MySQL单表查询实例详解
云崖先生 人气:01、准备数据
以下操作将在该表中进行
create table student ( id int unsigned primary key auto_increment, name char(12) not null, gender enum("male","famale") default "male", age tinyint unsigned not null, hoc_group char(12) not null, html tinyint unsigned not null, css tinyint unsigned not null, js tinyint unsigned not null, sanction enum("大处分","小处分","无") ); insert into student(name,gender,age,hoc_group,html,css,js,sanction) values ("Yunya","male",18,"first",88,93,76,"无"), ("Jack","male",17,"second",92,81,88,"无"), ("Bella","famale",17,"first",72,68,91,"小处分"), ("Dairis","famale",18,"third",89,54,43,"大处分"), ("Kyle","famale",19,"fifth",31,24,60,"大处分"), ("Alice","famale",16,"second",49,23,58,"无"), ("Ken","male",16,"third",33,62,17,"大处分"), ("Jason","male",21,"fourth",91,92,90,"无"), ("Tom","male",20,"fifth",88,72,91,"无"), ("Fiona","famale",19,"fourth",60,71,45,"无");
2、查询语法
SELECT DISTINCT(字段名1,字段名2...) FROM 表名 WHERE 条件 GROUP BY 字段名 HAVING 筛选 ORDER BY 字段名 asc/desc LIMIT 限制条数;
3、执行顺序
虽然查询的书写语法是上面那样的,但是其内部执行顺序却有些不太一样。
1.通过from找到将要查询的表
2.where规定查询条件,在表记录中逐行进行查询并筛选出符合规则的记录
3.将查到的记录进行字段分组group by,如果没有进行分组,则默认为一组
4.将分组得到的结果进行having筛选,可使用聚和函数(where时不可使用聚合函数)
5.执行select准备打印
6.执行distinct对打印结果进行去重
7.执行ordery by对结果进行排序
8.执行limit对打印结果的条数进行限制
4、select
select主要复负责打印相关的工作
4.1 全部查询
使用select * from 表名
可拿到该表下全部的数据
以下示例将展示使用全部查询拿到student表中所有记录
select * from student; +----+--------+--------+-----+-----------+------+-----+----+-----------+ | id | name | gender | age | hoc_group | html | css | js | sanction | +----+--------+--------+-----+-----------+------+-----+----+-----------+ | 1 | Yunya | male | 18 | first | 88 | 93 | 76 | 无 | | 2 | Jack | male | 17 | second | 92 | 81 | 88 | 无 | | 3 | Bella | famale | 17 | first | 72 | 68 | 91 | 小处分 | | 4 | Dairis | famale | 18 | third | 89 | 54 | 43 | 大处分 | | 5 | Kyle | famale | 19 | fifth | 31 | 24 | 60 | 大处分 | | 6 | Alice | famale | 16 | second | 49 | 23 | 58 | 无 | | 7 | Ken | male | 16 | third | 33 | 62 | 17 | 大处分 | | 8 | Jason | male | 21 | fourth | 91 | 92 | 90 | 无 | | 9 | Tom | male | 20 | fifth | 88 | 72 | 91 | 无 | | 10 | Fiona | famale | 19 | fourth | 60 | 71 | 45 | 无 | +----+--------+--------+-----+-----------+------+-----+----+-----------+
4.2 字段查询
使用select 字段名1,字段名2 from 表名可拿到特定字段下相应的数据
以下示例将展示使用字段查询拿到每个学生的HTML\CSS\JS成绩
select name,html,css,js from student; +--------+------+-----+----+ | name | html | css | js | +--------+------+-----+----+ | Yunya | 88 | 93 | 76 | | Jack | 92 | 81 | 88 | | Bella | 72 | 68 | 91 | | Dairis | 89 | 54 | 43 | | Kyle | 31 | 24 | 60 | | Alice | 49 | 23 | 58 | | Ken | 33 | 62 | 17 | | Jason | 91 | 92 | 90 | | Tom | 88 | 72 | 91 | | Fiona | 60 | 71 | 45 | +--------+------+-----+----+
4.3 as 别名
使用select 字段名1 as 别名1, 字段名2 as 别名2 from 表名可将查询到的记录字段修改一个别名
以下示例将展示修改name字段为姓名,修改gender字段为性别,修改age字段为年龄的操作
select name as "姓名", gender as "性别", age as "年龄" from student; +--------+--------+--------+ | 姓名 | 性别 | 年龄 | +--------+--------+--------+ | Yunya | male | 18 | | Jack | male | 17 | | Bella | famale | 17 | | Dairis | famale | 18 | | Kyle | famale | 19 | | Alice | famale | 16 | | Ken | male | 16 | | Jason | male | 21 | | Tom | male | 20 | | Fiona | famale | 19 | +--------+--------+--------+
4.4 distinct
使用select distinct(字段名1, 字段名2) from 表名可将查询到的记录做一个取消重复的操作
以下示例将展示使用去重功能来看有多少个小组
select distinct(hoc_group) from student; +-----------+ | hoc_group | +-----------+ | first | | second | | third | | fifth | | fourth | +-----------+
4.5 四则运算
查询结果可进行四则运算,以下示例将展示拿到每个同学三科总分的操作
select name, html+css+js as 总成绩 from student; +--------+-----------+ | name | 总成绩 | +--------+-----------+ | Yunya | 257 | | Jack | 261 | | Bella | 231 | | Dairis | 186 | | Kyle | 115 | | Alice | 130 | | Ken | 112 | | Jason | 273 | | Tom | 251 | | Fiona | 176 | +--------+-----------+
4.6 显示格式
使用concat()可将查询结果与任意字符串进行拼接
使用concat_ws()可指定连接符进行拼接,第一个参数是连接符
select concat("姓名->",name," ","性别->",gender) from student; # 合并成了一个字符串,注意用的空格分隔开的,不然会黏在一起 +--------------------------------------------------+ | concat("姓名->",name," ","性别->",gender) | +--------------------------------------------------+ | 姓名->Yunya 性别->male | | 姓名->Jack 性别->male | | 姓名->Bella 性别->famale | | 姓名->Dairis 性别->famale | | 姓名->Kyle 性别->famale | | 姓名->Alice 性别->famale | | 姓名->Ken 性别->male | | 姓名->Jason 性别->male | | 姓名->Tom 性别->male | | 姓名->Fiona 性别->famale | +--------------------------------------------------+
select concat_ws("|||",name,gender,age) from student; # 使用|||为每个字段进行分割 +----------------------------------+ | concat_ws("|||",name,gender,age) | +----------------------------------+ | Yunya|||male|||18 | | Jack|||male|||17 | | Bella|||famale|||17 | | Dairis|||famale|||18 | | Kyle|||famale|||19 | | Alice|||famale|||16 | | Ken|||male|||16 | | Jason|||male|||21 | | Tom|||male|||20 | | Fiona|||famale|||19 | +----------------------------------+
5、where
where条件是查询的第一道坎,能有效过滤出我们想要的任意数据
5.1、比较运算
使用比较运算符> < >= <= !=进行查询
以下示例将展示使用where过滤出js成绩大于80分的同学
select name, js from student where js > 80; +-------+----+ | name | js | +-------+----+ | Jack | 88 | | Bella | 91 | | Jason | 90 | | Tom | 91 | +-------+----+
5.2、逻辑运算
使用and or not可进行逻辑运算与多条件查询
以下示例将展示使用where多条件查询过滤出各科成绩都大于80分的同学
select name, html, css, js from student where html > 80 and css > 80 and js > 80; +-------+------+-----+----+ | name | html | css | js | +-------+------+-----+----+ | Jack | 92 | 81 | 88 | | Jason | 91 | 92 | 90 | +-------+------+-----+----+
5.3、成员运算
in可以在特定的值中进行获取,如in(80,90,100)则代表只取80或者90或者100的这几条记录。
以下示例将展示只取第一组first以及第二组second学生的个人信息
select name, gender, age, hoc_group from student where hoc_group in ("first","second"); +-------+--------+-----+-----------+ | name | gender | age | hoc_group | +-------+--------+-----+-----------+ | Yunya | male | 18 | first | | Jack | male | 17 | second | | Bella | famale | 17 | first | | Alice | famale | 16 | second | +-------+--------+-----+-----------+
5.4、between and
between and也是取区间的意思,
以下示例将展示使用between and过滤出Js成绩大于等于60并且小于80的同学
select name, js from student where js between 60 and 80; +-------+----+ | name | js | +-------+----+ | Yunya | 76 | | Kyle | 60 | +-------+----+
5.5、like
like是模糊查询,其中%代表任意多个字符(类似于贪婪匹配的通配符.*),_代表任意一个字符(类似于非贪婪匹配的通配符.*?)。
以下示例将展示使用like/%匹配出姓名以k开头的所有同学的名字
select name from student where name like "k%"; +------+ | name | +------+ | Kyle | | Ken | +------+
以下示例将展示使用like/_匹配出姓名以k开头并整体长度为3的同学的名字
select name from student where name like "k__"; +------+ | name | +------+ | Ken | +------+
5.6、正则匹配
使用RegExp可进行正则匹配,以下示例将展示使用正则匹配出名字中带有k的所有同学姓名
select name from student where name REGEXP "k+"; +------+ | name | +------+ | Jack | | Kyle | | Ken | +------+
6、group by
分组行为发生在where条件之后,我们可以将查询到的记录按照某个相同字段进行归类,一般分组都会配合聚合函数进行使用。
需要注意的是select语句是排在group by条件之后的,因此聚合函数也能在select语句中使用。
6.1、基本使用
以下示例将展示对hoc_group字段进行分组。
我们按照hoc_group字段进行分组,那么select查询的字段只能是hoc_group字段,想要获取组内的其他字段相关信息,需要借助函数来完成
select hoc_group from student group by hoc_group; +-----------+ | hoc_group | +-----------+ | fifth | | first | | fourth | | second | | third | +-----------+
如果不使用分组,则会产生重复的信息
mysql> select hoc_group from student; +-----------+ | hoc_group | +-----------+ | first | | second | | first | | third | | fifth | | second | | third | | fourth | | fifth | | fourth | +-----------+
6.2、group_concat
用什么字段名进行分组,在select查询时就只能查那个用于分组的字段,查询别的字段会抛出异常,会提示sql_mode异常。
我们将Js成绩大于80分的同学筛选出来并且按照gender字段进行分组,此外我们还想查看其所有满足条件同学的名字。
以下这样操作会抛出异常。
mysql> select gender,name from student where js > 80 group by gender; ERROR 1055 (42000): Expression #1 of SELECT list is not in GROUP BY clause and contains nonaggregated column 'school.student.name' which is not functionally dependent on columns in GROUP BY clause; this is incompatible with sql_mode=only_full_group_by
必须借助group_concat()函数来进行操作才能使我们的需求圆满完成。
select gender, group_concat(name) from student where js > 80 group by gender; +--------+--------------------+ | gender | group_concat(name) | +--------+--------------------+ | male | Jack,Jason,Tom | | famale | Bella | +--------+--------------------+
6.3、分组模式
ONLY_FULL_GROUP_BY要求select中的字段是在与group by中使用的字段
如果group by是主键或unique not null时可以在select中列出其他字段
#查看MySQL 5.7默认的sql_mode如下: mysql> select @@global.sql_mode; ONLY_FULL_GROUP_BY,STRICT_TRANS_TABLES,NO_ZERO_IN_DATE,NO_ZERO_DATE,ERROR_FOR_DIVISION_BY_ZERO,NO_AUTO_CREATE_USER,NO_ENGINE_SUBSTITUTION #设置sql_mole如下操作(我们可以去掉ONLY_FULL_GROUP_BY模式): mysql> set global sql_mode='STRICT_TRANS_TABLES,NO_ZERO_IN_DATE,NO_ZERO_DATE,ERROR_FOR_DIVISION_BY_ZERO,NO_AUTO_CREATE_USER,NO_ENGINE_SUBSTITUTION';
6.4、聚合函数
聚合函数可以在where执行后的所有语句中使用,比如having,select等。
聚合函数一般是同分组进行配套使用,以下是常用的聚合函数。
函数名 | 作用 |
---|---|
COUNT() | 对组内成员某一字段求个数 |
MAX() | 对组内成员某一字段求最大值 |
MIN() | 对组内成员某一字段求最小值 |
AVG() | 对组内成员某一字段求平均值 |
SUM() | 对组内成员某一字段求和 |
注意:不使用分组,则默认为一组 |
以下示例将展示求每组的成绩总和
select hoc_group, sum(js+html+css) from student group by hoc_group; +-----------+-------------------+ | hoc_group | sum(js+html+css) | +-----------+-------------------+ | fifth | 366 | | first | 488 | | fourth | 449 | | second | 391 | | third | 298 | +-----------+-------------------+
以下示例将展示整个班级的平均成绩及总成绩(round()用于四舍五入操作)
select round(avg(html+js+css)) as 平均分 ,sum(html+js+css) as 总分 from student; +-----------+--------+ | 平均分 | 总分 | +-----------+--------+ | 199 | 1992 | +-----------+--------+
以下示例将展示打印出总科成绩最高分数
select max(js+css+html) from student; +------------------+ | max(js+css+html) | +------------------+ | 273 | +------------------+
以下示例将展示查看本班有多少男生,多少女生
select gender, count(id) from student group by gender; +--------+-----------+ | gender | count(id) | +--------+-----------+ | male | 5 | | famale | 5 | +--------+-----------+
7、having
having也可用于过滤操作
7.1、区别差异
执行优先级从高到低:where> group by > having
where发生在分组group by之前,因而where中可以有任意字段,但是绝对不能使用聚合函数。
having发生在分组group by之后,因而having中可以使用分组的字段,无法直接取到其他字段,可以使用聚合函数
7.2、示例演示
以下示例将展示使用having过滤取出每组总分数大于400的小组
select hoc_group, sum(html+css+js) from student group by hoc_group having sum(html+css+js) > 400; +-----------+------------------+ | hoc_group | sum(html+css+js) | +-----------+------------------+ | first | 488 | | fourth | 449 | +-----------+------------------+
以下示例将展示使用having过滤取出有处分的同学。(可以使用分组的字段,但不能使用其他字段)
select sanction, group_concat(name) from student group by sanction having sanction != "无"; +-----------+--------------------+ | sanction | group_concat(name) | +-----------+--------------------+ | 大处分 | Dairis,Kyle,Ken | | 小处分 | Bella | +-----------+--------------------+
8、ordery by
ordery by用于对查询结果进行排序
默认的排序是按照主键进行排序的
8.1 asc
asc用于升序排列,以下示例将展示按照每位同学的年龄进行升序排列,如果年龄相同则依照总成绩进行升序排列。
select id, name, age, html+css+js as 总成绩 from student order by age, html+css+js asc; +----+--------+-----+-----------+ | id | name | age | 总成绩 | +----+--------+-----+-----------+ | 7 | Ken | 16 | 112 | | 6 | Alice | 16 | 130 | | 3 | Bella | 17 | 231 | | 2 | Jack | 17 | 261 | | 4 | Dairis | 18 | 186 | | 1 | Yunya | 18 | 257 | | 5 | Kyle | 19 | 115 | | 10 | Fiona | 19 | 176 | | 9 | Tom | 20 | 251 | | 8 | Jason | 21 | 273 | +----+--------+-----+-----------+
8.2、desc
desc用于降序排列,以下示例将展示按照每位同学的年龄进行降序排列。
select id, name, age, html+css+js as 总成绩 from student order by age desc; +----+--------+-----+-----------+ | id | name | age | 总成绩 | +----+--------+-----+-----------+ | 8 | Jason | 21 | 273 | | 9 | Tom | 20 | 251 | | 5 | Kyle | 19 | 115 | | 10 | Fiona | 19 | 176 | | 1 | Yunya | 18 | 257 | | 4 | Dairis | 18 | 186 | | 2 | Jack | 17 | 261 | | 3 | Bella | 17 | 231 | | 6 | Alice | 16 | 130 | | 7 | Ken | 16 | 112 | +----+--------+-----+-----------+
9、limit
limit用于控制显示的条数
9.1、示例演示
按照总成绩进行降序排序,只打印1-5名。
select id, name, age, html+css+js as 总成绩 from student order by html+css+js desc limit 5; +----+-------+-----+-----------+ | id | name | age | 总成绩 | +----+-------+-----+-----------+ | 8 | Jason | 21 | 273 | | 2 | Jack | 17 | 261 | | 1 | Yunya | 18 | 257 | | 9 | Tom | 20 | 251 | | 3 | Bella | 17 | 231 | +----+-------+-----+-----------+
按照总成绩进行降序排序,只打印6-8名。
select id, name, age, html+css+js as 总成绩 from student order by html+css+js desc limit 5,3; # 从第五名开始,打印三条。 6,7,8 +----+--------+-----+-----------+ | id | name | age | 总成绩 | +----+--------+-----+-----------+ | 4 | Dairis | 18 | 186 | | 10 | Fiona | 19 | 176 | | 6 | Alice | 16 | 130 | +----+--------+-----+-----------+
总结
加载全部内容