不会看 Explain执行计划,劝你简历别写熟悉 SQL优化

个人博客地址:http://www.chengxy-nds.top,技术资料自取

昨天中午在食堂,和部门的技术大牛们坐在一桌吃饭,作为一个卑微技术渣仔默默的吃着饭,听大佬们高谈阔论,研究各种高端技术,我TM也想说话可实在插不上嘴。

聊着聊着突然说到他上午面试了一个工作6年的程序员,表情挺复杂,他说:我看他简历写着熟悉SQL语句调优,就问了下 Explain 执行计划怎么看?结果这老哥一问三不知,工作6年这么基础的东西都不了解!

感受到了大佬的王之鄙视,回到工位我就开始默默写这个,哎~ 我TM也不太懂 Explain ,老哥你这是针对我啊!哭唧唧~
不会看 Explain执行计划,劝你简历别写熟悉 SQL优化_第1张图片

Explain有什么用

ExplainSQL语句一起使用时,MySQL 会显示来自优化器关于SQL执行的信息。也就是说,MySQL解释了它将如何处理该语句,包括如何连接表以及什么顺序连接表等。

  • 表的加载顺序
  • sql 的查询类型
  • 可能用到哪些索引,哪些索引又被实际使用
  • 表与表之间的引用关系
  • 一个表中有多少行被优化器查询

Explain有哪些信息

Explain 执行计划包含字段信息如下:分别是 idselect_typetablepartitionstypepossible_keyskeykey_lenrefrowsfilteredExtra 12个字段。

下边我们会结合具体的SQL示例,详细的解读每个字段以及每个字段中不同参数的含义,以下所有示例数据库版本为 MySQL.5.7.17

mysql> select version() from dual;
+------------+
| version()  |
+------------+
| 5.7.17-log |
+------------+

我们创建三张表 onetwothree,表之间的关系 one.two_id = two.two_id AND two.three_id = three.three_id

Explain执行计划详解

一、id

id: :表示查询中执行select子句或者操作表的顺序,id的值越大,代表优先级越高,越先执行id大致会出现 3种情况:

1、id相同

看到三条记录的id都相同,可以理解成这三个表为一组,具有同样的优先级,执行顺序由上而下,具体顺序由优化器决定。

mysql> EXPLAIN SELECT * FROM one o,two t, three r WHERE o.two_id = t.two_id AND t.three_id = r.three_id;
+----+-------------+-------+------------+--------+---------------+---------+---------+----------------------+------+----------+----------------------------------------------------+
| id | select_type | table | partitions | type   | possible_keys | key     | key_len | ref                  | rows | filtered | Extra                                              |
+----+-------------+-------+------------+--------+---------------+---------+---------+----------------------+------+----------+----------------------------------------------------+
|  1 | SIMPLE      | o     | NULL       | ALL    | NULL          | NULL    | NULL    | NULL                 |    2 |      100 | NULL                                               |
|  1 | SIMPLE      | t     | NULL       | ALL    | PRIMARY       | NULL    | NULL    | NULL                 |    2 |       50 | Using where; Using join buffer (Block Nested Loop) |
|  1 | SIMPLE      | r     | NULL       | eq_ref | PRIMARY       | PRIMARY | 4       | xin-slave.t.three_id |    1 |      100 | NULL                                               |
+----+-------------+-------+------------+--------+---------------+---------+---------+----------------------+------+----------+----------------------------------------------------+
2、id不同

如果我们的 SQL 中存在子查询,那么 id的序号会递增,id值越大优先级越高,越先被执行 。当三个表依次嵌套,发现最里层的子查询 id最大,最先执行。

mysql> EXPLAIN select * from one o where o.two_id = (select t.two_id from two t where t.three_id = (select r.three_id  from three r where r.three_name='我是第三表2'));
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | PRIMARY     | o     | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    2 |       50 | Using where |
|  2 | SUBQUERY    | t     | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    2 |       50 | Using where |
|  3 | SUBQUERY    | r     | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    2 |       50 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
3、以上两种同时存在

将上边的 SQL 稍微修改一下,增加一个子查询,发现 id的以上两种同时存在。相同id划分为一组,这样就有三个组,同组的从上往下顺序执行,不同组 id值越大,优先级越高,越先执行。

mysql>  EXPLAIN select * from one o where o.two_id = (select t.two_id from two t where t.three_id = (select r.three_id  from three r where r.three_name='我是第三表2')) AND o.one_id in(select one_id from one where o.one_name="我是第一表2");
+----+-------------+-------+------------+--------+---------------+---------+---------+--------------------+------+----------+-------------+
| id | select_type | table | partitions | type   | possible_keys | key     | key_len | ref                | rows | filtered | Extra       |
+----+-------------+-------+------------+--------+---------------+---------+---------+--------------------+------+----------+-------------+
|  1 | PRIMARY     | o     | NULL       | ALL    | PRIMARY       | NULL    | NULL    | NULL               |    2 |       50 | Using where |
|  1 | PRIMARY     | one   | NULL       | eq_ref | PRIMARY       | PRIMARY | 4       | xin-slave.o.one_id |    1 |      100 | Using index |
|  2 | SUBQUERY    | t     | NULL       | ALL    | NULL          | NULL    | NULL    | NULL               |    2 |       50 | Using where |
|  3 | SUBQUERY    | r     | NULL       | ALL    | NULL          | NULL    | NULL    | NULL               |    2 |       50 | Using where |
+----+-------------+-------+------------+--------+---------------+---------+---------+--------------------+------+----------+-------------+

二、select_type

select_type:表示 select 查询的类型,主要是用于区分各种复杂的查询,例如:普通查询联合查询子查询等。

1、SIMPLE

SIMPLE:表示最简单的 select 查询语句,也就是在查询中不包含子查询或者 union交并差集等操作。

2、PRIMARY

PRIMARY:当查询语句中包含任何复杂的子部分,最外层查询则被标记为PRIMARY

3、SUBQUERY

SUBQUERY:当 selectwhere 列表中包含了子查询,该子查询被标记为:SUBQUERY

4、DERIVED

DERIVED:表示包含在from子句中的子查询的select,在我们的 from 列表中包含的子查询会被标记为derived

5、UNION

UNION:如果union后边又出现的select 语句,则会被标记为union;若 union 包含在 from 子句的子查询中,外层 select 将被标记为 derived

6、UNION RESULT

UNION RESULT:代表从union的临时表中读取数据,而table列的表示用第一个和第四个select的结果进行union操作。

mysql> EXPLAIN select t.two_name, ( select one.one_id from one) o from (select two_id,two_name from two where two_name ='') t  union (select r.three_name,r.three_id from three r);

+------+--------------+------------+------------+-------+---------------+---------+---------+------+------+----------+-----------------+
| id   | select_type  | table      | partitions | type  | possible_keys | key     | key_len | ref  | rows | filtered | Extra           |
+------+--------------+------------+------------+-------+---------------+---------+---------+------+------+----------+-----------------+
|    1 | PRIMARY      | two        | NULL       | ALL   | NULL          | NULL    | NULL    | NULL |    2 |       50 | Using where     |
|    2 | SUBQUERY     | one        | NULL       | index | NULL          | PRIMARY | 4       | NULL |    2 |      100 | Using index     |
|    4 | UNION        | r          | NULL       | ALL   | NULL          | NULL    | NULL    | NULL |    2 |      100 | NULL            |
| NULL | UNION RESULT | <union1,4> | NULL       | ALL   | NULL          | NULL    | NULL    | NULL | NULL | NULL     | Using temporary |
+------+--------------+------------+------------+-------+---------------+---------+---------+------+------+----------+-----------------+

三、table

查询的表名,并不一定是真实存在的表,有别名显示别名,也可能为临时表,例如上边的DERIVED等。

四、partitions

查询时匹配到的分区信息,对于非分区表值为NULL,当查询的是分区表时,partitions显示分区表命中的分区情况。

+----+-------------+----------------+---------------------------------+-------+---------------+---------+---------+------+------+----------+-------------+
| id | select_type | table          | partitions                      | type  | possible_keys | key     | key_len | ref  | rows | filtered | Extra       |
+----+-------------+----------------+---------------------------------+-------+---------------+---------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | one            | p201801,p201802,p201803,p300012 | index | NULL          | PRIMARY | 9       | NULL |    3 |      100 | Using index |
+----+-------------+----------------+---------------------------------+-------+---------------+---------+---------+------+------+----------+-------------+

五、type

type:查询使用了何种类型,它在 SQL优化中是一个非常重要的指标,以下性能从好到坏依次是:system > const > eq_ref > ref > ref_or_null > index_merge > unique_subquery > index_subquery > range > index > ALL

1、system

system: 当表仅有一行记录时(系统表),数据量很少,往往不需要进行磁盘IO,速度非常快。

2、const

const:表示查询时命中 primary key 主键或者 unique 唯一索引,或者被连接的部分是一个常量(const)值。这类扫描效率极高,返回数据量少,速度非常快。

mysql> EXPLAIN SELECT * from three where three_id=1;
+----+-------------+-------+------------+-------+---------------+---------+---------+-------+------+----------+-------+
| id | select_type | table | partitions | type  | possible_keys | key     | key_len | ref   | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+---------+---------+-------+------+----------+-------+
|  1 | SIMPLE      | three | NULL       | const | PRIMARY       | PRIMARY | 4       | const |    1 |      100 | NULL  |
+----+-------------+-------+------------+-------+---------------+---------+---------+-------+------+----------+-------+
3、eq_ref

eq_ref:查询时命中主键primary key 或者 unique key索引, type 就是 eq_ref

mysql> EXPLAIN select o.one_name from one o ,two t where o.one_id = t.two_id ; 
+----+-------------+-------+------------+--------+---------------+----------+---------+--------------------+------+----------+-------------+
| id | select_type | table | partitions | type   | possible_keys | key      | key_len | ref                | rows | filtered | Extra       |
+----+-------------+-------+------------+--------+---------------+----------+---------+--------------------+------+----------+-------------+
|  1 | SIMPLE      | o     | NULL       | index  | PRIMARY       | idx_name | 768     | NULL               |    2 |      100 | Using index |
|  1 | SIMPLE      | t     | NULL       | eq_ref | PRIMARY       | PRIMARY  | 4       | xin-slave.o.one_id |    1 |      100 | Using index |
+----+-------------+-------+------------+--------+---------------+----------+---------+--------------------+------+----------+-------------+
4、ref

ref:区别于eq_refref表示使用非唯一性索引,会找到很多个符合条件的行。

mysql> select o.one_id from one o where o.one_name = "xin" ; 
+--------+
| one_id |
+--------+
|      1 |
|      3 |
+--------+
mysql> EXPLAIN select o.one_id from one o where o.one_name = "xin" ; 
+----+-------------+-------+------------+------+---------------+----------+---------+-------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key      | key_len | ref   | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+----------+---------+-------+------+----------+-------------+
|  1 | SIMPLE      | o     | NULL       | ref  | idx_name      | idx_name | 768     | const |    1 |      100 | Using index |
+----+-------------+-------+------------+------+---------------+----------+---------+-------+------+----------+-------------+
5、ref_or_null

ref_or_null:这种连接类型类似于 ref,区别在于 MySQL会额外搜索包含NULL值的行。

mysql> EXPLAIN select o.one_id from one o where o.one_name = "xin" OR o.one_name IS NULL; 
+----+-------------+-------+------------+-------------+---------------+----------+---------+-------+------+----------+--------------------------+
| id | select_type | table | partitions | type        | possible_keys | key      | key_len | ref   | rows | filtered | Extra                    |
+----+-------------+-------+------------+-------------+---------------+----------+---------+-------+------+----------+--------------------------+
|  1 | SIMPLE      | o     | NULL       | ref_or_null | idx_name      | idx_name | 768     | const |    3 |      100 | Using where; Using index |
+----+-------------+-------+------------+-------------+---------------+----------+---------+-------+------+----------+--------------------------+
6、index_merge

index_merge:使用了索引合并优化方法,查询使用了两个以上的索引。

下边示例中同时使用到主键one_id 和 字段one_nameidx_name 索引 。

mysql> EXPLAIN select * from one o where o.one_id >1 and o.one_name ='xin'; 
+----+-------------+-------+------------+-------------+------------------+------------------+---------+------+------+----------+------------------------------------------------+
| id | select_type | table | partitions | type        | possible_keys    | key              | key_len | ref  | rows | filtered | Extra                                          |
+----+-------------+-------+------------+-------------+------------------+------------------+---------+------+------+----------+------------------------------------------------+
|  1 | SIMPLE      | o     | NULL       | index_merge | PRIMARY,idx_name | idx_name,PRIMARY | 772,4   | NULL |    1 |      100 | Using intersect(idx_name,PRIMARY); Using where |
+----+-------------+-------+------------+-------------+------------------+------------------+---------+------+------+----------+------------------------------------------------+
7、unique_subquery

unique_subquery:替换下面的 IN子查询,子查询返回不重复的集合。

value IN (SELECT primary_key FROM single_table WHERE some_expr)
8、index_subquery

index_subquery:区别于unique_subquery,用于非唯一索引,可以返回重复值。

value IN (SELECT key_column FROM single_table WHERE some_expr)
9、range

range:使用索引选择行,仅检索给定范围内的行。简单点说就是针对一个有索引的字段,给定范围检索数据。在where语句中使用 bettween...and<><=in 等条件查询 type 都是 range

举个栗子:three表中three_id为唯一主键,user_id普通字段未建索引。

mysql> EXPLAIN SELECT * from three where three_id BETWEEN 2 AND 3;
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type  | possible_keys | key     | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | three | NULL       | range | PRIMARY       | PRIMARY | 4       | NULL |    1 |      100 | Using where |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+

从结果中看到只有对设置了索引的字段,做范围检索 type 才是 range

mysql> EXPLAIN SELECT * from three where user_id BETWEEN 2 AND 3;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | three | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |    33.33 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
10、index

indexIndexALL 其实都是读全表,区别在于index是遍历索引树读取,而ALL是从硬盘中读取。

下边示例:three_id 为主键,不带 where 条件全表查询 ,type结果为index

mysql> EXPLAIN SELECT three_id from three ;
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type  | possible_keys | key     | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | three | NULL       | index | NULL          | PRIMARY | 4       | NULL |    1 |      100 | Using index |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
11、ALL

ALL:将遍历全表以找到匹配的行,性能最差。

mysql> EXPLAIN SELECT * from two ;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+
|  1 | SIMPLE      | two   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    2 |      100 | NULL  |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+

六、possible_keys

possible_keys:表示在MySQL中通过哪些索引,能让我们在表中找到想要的记录,一旦查询涉及到的某个字段上存在索引,则索引将被列出,但这个索引并不定一会是最终查询数据时所被用到的索引。具体请参考上边的例子。

七、key

key:区别于possible_keys,key是查询中实际使用到的索引,若没有使用索引,显示为NULL。具体请参考上边的例子。

typeindex_merge 时,可能会显示多个索引。

八、key_len

key_len:表示查询用到的索引长度(字节数),原则上长度越短越好 。

  • 单列索引,那么需要将整个索引长度算进去;
  • 多列索引,不是所有列都能用到,需要计算查询中实际用到的列。

注意:key_len只计算where条件中用到的索引长度,而排序和分组即便是用到了索引,也不会计算到key_len中。

九、ref

ref:常见的有:constfuncnull,字段名。

  • 当使用常量等值查询,显示const
  • 当关联查询时,会显示相应关联表的关联字段
  • 如果查询条件使用了表达式函数,或者条件列发生内部隐式转换,可能显示为func
  • 其他情况null

十、rows

rows:以表的统计信息和索引使用情况,估算要找到我们所需的记录,需要读取的行数。

这是评估SQL 性能的一个比较重要的数据,mysql需要扫描的行数,很直观的显示 SQL 性能的好坏,一般情况下 rows 值越小越好。

mysql> EXPLAIN SELECT * from three;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+
|  1 | SIMPLE      | three | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |      100 | NULL  |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+

十一、filtered

filtered 这个是一个百分比的值,表里符合条件的记录数的百分比。简单点说,这个字段表示存储引擎返回的数据在经过过滤后,剩下满足条件的记录数量的比例。

MySQL.5.7版本以前想要显示filtered需要使用explain extended命令。MySQL.5.7后,默认explain直接显示partitionsfiltered的信息。

十二、Extra

Extra :不适合在其他列中显示的信息,Explain 中的很多额外的信息会在 Extra 字段显示。

1、Using index

Using index:我们在相应的 select 操作中使用了覆盖索引,通俗一点讲就是查询的列被索引覆盖,使用到覆盖索引查询速度会非常快,SQl优化中理想的状态。

什么又是覆盖索引?

一条 SQL只需要通过索引就可以返回,我们所需要查询的数据(一个或几个字段),而不必通过二级索引,查到主键之后再通过主键查询整行数据(select * )。

one_id表为主键

mysql> EXPLAIN SELECT one_id from one ;
+----+-------------+-------+------------+-------+---------------+------------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type  | possible_keys | key        | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+-------+---------------+------------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | one   | NULL       | index | NULL          | idx_two_id | 5       | NULL |    3 |      100 | Using index |
+----+-------------+-------+------------+-------+---------------+------------+---------+------+------+----------+-------------+

注意:想要使用到覆盖索引,我们在 select 时只取出需要的字段,不可select *,而且该字段建了索引。

mysql> EXPLAIN SELECT * from one ;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+
|  1 | SIMPLE      | one   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |      100 | NULL  |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+
2、Using where

Using where:查询时未找到可用的索引,进而通过where条件过滤获取所需数据,但要注意的是并不是所有带where语句的查询都会显示Using where

下边示例create_time 并未用到索引,typeALL,即MySQL通过全表扫描后再按where条件筛选数据。

mysql> EXPLAIN SELECT one_name from one where create_time ='2020-05-18';
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | one   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |    33.33 | Using where |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
3、Using temporary

Using temporary:表示查询后结果需要使用临时表来存储,一般在排序或者分组查询时用到。

mysql> EXPLAIN SELECT one_name from one where one_id in (1,2) group by one_name;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | one   | NULL       | range| NULL          | NULL | NULL    | NULL |    3 |    33.33 | Using where; Using temporary; Using filesort |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+
4、Using filesort

Using filesort:表示无法利用索引完成的排序操作,也就是ORDER BY的字段没有索引,通常这样的SQL都是需要优化的。

mysql> EXPLAIN SELECT one_id from one  ORDER BY create_time;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra          |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------+
|  1 | SIMPLE      | one   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |      100 | Using filesort |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------+

如果ORDER BY字段有索引就会用到覆盖索引,相比执行速度快很多。

mysql> EXPLAIN SELECT one_id from one  ORDER BY one_id;
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
| id | select_type | table | partitions | type  | possible_keys | key     | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | one   | NULL       | index | NULL          | PRIMARY | 4       | NULL |    3 |      100 | Using index |
+----+-------------+-------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
5、Using join buffer

Using join buffer:在我们联表查询的时候,如果表的连接条件没有用到索引,需要有一个连接缓冲区来存储中间结果。

先看一下有索引的情况:连接条件 one_nametwo_name 都用到索引。

mysql> EXPLAIN SELECT one_name from one o,two t where o.one_name = t.two_name;
+----+-------------+-------+------------+-------+---------------+----------+---------+----------------------+------+----------+--------------------------+
| id | select_type | table | partitions | type  | possible_keys | key      | key_len | ref                  | rows | filtered | Extra                    |
+----+-------------+-------+------------+-------+---------------+----------+---------+----------------------+------+----------+--------------------------+
|  1 | SIMPLE      | o     | NULL       | index | idx_name      | idx_name | 768     | NULL                 |    3 |      100 | Using where; Using index |
|  1 | SIMPLE      | t     | NULL       | ref   | idx_name      | idx_name | 768     | xin-slave.o.one_name |    1 |      100 | Using index              |
+----+-------------+-------+------------+-------+---------------+----------+---------+----------------------+------+----------+--------------------------+

接下来删掉 连接条件 one_nametwo_name 的字段索引。发现Extra 列变成 Using join buffertype均为全表扫描,这也是SQL优化中需要注意的地方。

mysql> EXPLAIN SELECT one_name from one o,two t where o.one_name = t.two_name;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra                                              |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
|  1 | SIMPLE      | t     | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    2 |      100 | NULL                                               |
|  1 | SIMPLE      | o     | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |    33.33 | Using where; Using join buffer (Block Nested Loop) |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
6、Impossible where

Impossible where:表示在我们用不太正确的where语句,导致没有符合条件的行。

mysql> EXPLAIN SELECT one_name from one WHERE 1=2;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+------------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra            |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+------------------+
|  1 | SIMPLE      | NULL  | NULL       | NULL | NULL          | NULL | NULL    | NULL | NULL | NULL     | Impossible WHERE |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+------------------+
7、No tables used

No tables used:我们的查询语句中没有FROM子句,或者有 FROM DUAL子句。

mysql> EXPLAIN select now();
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra          |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------+
|  1 | SIMPLE      | NULL  | NULL       | NULL | NULL          | NULL | NULL    | NULL | NULL | NULL     | No tables used |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------+

Extra列的信息非常非常多,这里就不再一一列举了,详见 MySQL官方文档 :https://dev.mysql.com/doc/refman/5.7/en/explain-output.html#jointype_index_merge
不会看 Explain执行计划,劝你简历别写熟悉 SQL优化_第2张图片

总结

上边只是简单介绍了下 Explain 执行计划各个列的含义,了解它不仅仅是要应付面试,在实际开发中也经常会用到。比如对慢SQL进行分析,如果连执行计划结果都不会看,那还谈什么SQL优化呢?

整理了几百本各类技术电子书和视频课程 ,送给小伙伴们。同名公号内回【666】自行领取。和一些小伙伴们建了一个技术交流群,一起探讨技术、分享技术资料,旨在共同学习进步,如果感兴趣就加入我们吧!

你可能感兴趣的:(mysql)