一 单表查询,以下是表内容
1 一 having 过滤
2 1.1 having和where
3 select * from emp where id > 15;
4 解析过程;from > where 找到数据 > 分组(没有默认一个组)> select 打印 where是出结果之前
5 select * from emp having id > 15;
6 解析过程;from > where 找到数据(没有约束条件,就是整个表)) > 分组(没有默认一个组)> select 打印 > having where是出结果之后
7 上面2个输出是一样的,因为没有分组
8 select depart_id,count(id) from emp group by depart_id; 分组完后3个组,就是3个记录,就要通过聚合取值,通过分组字段取值
9
10 select depart_id,count(id) from emp group by depart_id where depart_id=1; 报错,这里改成having就好了
11 from emp group by depart_id 得出一张虚拟的表在内存里面,有2个字段depart_id,count(id)
12 在往后接就是针对这块虚拟的表,也就是where针对的是这个虚拟表,所以报错,因为where针对的是硬盘表
13
14 小结;where和having的区别
15 where是约束来自数据库的数据,是返回结果之前起作用的,从硬盘把数据where约束拿到内存,在分组之前
16 having是过滤声明,是结果已经到内存,在分组之后
17
18
19
20 1.2 聚合 把内容压成一个字段一条记录
21 select depart_id,count(id) from emp group by depart_id having name like 'ego%'; 报错,结果针对虚拟表,只有depart_id,count(id)2个字段
22 select depart_id,count(id) from emp group by depart_id having count(id) > 2; 在having里面的聚合最后就是字符串字段
23 from > where > groupby > 聚合 > select > having
24 select count(id) from emp where id > 15; 解析 from > where > groupby(默认是一组) > 聚合 > select
25 select count(id) from emp; 解析 from > where (没有约束条件,就是整个表) > groupby(默认是一组) > 聚合 > select
26 select count(id) from emp having id > 15; 报错 解析 因为having是groupby后的结果,只有count(id)这个字段
27 select * from emp having avg(salary) > 10000; 报错 聚合把内容压成一个字段一条记录,这条记录的名字是avg(salary)
28 select avg(salary) from emp;
29 select max(salary) from emp having avg(salary) > 10000;
30 解析 from > where (没有约束条件,就是整个表) > groupby(默认是一组) > 聚合max > 聚合avg > select > having
31 也就是过滤针对的是select打印做出的限制,限制为真,就打印select,否则为空
32 select 10000 from emp having 20000 > 10000; 从表中拿到数据,然后where,groupby然后判断过滤后面的,如果为真,就输出select
33
34 **顺序
35 select max(salary) from emp where id > 2 group by depart_id having avg(salary) > 10000;
36 from emp > where id>2 到内存 > group by depart_id 只有depart_id字段(select只能靠聚合) > 算出结果max(salary) avg(salary) > having 满足 > select
37 depart_id max(id) avg(salary)
38 1 3 10
39 2 1 11
40
41 1.3 练习
42 1. 查询各岗位内包含的员工个数小于2的岗位名、岗位内包含员工名字、个数
43 select post,count(id),group_concat(name) from emp group by post having count(id) < 2;
44 3. 查询各岗位平均薪资大于10000的岗位名、平均工资
45 select post,avg(salary),group_concat(name) from emp group by post having avg(salary) > 10000;
46 select post 部门,avg(salary) 平均工资,group_concat(name) 职工 from emp group by post having avg(salary) > 10000;
47 3. 查询各岗位平均薪资大于10000且小于20000的岗位名、平均工资
48 select post,avg(salary),group_concat(name) from emp group by post having avg(salary) between 10000 and 20000);
49
50
51
52 二 order by 关键字 排序 默认升序,asc
53 2.1 select * from emp order by salary; 按照薪资排序,从小到大,升序
54 select * from emp order by salary asc; 升序
55 select * from emp order by salary desc; 降序
56
57 年龄从小到大,出现相同再按照薪资去排
58 select * from emp order by age,salary;
59 select * from emp order by age(asc,desc),salary(asc,desc);
60
61 2.2 优先级
62 在having后面执行的,对打印出来的限制,也就是在select之前
63
64 2.3 练习
65 1. 查询所有员工信息,先按照age升序排序,如果age相同则按照hire_date降序排序
66 select * from employee ORDER BY age asc,hire_date desc;
67 2. 查询各岗位平均薪资大于10000的岗位名、平均工资,结果按平均薪资升序排列
68 select post 岗位名,avg(salary) 平均工资 from emp group by post having avg(salary) > 10000 order by 平均工资 asc;
69 3. 查询各岗位平均薪资大于10000的岗位名、平均工资,结果按平均薪资降序排列
70 select post 岗位名,avg(salary) 平均工资 from emp group by post having avg(salary) > 10000 order by 平均工资 desc;
71
72 ** select name from emp where id > 5 having id > 16;
73 ** select 后面决定了有什么字段,having,等限制打印的一定要限制后面的字段,不然就报错,因为没有这个字段
74
75
76 三 limit 限制查询记录数
77 3.1 select * from emp limit 3; 打印升序(默认就是升序)前三条
78 select * from emp order by id desc limit 3;打印后三条
79
80 3.2 分页查询,从哪开始,往后取几条
81 select * from emp limit 0,3; 0代表第一条,3代表查询几条
82 select * from emp limit 3,3; 3代表第四条
83 select * from emp limit 6,3;
84
85 四 使用正则表达式查询
86 4.1 like 模糊查询,% 匹配多个 _ 匹配一个
87 select * from emp where name like 'eg%';
88 select * from emp where name like '_g%';
89
90 4.2 正则 regexp 支持正则匹配
91 select * from emp where name regexp '^eg';
92 select * from emp where name regexp 'eg';
93
94
95 五 distinct 去重
96 select distinct sex from emp;
97
98 六 数学运算符
99 select name,salary from emp; 每个人名字,薪资
100 select name,salary*12 年薪 from emp;
二 多表查询,一下是表内容
1 多表查询
2 一 简单查询 多张表 提取相同的 假设department id 和 employee dep_id 关联
3 1.1 select * from department,employee; 结果显示一张表上,是左右字段一次排开拼接这种
4 遍历第一张表,循环第二张,配对一遍,显示出来,拼接方式叫做 笛卡尔积
5 select * from department,employee where department.id=employee.dep_id;
6 select * from employee,department where department.id=employee.dep_id; 提取2表相同的部分
7
8
9 二 连表操作
10 2.1 内链接 inner join department on
11 按照on条件只取2表的相同部分,链接成一张虚拟的表
12 select * from employee inner join department on department.id=employee.dep_id; 提取2表相同的部分
13
14 2.2 左链接 left join department on
15 按照on条件取2表的相同部分基础上,保留左表有,右边没有的记录
16 select * from employee left join department on department.id=employee.dep_id;
17
18 2.3 右链接 right join department on
19 按照on条件取2表的相同部分基础上,保留右表有,左边没有的记录
20 select * from employee right join department on department.id=employee.dep_id;
21
22 2.4 全链接 full join 在mysql里面没有这个功能,只是个概念,还要考左,右链接拼凑出来
23 有对应关系,和没对应关系的都取出来
24 select * from employee left join department on department.id=employee.dep_id
25 union 联合的意思,将2表联合,去掉重复的部分
26 select * from employee right join department on department.id=employee.dep_id;
27
28 ** 优先级
29 from > join on 后面的条件联合 > where > gourp by > 聚合 > having > order by > limit > select
30
31 三 子查询
32 3.1 带in关键字的子查询
33 select id from department where name in ('技术','销售'); 查看2个部门的d
34 select * from employee where dep_id in (200.202); 查询部门id对应的员工
35 问题;怎么知道员工的id,应该是从部门表得到id
36 select * from employee where dep_id in (select id from department where name in ('技术','销售'));
37
38 3.2 练习
39 1. 查询平均年龄在25岁以上的部门名
40 select name from department where id in (select dep_id from employee group by dep_id having avg(age) > 25);
41 2. 查看技术部员工姓名
42 select name from employee where dep_id in (select id from department where name='技术');
43 3. 查看人数小于2的部门,不准确,因为还有空的存在,看 4 5 题
44 select name from department where id in (select dep_id from employee group by dep_id having count(dep_id) < 2);
45 4. 提取空部门,就是没有人的部门
46 select distinct dep_id from employee; 去重,有人的部门
47 select name from department where id not in (select distinct dep_id from employee); 非有人的部门,空部门
48 5. 查看人数小于2的部门,但是有空的情况
49 select name from department where id in (select dep_id from employee group by dep_id having count(dep_id) < 2)
50 union
51 select name from department where id not in (select distinct dep_id from employee); 没有人的部门
52
53 或者:
54 select name from department where id in
55 (
56 select dep_id from employee group by dep_id having count(id) < 2
57 union
58 select id from department where id not in (select distinct dep_id from employee)
59 );
60
61 3.3 exists存在,判断
62 select * from employee where exists (select id from department where name='技术');
63 select id from department where name='技术' 存在 就打印前面的select,不存在就返回空
64
65 ** 子查询,括号内的查询都要select出2表相同的字段,这样才可以,语法:查询 where 相同字段 in 子查询
三 总结&表创建过程
1 总结;
2 查询语法
3 SELECT 字段1,字段2... FROM 表名
4 JOIN ON
5 WHERE 条件
6 GROUP BY field
7 HAVING 筛选
8 ORDER BY field
9 LIMIT 限制条数
10
11 关键字的执行优先级
12 from
13 join on
14 where
15 ---------------
16 group by
17 ---------------
18 聚合
19 having
20 select
21 ---------------
22 限制打印
23 distinct
24 order by
25 limit
单表创建
1 #创建表
2 create table employee(
3 id int not null unique auto_increment,
4 name varchar(20) not null,
5 sex enum('male','female') not null default 'male', #大部分是男的
6 age int(3) unsigned not null default 28,
7 hire_date date not null,
8 post varchar(50),
9 post_comment varchar(100),
10 salary double(15,2),
11 office int, #一个部门一个屋子
12 depart_id int
13 );
14
15 #插入记录
16 #三个部门:教学,销售,运营
17 insert into employee(name,sex,age,hire_date,post,salary,office,depart_id) values
18 ('egon','male',18,'20170301','老男孩驻沙河办事处外交大使',7300.33,401,1), #以下是教学部
19 ('alex','male',78,'20150302','teacher',1000000.31,401,1),
20 ('wupeiqi','male',81,'20130305','teacher',8300,401,1),
21 ('yuanhao','male',73,'20140701','teacher',3500,401,1),
22 ('liwenzhou','male',28,'20121101','teacher',2100,401,1),
23 ('jingliyang','female',18,'20110211','teacher',9000,401,1),
24 ('jinxin','male',18,'19000301','teacher',30000,401,1),
25 ('成龙','male',48,'20101111','teacher',10000,401,1),
26
27 ('歪歪','female',48,'20150311','sale',3000.13,402,2),#以下是销售部门
28 ('丫丫','female',38,'20101101','sale',2000.35,402,2),
29 ('丁丁','female',18,'20110312','sale',1000.37,402,2),
30 ('星星','female',18,'20160513','sale',3000.29,402,2),
31 ('格格','female',28,'20170127','sale',4000.33,402,2),
32
33 ('张野','male',28,'20160311','operation',10000.13,403,3), #以下是运营部门
34 ('程咬金','male',18,'19970312','operation',20000,403,3),
35 ('程咬银','female',18,'20130311','operation',19000,403,3),
36 ('程咬铜','male',18,'20150411','operation',18000,403,3),
37 ('程咬铁','female',18,'20140512','operation',17000,403,3)
38 ;
多表创建
1 多表查询
2 整体是一个数据,方便管理,节省空间拆成不同的表
3 company.employee
4 company.department
5
6 ==============================被关联表
7 创建 被关联表
8 create table department(
9 id int,
10 name varchar(20)
11 )charset utf8;
12 插入数据
13 insert into department values
14 (200,'技术'),
15 (201,'人力资源'),
16 (202,'销售'),
17 (203,'运营');
18
19
20 ===============================关联表
21 创建 关联表
22 create table employee(
23 id int primary key auto_increment,
24 name varchar(20),
25 sex enum('male','female') not null default 'male',
26 age int,
27 dep_id int
28 )charset utf8;
29 插入数据
30 insert into employee(name,sex,age,dep_id) values
31 ('egon','male',18,200),
32 ('alex','female',48,201),
33 ('wupeiqi','male',38,201),
34 ('yuanhao','female',28,202),
35 ('liwenzhou','male',18,200),
36 ('jingliyang','female',18,204)
37 ;
38
39 ** 暂时不加入外键,如果假如外键,关联表在拆入数据的时候,插入204,上面的表没有下面的表对应关系
40 上有下没有,下有上没有,就会报错
41 上表有203下面没有,下面204上面没有
42 ** 虽然没有外键,咱们假设department id 和 employee dep_id 关联