱动的表上连接字段没有索引，被驱动表会走全表扫描。
#且全表扫描的次数等于驱动集合的记录数。
总结：
- 嵌套循环通常选择结果集小的那个表作为驱动表。
- 嵌套循环的被驱动集合的表上连接字段一定要有索引，否则会执行全表扫描。
- 被驱动集合的循环次数等于驱动集合的记录数。
- 嵌套循环通常适合于一大一小两个集合，且被驱动的集合的表的连接字段上有索引。
2、块嵌套循环-Block Nested-Loop
块嵌套循环是嵌套循环的一个变体。块嵌套循环在执行外层循环时，并不会取一行记录就去执行内存循环，而是取一行记录就会存放到join_buffer中，当join_buffer满了才开始执行内层循环。
内层循环取一条记录就会跟外层循环存放在join_buffer中的每一条记录做比较。当做完一轮比较后，清空join_buffer，执行外层循环的下一批操作，直到所有的记录都join完。
块嵌套循环大大减少了内层循环的次数。
伪代码:
for each row in t1 matching range {
store used columns from t1 in join buffer
if buffer is full {
for each row in t2 matching reference key {
if row satisfies join conditions, send to client
}
empty join buffer
}
}
块嵌套循环的执行计划如下：
mysql> desc select * from emps e join sals s where e.id'2001-06-22';
+----+-------------+-------+------------+-------+---------------+---------+---------+------+---------+----------+----------------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+---------+----------+----------------------------------------------------+
| 1 | SIMPLE | e | NULL | range | PRIMARY | PRIMARY | 4 | NULL | 10 | 100.00 | Using where |
| 1 | SIMPLE | s | NULL | ALL | NULL | NULL | NULL | NULL | 2837430 | 3.33 | Using where; Using join buffer (Block Nested Loop) |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+---------+----------+----------------------------------------------------+
2 rows in set, 1 warning (0.00 sec)
join_buffer_size控制缓冲区的大小。每个连接都会分配一个join_buffer，一个查询中可能会分配多个join_buffer。
8.0.18以后块嵌套循环被hash join取代，之前用于块嵌套循环的场景会使用hash join。
块嵌套循环可以由优化器参数block_nested_loop控制开启和关闭。
mysql> set optimizer_switch='block_nested_loop=on|off';
