Home  >  Article  >  Database  >  MySQL execution plan explanation and index data structure deduction

MySQL execution plan explanation and index data structure deduction

coldplay.xixi
coldplay.xixiforward
2020-11-13 17:08:072619browse

mysql tutorialThe column introduces the execution plan explain and index data structure

MySQL execution plan explanation and index data structure deduction

Preparation work

First build the database table, the MySQL table for demonstration, and the table creation statement:

CREATE TABLE `emp` (  `id` int(11) NOT NULL AUTO_INCREMENT COMMENT '主键',  `empno` int(11) DEFAULT NULL COMMENT '雇员工号',  `ename` varchar(255) DEFAULT NULL COMMENT '雇员姓名',  `job` varchar(255) DEFAULT NULL COMMENT '工作',  `mgr` varchar(255) DEFAULT NULL COMMENT '经理的工号',  `hiredate` date DEFAULT NULL COMMENT '雇用日期',  `sal` double DEFAULT NULL COMMENT '工资',  `comm` double DEFAULT NULL COMMENT '津贴',  `deptno` int(11) DEFAULT NULL COMMENT '所属部门号',
  PRIMARY KEY (`id`) USING BTREE
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COMMENT='雇员表';CREATE TABLE `dept` (  `id` int(11) NOT NULL AUTO_INCREMENT COMMENT '主键',  `deptno` int(11) DEFAULT NULL COMMENT '部门号',  `dname` varchar(255) DEFAULT NULL COMMENT '部门名称',  `loc` varchar(255) DEFAULT NULL COMMENT '地址',
  PRIMARY KEY (`id`) USING BTREE
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COMMENT='部门表';CREATE TABLE `salgrade` (  `id` int(11) NOT NULL COMMENT '主键',  `grade` varchar(255) DEFAULT NULL COMMENT '等级',  `lowsal` varchar(255) DEFAULT NULL COMMENT '最低工资',  `hisal` varchar(255) DEFAULT NULL COMMENT '最高工资',
  PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COMMENT='工资等级表';CREATE TABLE `bonus` (  `id` int(11) NOT NULL COMMENT '主键',  `ename` varchar(255) DEFAULT NULL COMMENT '雇员姓名',  `job` varchar(255) DEFAULT NULL COMMENT '工作',  `sal` double DEFAULT NULL COMMENT '工资',  `comm` double DEFAULT NULL COMMENT '津贴',
  PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COMMENT='奖金表';复制代码

The subsequent execution plan, query optimization, index optimization and other knowledge drills are based on the above table to operate.

MySQL Execution Plan

To perform SQL tuning, you have to know how the SQL statement to be tuned is executed, and check the specific execution process of the SQL statement to speed up the SQL statement. execution efficiency.

You can use the explain SQL statement to simulate the optimizer executing a SQL query statement, so as to know how MySQL processes SQL statements.

For explain, you can read the official website introduction.

Explain’s output format

mysql> explain select * from emp;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+|  1 | SIMPLE      | emp   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    1 |   100.00 | NULL  |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+复制代码

Explanation of fields id, select_type and other fields:

Column Meaning
id The SELECT identifier )
select_type The SELECT type(The SELECT type)
table The table for the output row (output the row's table name)
partitions The matching partitions (matching partitions)
type The join type
possible_keys The possible indexes to choose (possible index selection)
key The index actually chosen
key_len The length of the chosen key (the length of the selected key)
ref The columns compared to the index (the columns compared with the index)
rows Estimate of rows to be examined
filtered Percentage of rows filtered by table condition (percentage of rows filtered by table condition)
extra Additional information

id

select查询的序列号,包含一组数字,表示查询中执行select子句或者操作表的顺序。

id号分为三类:

  • 如果id相同,那么执行顺序从上到下
mysql> explain select * from emp e join dept d on e.deptno = d.deptno join salgrade sg on e.sal between sg.lowsal and sg.hisal;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra                                              |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+
|  1 | SIMPLE      | e     | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    1 |   100.00 | NULL                                               |
|  1 | SIMPLE      | d     | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    1 |   100.00 | Using where; Using join buffer (Block Nested Loop) |
|  1 | SIMPLE      | sg    | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    1 |   100.00 | Using where; Using join buffer (Block Nested Loop) |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+复制代码

这个查询,用explain执行一下,id序号都是1,那么MySQL的执行顺序就是从上到下执行的。

  • 如果id不同,如果是子查询,id的序号会递增,id值越大优先级越高,越先被执行
mysql> explain select * from emp e where e.deptno in (select d.deptno from dept d where d.dname = 'SALEDept');
+----+--------------+-------------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+| id | select_type  | table       | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra                                              |
+----+--------------+-------------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+|  1 | SIMPLE       | <subquery2> | NULL       | ALL  | NULL          | NULL | NULL    | NULL | NULL |   100.00 | NULL                                               |
|  1 | SIMPLE       | e           | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    2 |    50.00 | Using where; Using join buffer (Block Nested Loop) |
|  2 | MATERIALIZED | d           | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    1 |   100.00 | Using where                                        |
+----+--------------+-------------+------------+------+---------------+------+---------+------+------+----------+----------------------------------------------------+复制代码</subquery2>

这个例子的执行顺序是先执行id为2的,然后执行id为1的。

  • id相同和不同的,同时存在:相同的可以认为是一组,从上往下顺序执行,在所有组中,id值越大,优先级越高,越先执行

还是上面那个例子,先执行id为2的,然后按顺序从上往下执行id为1的。

select_type

主要用来分辨查询的类型,是普通查询还是联合查询还是子查询。

select_type Value JSON Name Meaning
SIMPLE None Simple SELECT (not using UNION or subqueries)
PRIMARY None Outermost SELECT
UNION None Second or later SELECT statement in a UNION
DEPENDENT UNION dependent (true) Second or later SELECT statement in a UNION, dependent on outer query
UNION RESULT union_result Result of a UNION.
SUBQUERY None First SELECT in subquery
DEPENDENT SUBQUERY dependent (true) First SELECT in subquery, dependent on outer query
DERIVED None Derived table
MATERIALIZED materialized_from_subquery Materialized subquery
UNCACHEABLE SUBQUERY cacheable (false) A subquery for which the result cannot be cached and must be re-evaluated for each row of the outer query
UNCACHEABLE UNION cacheable (false) The second or later select in a UNION that belongs to an uncacheable subquery (see UNCACHEABLE SUBQUERY)
  • SIMPLE 简单的查询,不包含子查询和union
mysql> explain select * from emp;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+|  1 | SIMPLE      | emp   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |   100.00 | NULL  |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------+复制代码
  • primary 查询中若包含任何复杂的子查询,最外层查询则被标记为Primary
  • union 若第二个select出现在union之后,则被标记为union
mysql> explain select * from emp where deptno = 1001 union select * from emp where sal  | NULL       | ALL  | NULL          | NULL | NULL    | NULL | NULL |     NULL | Using temporary |
+----+--------------+------------+------------+------+---------------+------+---------+------+------+----------+-----------------+复制代码

这条语句的select_type包含了primaryunion

  • dependent union 跟union类似,此处的depentent表示union或union all联合而成的结果会受外部表影响
  • union result 从union表获取结果的select
  • dependent subquery subquery的子查询要受到外部表查询的影响
mysql> explain select * from emp e where e.empno  in ( select empno from emp where deptno = 1001 union select empno from emp where sal  | NULL       | ALL  | NULL          | NULL | NULL    | NULL | NULL |     NULL | Using temporary |
+----+--------------------+------------+------------+------+---------------+------+---------+------+------+----------+-----------------+复制代码

这条SQL执行包含了PRIMARYDEPENDENT SUBQUERYDEPENDENT UNIONUNION RESULT

  • subquery 在select或者where列表中包含子查询

举例:

mysql> explain select * from emp where sal > (select avg(sal) from emp) ;
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+| id | select_type | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+|  1 | PRIMARY     | emp   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    4 |    33.33 | Using where |
|  2 | SUBQUERY    | emp   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    4 |   100.00 | NULL        |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+复制代码
  • DERIVED from子句中出现的子查询,也叫做派生表
  • MATERIALIZED Materialized subquery?
  • UNCACHEABLE SUBQUERY 表示使用子查询的结果不能被缓存

例如:

mysql> explain select * from emp where empno = (select empno from emp where deptno=@@sort_buffer_size);
+----+----------------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+| id | select_type          | table | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+----------------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+|  1 | PRIMARY              | emp   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    4 |   100.00 | Using where |
|  2 | UNCACHEABLE SUBQUERY | emp   | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    4 |    25.00 | Using where |
+----+----------------------+-------+------------+------+---------------+------+---------+------+------+----------+-------------+复制代码
  • uncacheable union 表示union的查询结果不能被缓存

table

对应行正在访问哪一个表,表名或者别名,可能是临时表或者union合并结果集。

  1. 如果是具体的表名,则表明从实际的物理表中获取数据,当然也可以是表的别名
  2. 表名是derivedN的形式,表示使用了id为N的查询产生的衍生表
  3. 当有union result的时候,表名是union n1,n2等的形式,n1,n2表示参与union的id

type

type显示的是访问类型,访问类型表示我是以何种方式去访问我们的数据,最容易想到的是全表扫描,直接暴力的遍历一张表去寻找需要的数据,效率非常低下。

访问的类型有很多,效率从最好到最坏依次是:

system > const > eq_ref > ref > fulltext > ref_or_null > index_merge > unique_subquery > index_subquery > range > index > ALL

一般情况下,得保证查询至少达到range级别,最好能达到ref

  • all 全表扫描,一般情况下出现这样的sql语句而且数据量比较大的话那么就需要进行优化

通常,可以通过添加索引来避免ALL

  • index 全索引扫描这个比all的效率要好,主要有两种情况:
    • 一种是当前的查询时覆盖索引,即我们需要的数据在索引中就可以索取
    • 一是使用了索引进行排序,这样就避免数据的重排序
  • range 表示利用索引查询的时候限制了范围,在指定范围内进行查询,这样避免了index的全索引扫描,适用的操作符: =, , >, >=,

官网上举例如下:

SELECT * FROM tbl_name WHERE key_column = 10;

SELECT * FROM tbl_name WHERE key_column BETWEEN 10 and 20;

SELECT * FROM tbl_name WHERE key_column IN (10,20,30);

SELECT * FROM tbl_name WHERE key_part1 = 10 AND key_part2 IN (10,20,30);

  • index_subquery 利用索引来关联子查询,不再扫描全表

value IN (SELECT key_column FROM single_table WHERE some_expr)

  • unique_subquery 该连接类型类似与index_subquery,使用的是唯一索引

value IN (SELECT primary_key FROM single_table WHERE some_expr)

  • index_merge 在查询过程中需要多个索引组合使用
  • ref_or_null 对于某个字段既需要关联条件,也需要null值的情况下,查询优化器会选择这种访问方式

SELECT * FROM ref_table

WHERE key_column=expr OR key_column IS NULL;

  • fulltext 使用FULLTEXT索引执行join
  • ref 使用了非唯一性索引进行数据的查找

SELECT * FROM ref_table WHERE key_column=expr;

SELECT * FROM ref_table,other_table WHERE ref_table.key_column=other_table.column;

SELECT * FROM ref_table,other_table WHERE ref_table.key_column_part1=other_table.column AND ref_table.key_column_part2=1;

  • eq_ref 使用唯一性索引进行数据查找

SELECT * FROM ref_table,other_table WHERE ref_table.key_column=other_table.column;

SELECT * FROM ref_table,other_table WHERE ref_table.key_column_part1=other_table.column AND ref_table.key_column_part2=1;

  • const 这个表至多有一个匹配行

SELECT * FROM tbl_name WHERE primary_key=1;

SELECT * FROM tbl_name WHERE primary_key_part1=1 AND primary_key_part2=2;

例如:

mysql> explain select * from emp where id = 1;
+----+-------------+-------+------------+-------+---------------+---------+---------+-------+------+----------+-------+| id | select_type | table | partitions | type  | possible_keys | key     | key_len | ref   | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+---------+---------+-------+------+----------+-------+|  1 | SIMPLE      | emp   | NULL       | const | PRIMARY       | PRIMARY | 4       | const |    1 |   100.00 | NULL  |
+----+-------------+-------+------------+-------+---------------+---------+---------+-------+------+----------+-------+复制代码
  • system 表只有一行记录(等于系统表),这是const类型的特例,平时不会出现

possible_keys

显示可能应用在这张表中的索引,一个或多个,查询涉及到的字段上若存在索引,则该索引将被列出,但不一定被查询实际使用

key

实际使用的索引,如果为null,则没有使用索引,查询中若使用了覆盖索引,则该索引和查询的select字段重叠

key_len

表示索引中使用的字节数,可以通过key_len计算查询中使用的索引长度,在不损失精度的情况下长度越短越好

ref

显示索引的哪一列被使用了,如果可能的话,是一个常数

rows

根据表的统计信息及索引使用情况,大致估算出找出所需记录需要读取的行数,此参数很重要,直接反应的sql找了多少数据,在完成目的的情况下越少越好

extra

包含额外的信息

  • using filesort 说明mysql无法利用索引进行排序,只能利用排序算法进行排序,会消耗额外的位置
  • using temporary 建立临时表来保存中间结果,查询完成之后把临时表删除
  • using index 这个表示当前的查询是覆盖索引的,直接从索引中读取数据,而不用访问数据表。如果同时出现using where 表明索引被用来执行索引键值的查找,如果没有,表示索引被用来读取数据,而不是真的查找
  • using where 使用where进行条件过滤
  • using join buffer 使用连接缓存
  • impossible where where语句的结果总是false

MySQL索引基本知识

想要了解索引的优化方式,必须要对索引的底层原理有所了解。

索引的优点

  1. 大大减少了服务器需要扫描的数据量
  2. 帮助服务器避免排序和临时表
  3. 将随机io变成顺序io(提升效率)

索引的用处

  1. 快速查找匹配WHERE子句的行
  2. 从consideration中消除行,如果可以在多个索引之间进行选择,mysql通常会使用找到最少行的索引
  3. 如果表具有多列索引,则优化器可以使用索引的任何最左前缀来查找行
  4. 当有表连接的时候,从其他表检索行数据
  5. 查找特定索引列的min或max值
  6. 如果排序或分组时在可用索引的最左前缀上完成的,则对表进行排序和分组
  7. 在某些情况下,可以优化查询以检索值而无需查询数据行

MySQL execution plan explanation and index data structure deduction

MySQL execution plan explanation and index data structure deduction

MySQL索引数据结构推演

索引用于快速查找具有特定列值的行。

如果没有索引,MySQL必须从第一行开始,然后通读整个表以找到相关的行。

表越大花费的时间越多,如果表中有相关列的索引,MySQL可以快速确定要在数据文件中间查找的位置,而不必查看所有数据。这比顺序读取每一行要快得多。

既然MySQL索引能帮助我们快速查询到数据,那么它的底层是怎么存储数据的呢?

几种可能的存储结构

hash

hash表的索引格式

Disadvantages of storing data in hash tables:

  1. If you use hash storage, you need to add all data files to the memory, which consumes more memory space
  2. If all queries are equivalent queries, then hashing is indeed very fast, but in the actual working environment, the range search data is more, rather than equivalent queries, in this case hash is not suitable

In fact, when the MySQL storage engine is memory, the index data structure uses a hash table.

Binary tree

The structure of the binary tree is like this:

MySQL execution plan explanation and index data structure deduction

The binary tree will cause data loss due to the depth of the tree. Tilt, if the depth of the tree is too deep, it will cause more IO times and affect the efficiency of data reading.

AVL tree Needs to be rotated, see the legend:

MySQL execution plan explanation and index data structure deduction

Red-black tree There are more operations besides rotation A color-changing function (in order to reduce rotation), although the insertion speed is fast, the query efficiency is lost.

MySQL execution plan explanation and index data structure deduction

Binary tree, AVL tree, red-black tree will all be caused by the depth of the tree being too deep. The number of IO times increases, which affects the efficiency of data reading.

Let’s take a look at B tree

Features of B tree:

  • All key values ​​are distributed throughout the tree
  • The search may end at a non-leaf node, and a search is performed within the complete set of keywords. The performance is close to binary search
  • Each node has at most m subtrees
  • The root node has at least 2 subtrees Tree
  • The branch node has at least m/2 subtrees (all branch nodes except the root node and leaf nodes)
  • All leaf nodes are on the same level, and each node can have at most m -1 key, and arranged in ascending order

MySQL execution plan explanation and index data structure deduction

Legend description:

Each node occupies one disk block, There are two ascending-order keys on a node and three pointers to the root node of the subtree. The pointers store the address of the disk block where the child node is located.

The three ranges divided by the two keywords correspond to the ranges of the data of the subtree pointed to by the three pointers.

Taking the root node as an example, the keywords are 16 and 34. The data range of the subtree pointed by the P1 pointer is less than 16, the data range of the subtree pointed by the P2 pointer is 16~34, and the data range pointed by the P3 pointer The data range of the subtree is greater than 34.

Keyword search process:

1. Find disk block 1 based on the root node and read it into memory. [Disk I/O operation 1st time]

2. Compare keyword 28 in the interval (16,34), find the pointer P2 of disk block 1.

3. Find disk block 3 according to the P2 pointer and read it into the memory. [Disk I/O operation 2nd time]

4. Compare keyword 28 in the interval (25,31), find the pointer P2 of disk block 3.

5. Find disk block 8 according to the P2 pointer and read it into the memory. [Disk I/O operation 3rd time]

6. Find keyword 28 in the keyword list in disk block 8.

From this, we can know the shortcomings of B-tree storage:

  • Each node has a key and also contains data, and the storage space of each page is limited. If the data is relatively large, it will cause the number of keys stored in each node to become smaller
  • When the amount of stored data is large, it will lead to a larger depth, increase the number of disk IO times during query, and thus affect query performance

So what is the MySQL index data structure?

Official website: Most MySQL indexes (PRIMARY KEY, UNIQUE, INDEX, and FULLTEXT) are stored in B-trees

Don't get me wrong, in fact, the storage structure of the MySQL index is B tree. After our analysis above, we know that B tree is inappropriate.

mysql index data structure---B Tree

B Tree is an optimization based on BTree, with the following changes:

1. Each node of B Tree can contain more nodes. There are two reasons for this. The first reason is to reduce the height of the tree. The second reason is to change the data range into multiple intervals. The more, the faster the data retrieval is.

2. Non-leaf nodes store keys, and leaf nodes store keys and data.

3. Two pointers of leaf nodes are connected to each other (in line with the read-ahead characteristics of the disk), and the sequential query performance is higher.

B tree storage search diagram:

MySQL execution plan explanation and index data structure deduction

注意:

在B+Tree上有两个头指针,一个指向根节点,另一个指向关键字最小的叶子节点,而且所有叶子节点(即数据节点)之间是一种链式环结构。

因此可以对 B+Tree 进行两种查找运算:一种是对于主键的范围查找和分页查找,另一种是从根节点开始,进行随机查找。

由于B+树叶子结点只存放data,根节点只存放key,那么我们计算一下,即使只有3层B+树,也能制成千万级别的数据。

你得知道的技(zhuang)术(b)名词

假设有这样一个表如下,其中id是主键:

mysql> select * from stu;
+------+---------+------+| id   | name    | age  |
+------+---------+------+|    1 | Jack Ma |   18 |
|    2 | Pony    |   19 |
+------+---------+------+复制代码

回表

我们对普通列建普通索引,这时候我们来查:

select * from stu where name='Pony';复制代码

由于name建了索引,查询时先找nameB+树,找到主键id后,再找主键idB+树,从而找到整行记录。

这个最终会回到主键上来查找B+树,这个就是回表

覆盖索引

如果是这个查询:

mysql> select id from stu where name='Pony';复制代码

就没有回表了,因为直接找到主键id,返回就完了,不需要再找其他的了。

没有回表就叫覆盖索引

最左匹配

再来以nameage两个字段建组合索引(name, age),然后有这样一个查询:

select * from stu where name=? and age=?复制代码

这时按照组合索引(name, age)查询,先匹配name,再匹配age,如果查询变成这样:

select * from stu where age=?复制代码

直接不按name查了,此时索引不会生效,也就是不会按照索引查询---这就是最左匹配原则。

加入我就要按age查,还要有索引来优化呢?可以这样做:

  • (推荐)把组合索引(name, age)换个顺序,建(age, name)索引
  • 或者直接把age字段单独建个索引

索引下推

可能也叫谓词下推。。。

select t1.name,t2.name from t1 join t2 on t1.id=t2.id复制代码

t1有10条记录,t2有20条记录。

我们猜想一下,这个要么按这个方式执行:

先t1,t2按id合并(合并后20条),然后再查t1.name,t2.name

或者:

先把t1.name,t2.name找出来,再按照id关联

如果不使用索引条件下推优化的话,MySQL只能根据索引查询出t1,t2合并后的所有行,然后再依次比较是否符合全部条件。

当使用了索引条件下推优化技术后,可以通过索引中存储的数据判断当前索引对应的数据是否符合条件,只有符合条件的数据才将整行数据查询出来。

小结

  1. Explain 为了知道优化SQL语句的执行,需要查看SQL语句的具体执行过程,以加快SQL语句的执行效率。
  2. 索引优点及用处。
  3. 索引采用的数据结构是B+树。
  4. 回表,覆盖索引,最左匹配和索引下推。

更多相关免费学习推荐:mysql教程(视频)

The above is the detailed content of MySQL execution plan explanation and index data structure deduction. For more information, please follow other related articles on the PHP Chinese website!

Statement:
This article is reproduced at:juejin.im. If there is any infringement, please contact admin@php.cn delete