관련 무료 학습 권장 사항: mysql 비디오 튜토리얼
1.查询语句的基本操作 - select - from - where:约束条件 - group by:分组 - having:过滤 - distinct:去重 - order by:排序 - limit:限制查询记录的数量 - 聚合函数: count(计数) max(最大值) min(最小值) avg(平均值) sum(求和)
# 创建一张部门表 create table emp( id int not null unique auto_increment, name varchar(20) not null, sex enum('male','female') not null default 'male', #大部分是男的 age int(3) unsigned not null default 28, hire_date date not null, post varchar(50), post_comment varchar(100), salary double(15,2), office int, # 一个部门一个屋子 depart_id int ); # 插入记录 # 三个部门:教学,销售,运营 insert into emp(name,sex,age,hire_date,post,salary,office,depart_id) values ('tank','male',17,'20170301','张江第一帅形象代言部门',7300.33,401,1), # 以下是教学部 ('egon','male',78,'20150302','teacher',1000000.31,401,1), ('kevin','male',81,'20130305','teacher',8300,401,1), ('jason','male',73,'20140701','teacher',3500,401,1), ('owen','male',28,'20121101','teacher',2100,401,1), ('jerry','female',18,'20110211','teacher',9000,401,1), ('大饼','male',18,'19000301','teacher',30000,401,1), ('sean','male',48,'20101111','teacher',10000,401,1), ('歪歪','female',48,'20150311','sale',3000.13,402,2),# 以下是销售部门 ('丫丫','female',38,'20101101','sale',2000.35,402,2), ('丁丁','female',18,'20110312','sale',1000.37,402,2), ('星星','female',18,'20160513','sale',3000.29,402,2), ('格格','female',28,'20170127','sale',4000.33,402,2), ('张野','male',28,'20160311','operation',10000.13,403,3), # 以下是运营部门 ('程咬金','male',18,'19970312','operation',20000,403,3), ('程咬银','female',18,'20130311','operation',19000,403,3), ('程咬铜','male',18,'20150411','operation',18000,403,3), ('程咬铁','female',18,'20140512','operation',17000,403,3); # PS:如果在windows系统中,插入中文字符,select的结果为空白,可以将所有字符编码统一设置成gbk - select * from emp; # 若数据比较多,比较凌乱,可以在表后面+ \G - select * from emp\G
요점:
- select * from emp; # 若数据比较多,比较凌乱,可以在表后面+ \G - select * from emp\G
写SQL语句必须遵循两点: - 书写顺序: # 获取id为 4、5的两条记录 select * from emp where id > 3 and id < 6; - select - from - where - 执行顺序: 比如: 图书管理员,得先找到是哪一个图书馆(哪张表), 再找这本书在图书馆的哪个位置(哪一条记录), 最后查找这个本书中某一页(哪些字段值); select * from emp where id > 3 and id < 6; - from ---> 找到图书馆 - where ---> 找到书的位置 - select ---> 找到书本中的某一页 注意: 必须记住SQL语句的 书写顺序 与 执行顺序(*******);
# PS: 根据执行顺序来书写 SQL语句,一步一步来写; # 1.查询id大于等于3小于等于6的数据(and:与) select * from emp where id >= 3 and id <= 6; mysql> SELECT * FROM EMP WHERE ID > 3 AND ID <= 6; +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ | id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id | +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ | 4 | jason | male | 73 | 2014-07-01 | teacher | NULL | 3500.00 | 401 | 1 | | 5 | owen | male | 28 | 2012-11-01 | teacher | NULL | 2100.00 | 401 | 1 | | 6 | jerry | female | 18 | 2011-02-11 | teacher | NULL | 9000.00 | 401 | 1 | +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ 3 rows in set (0.00 sec) #可以使用between()and() :两者之间 mysql> select * from emp where id between 3 and 6; +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ | id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id | +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ | 3 | kevin | male | 81 | 2013-03-05 | teacher | NULL | 8300.00 | 401 | 1 | | 4 | jason | male | 73 | 2014-07-01 | teacher | NULL | 3500.00 | 401 | 1 | | 5 | owen | male | 28 | 2012-11-01 | teacher | NULL | 2100.00 | 401 | 1 | | 6 | jerry | female | 18 | 2011-02-11 | teacher | NULL | 9000.00 | 401 | 1 | +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ 4 rows in set (0.00 sec) # 2.查询薪资是20000或者18000或者17000的数据 # or: 或者 select * from emp where salary=20000 or salary=18000 or salary=17000; # in: 在什么里 select * from emp where salary in (20000, 18000, 17000); mysql> select * from emp where salary = 20000 or salary = 18000 or salary = 17000; +----+-----------+--------+-----+------------+-----------+--------------+----------+--------+-----------+ | id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id | +----+-----------+--------+-----+------------+-----------+--------------+----------+--------+-----------+ | 15 | 程咬金 | male | 18 | 1997-03-12 | operation | NULL | 20000.00 | 403 | 3 | | 17 | 程咬铜 | male | 18 | 2015-04-11 | operation | NULL | 18000.00 | 403 | 3 | | 18 | 程咬铁 | female | 18 | 2014-05-12 | operation | NULL | 17000.00 | 403 | 3 | +----+-----------+--------+-----+------------+-----------+--------------+----------+--------+-----------+ 3 rows in set (0.00 sec) # 也可以使用in(3,4,5,6) select * from emp where id in(3,4,5,6); +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ | id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id | +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ | 3 | kevin | male | 81 | 2013-03-05 | teacher | NULL | 8300.00 | 401 | 1 | | 4 | jason | male | 73 | 2014-07-01 | teacher | NULL | 3500.00 | 401 | 1 | | 5 | owen | male | 28 | 2012-11-01 | teacher | NULL | 2100.00 | 401 | 1 | | 6 | jerry | female | 18 | 2011-02-11 | teacher | NULL | 9000.00 | 401 | 1 | +----+-------+--------+-----+------------+---------+--------------+---------+--------+-----------+ 4 rows in set (0.00 sec) # 3.查询员工姓名中包含o字母 的 员工姓名和薪资 # like: 模糊匹配 # %: 匹配0个或多个任意字符(若前后都有%(%o%),查所有带有索引字符的记录,若头有%而尾没有(%o),则查所有尾带有索引字符的记录,若尾有%而头没有(o%),则查所有头带有索引字符的记录,若前后都没有%(o),则只会查询只有索引字符的字符) # _: 匹配一个任意字符 select name, salary from emp where name like "%o%"; +-------+------------+ | name | salary | +-------+------------+ | egon | 1000000.31 | | jason | 3500.00 | | owen | 2100.00 | +-------+------------+ 3 rows in set (0.00 sec) mysql> select name, salary from emp where name like "o"; +------+--------+ | name | salary | +------+--------+ | o | NULL | +------+--------+ 1 row in set (0.00 sec) # 4.查找名字个数为3个的员工 名字 与 薪资 select name, salary from emp where name like "___"; +-----------+----------+ | name | salary | +-----------+----------+ | 程咬金 | 20000.00 | | 程咬银 | 19000.00 | | 程咬铜 | 18000.00 | | 程咬铁 | 17000.00 | +-----------+----------+ 4 rows in set (0.00 sec) # 或者 # char_length(name): 计算名字字符的长度 select name, salary from emp where char_length(name) = 4; mysql> select name, salary from emp where char_length(name) = 3; +-----------+----------+ | name | salary | +-----------+----------+ | 程咬金 | 20000.00 | | 程咬银 | 19000.00 | | 程咬铜 | 18000.00 | | 程咬铁 | 17000.00 | +-----------+----------+ 4 rows in set (0.00 sec) # 5.查询id小于3或者大于6的数据 # not in: 不再什么什么中 select * from emp where id not in (3, 4, 5, 6); select * from emp where id not between 3 and 6; # 6.查询薪资不在20000,18000,17000范围的数据 select * from emp where salary not in (20000, 18000, 17000); # 7.查询岗位描述为空的 员工名 与 岗位名 post_comment # 用等于号无法判定空字段 select name, post from emp where post_comment = null; select * from emp where post_comment = null; Empty set (0.00 sec) # 注意: 针对null的值 需要使用 is select name, post from emp where post_comment is null; select name, post_comment from emp where post_comment is null; +-----------+--------------+ | name | post_comment | +-----------+--------------+ | tank | NULL | | egon | NULL | | kevin | NULL | | jason | NULL | | owen | NULL | | jerry | NULL | | 大饼 | NULL | | sean | NULL | | 歪歪 | NULL | | 丫丫 | NULL | | 丁丁 | NULL | | 星星 | NULL | | 格格 | NULL | | 张野 | NULL | | 程咬金 | NULL | | 程咬银 | NULL | | 程咬铜 | NULL | | 程咬铁 | NULL | | o | NULL | +-----------+--------------+ 19 rows in set (0.00 sec)
집계 함수 뒤에는 다음이 있어야 합니다. 그룹별 Behind(실행 순서); - 书写顺序:
- select
- from
- where
- group by
- 执行顺序:
- from
- where
- group by
- select
# 什么叫分组
# 比如: 一张员工表中有性别字段,可以根据性别分组,一组是男性,一组是女性,或者是根据部门分组,有教学部、销售部等...
# 1.根据部门分组
# 非严格模式下可以获取 分组条件post 以外的字段数据
select post, salary from emp group by post;
# 设置严格模式:
show variables like "%mode%";
# 全局设置: 永久有效
set global sql_mode="strict_trans_tables,only_full_group_by";
"""
设置sql_mode为only_full_group_by,意味着以后但凡分组,只能取到分组的依据,
不应该在去取组里面的单个元素的值,那样的话分组就没有意义了,并且会报错,因为不分组就是对单个元素信息的随意获取
"""
mysql> select post, salary from emp group by post;
ERROR 1046 (3D000): No database selected
# 可以同聚合函数,间接获取其他字段数据
聚合函数:
count: 计数
max: 最大值
min: 最小值
avg: 平均值
sum: 求和
group_concat(name): 可以将分组后的 所有名字获取并进行拼接
# 指定以:拼接
select post, group_concat(name) from emp group by post;
select post, group_concat(name, ':') from emp group by post;
2.获取每个 部门 的最高工资
select post,max(salary) from emp group by post;
mysql> select post,max(salary) from emp group by post;
+-----------------------------------+-------------+
| post | max(salary) |
+-----------------------------------+-------------+
| NULL | NULL |
| operation | 20000.00 |
| sale | 4000.33 |
| teacher | 1000000.31 |
| 张江第一帅形象代言部门 | 7300.33 |
+-----------------------------------+-------------+
5 rows in set (0.00 sec)
补充: as 别名: 可以给字段 加一个 别名
select post as '部门', max(salary) as '薪资'
from emp group by post;
# 也可以简写,但是不推荐
select post '部门', max(salary) '薪资'
from emp group by post;
mysql> select post as '部门', max(salary) as'薪资'
-> from emp group by post;
+-----------------------------------+------------+
| 部门 | 薪资 |
+-----------------------------------+------------+
| NULL | NULL |
| operation | 20000.00 |
| sale | 4000.33 |
| teacher | 1000000.31 |
| 张江第一帅形象代言部门 | 7300.33 |
+-----------------------------------+------------+
5 rows in set (0.00 sec)
3.每个部门的最低工资
select post, min(salary) from emp group by post;
4.每个部门的平均工资
select post, avg(salary) from emp group by post;
5.每个部门的工资总和
select post, sum(salary) from emp group by post;
6.每个部门的员工个数
# count(): 括号中可以填任意非空值
select post, count(salary) from emp group by post;
select post, count(post_comment) from emp group by post;
查询岗位名以及各岗位内包含的员工个数
select post, count(id) from emp group by post;
查询公司内男员工和女员工的个数
select sex, count(*) from emp group by sex;
查询男员工与男员工的平均薪资,女员工与女员工的平均薪资
select sex, avg(salary) from emp group by sex;
.统计各部门年龄在30岁以上的员工平均工资:
# 步骤: 先找到表,再找年龄30岁以上,再根据部门分组,最后求平均薪资;
select post, avg(salary) from emp where age > 30 group by post;
# 聚合函数: 若没有group by 分组,默认将查出来的数据当做一个分组, 也能使用;
select max(salary) from emp;
# group_concat(name): 可以将分组后的 所有名字获取并进行拼接
# 默认以, 拼接
select post, group_concat(name) from emp group by post;
# 指定以:拼接
select post, group_concat(name, ':') from emp group by post;
#
select post, group_concat('Name: ', name) from emp group by post;
select post, group_coucat(name) from emp group by post;
mysql> select post, group_concat(name) from emp group by post;
+-----------------------------------+------------------------------------------------+
| post | group_concat(name) |
+-----------------------------------+------------------------------------------------+
| NULL | o |
| operation | 程咬铁,程咬铜,程咬银,程咬金,张野 |
| sale | 格格,星星,丁丁,丫丫,歪歪 |
| teacher | sean,大饼,jerry,owen,jason,kevin,egon |
| 张江第一帅形象代言部门 | tank |
+-----------------------------------+------------------------------------------------+
5 rows in set (0.00 sec)
# 给每个部门的员工名字前 + NB_
select post, group_concat('NB_', name) from emp group by post;
# 拼接部门员工名字+薪资
select post, group_concat(name, ":", salary) from emp group by post;
# 4.补充concat(不分组时用)拼接字符串达到更好的显示效果 as语法并用 select concat('Name: ', name) as '名字', concat('Sal: ', salary) as '薪资' from emp; mysql> select concat('name: ', name) as '名字',concat('sal: ',salary) as '薪资' from emp; +-----------------+-----------------+ | 名字 | 薪资 | +-----------------+-----------------+ | name: tank | sal: 7300.33 | | name: egon | sal: 1000000.31 | | name: kevin | sal: 8300.00 | | name: jason | sal: 3500.00 | | name: owen | sal: 2100.00 | | name: jerry | sal: 9000.00 | | name: 大饼 | sal: 30000.00 | | name: sean | sal: 10000.00 | | name: 歪歪 | sal: 3000.13 | | name: 丫丫 | sal: 2000.35 | | name: 丁丁 | sal: 1000.37 | | name: 星星 | sal: 3000.29 | | name: 格格 | sal: 4000.33 | | name: 张野 | sal: 10000.13 | | name: 程咬金 | sal: 20000.00 | | name: 程咬银 | sal: 19000.00 | | name: 程咬铜 | sal: 18000.00 | | name: 程咬铁 | sal: 17000.00 | | name: o | NULL | +-----------------+-----------------+ 19 rows in set (0.00 sec)
1.having与where语法一样,只不过having必需要在group by后使用; 2.where 不能使用聚合函数,但having可以; - 书写顺序: - select - from - where - group by - having - 执行顺序: - from - where - group by - having - select 1、统计各 部门 年龄在30岁以上的员工平均工资,并且保留平均工资大于10000的部门; select post, avg(salary) from emp where age > 30 group by post having avg(salary) > 10000; mysql> select post, avg(salary) from emp where age > 30 group by post having avg(salary) > 10000; +---------+---------------+ | post | avg(salary) | +---------+---------------+ | teacher | 255450.077500 | +---------+---------------+ 1 row in set (0.00 sec)
# distinct: 去重 - 书写顺序: - select - distinct - from - where - group by - having 执行顺序: - from - where - group by - having - select - distinct # 注意: 查询的字段值必须是重复的才有效,只要有一个字段值是不重复的就没有效果。 # 若所查字段内有不重复的字段记录就不会去重 select distinct id, post from emp; mysql> select distinct id, post from emp; +----+-----------------------------------+ | id | post | +----+-----------------------------------+ | 1 | 张江第一帅形象代言部门 | | 2 | teacher | | 3 | teacher | | 4 | teacher | | 5 | teacher | | 6 | teacher | | 7 | teacher | | 8 | teacher | | 9 | sale | | 10 | sale | | 11 | sale | | 12 | sale | | 13 | sale | | 14 | operation | | 15 | operation | | 16 | operation | | 17 | operation | | 18 | operation | | 19 | NULL | +----+-----------------------------------+ 19 rows in set (0.00 sec) select distinct post from emp; mysql> select distinct post from emp; +-----------------------------------+ | post | +-----------------------------------+ | 张江第一帅形象代言部门 | | teacher | | sale | | operation | | NULL | +-----------------------------------+ 5 rows in set (0.00 sec)
- 书写顺序: - select - from - where - group by - having - order by 执行顺序: - from - where - group by - having - select - order by # 通过select 查出来的数据再进行排序 # order by 默认升序 # asc升序 # desc降序 # 1、根据薪资进行升序 select name, salary from emp order by salary; # 默认升序 mysql> select name, salary from emp order by salary; +-----------+------------+ | name | salary | +-----------+------------+ | o | NULL | | 丁丁 | 1000.37 | | 丫丫 | 2000.35 | | owen | 2100.00 | | 歪歪 | 3000.13 | | 星星 | 3000.29 | | jason | 3500.00 | | 格格 | 4000.33 | | tank | 7300.33 | | kevin | 8300.00 | | jerry | 9000.00 | | sean | 10000.00 | | 张野 | 10000.13 | | 程咬铁 | 17000.00 | | 程咬铜 | 18000.00 | | 程咬银 | 19000.00 | | 程咬金 | 20000.00 | | 大饼 | 30000.00 | | egon | 1000000.31 | +-----------+------------+ 19 rows in set (0.00 sec) 2.select * from emp order by salary asc; # 指定升序 3.select name, salary from emp order by salary desc; # 指定降序 mysql> select name, salary from emp order by salary desc; +-----------+------------+ | name | salary | +-----------+------------+ | egon | 1000000.31 | | 大饼 | 30000.00 | | 程咬金 | 20000.00 | | 程咬银 | 19000.00 | | 程咬铜 | 18000.00 | | 程咬铁 | 17000.00 | | 张野 | 10000.13 | | sean | 10000.00 | | jerry | 9000.00 | | kevin | 8300.00 | | tank | 7300.33 | | 格格 | 4000.33 | | jason | 3500.00 | | 星星 | 3000.29 | | 歪歪 | 3000.13 | | owen | 2100.00 | | 丫丫 | 2000.35 | | 丁丁 | 1000.37 | | o | NULL | +-----------+------------+ 19 rows in set (0.00 sec) # 先按照age升序,再按照salary降序 select age, salary from emp order by age asc, salary desc; # 统计 各部门(分组) 年龄在10岁以上的员工平均工资,并且保留平均工资大于1000的部门,然后对平均工资进行升序序 select post, avg(salary) from emp where age > 10 group by post having avg(salary) > 1000 order by avg(salary);
书写顺序: - select - from - order by - limit 执行顺序: - from - select - order by - limit # 1、从第一条开始,获取4条记录; select * from emp limit 4; mysql> select * from emp limit 4; +----+-------+------+-----+------------+-----------------------------------+--------------+------------+--------+-----------+ | id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id | +----+-------+------+-----+------------+-----------------------------------+--------------+------------+--------+-----------+ | 1 | tank | male | 17 | 2017-03-01 | 张江第一帅形象代言部门 | NULL | 7300.33 | 401 | 1 | | 2 | egon | male | 78 | 2015-03-02 | teacher | NULL | 1000000.31 | 401 | 1 | | 3 | kevin | male | 81 | 2013-03-05 | teacher | NULL | 8300.00 | 401 | 1 | | 4 | jason | male | 73 | 2014-07-01 | teacher | NULL | 3500.00 | 401 | 1 | +----+-------+------+-----+------------+-----------------------------------+--------------+------------+--------+-----------+ 4 rows in set (0.00 sec) # 2、limit可以有两个参数, 参数1:是限制的开始位置, 参数2:是从开始位置展示的条数; select * from emp limit 0, 4; select * from emp limit 4, 4; mysql> select * from emp limit 4, 4; +----+--------+--------+-----+------------+---------+--------------+----------+--------+-----------+ | id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id | +----+--------+--------+-----+------------+---------+--------------+----------+--------+-----------+ | 5 | owen | male | 28 | 2012-11-01 | teacher | NULL | 2100.00 | 401 | 1 | | 6 | jerry | female | 18 | 2011-02-11 | teacher | NULL | 9000.00 | 401 | 1 | | 7 | 大饼 | male | 18 | 1900-03-01 | teacher | NULL | 30000.00 | 401 | 1 | | 8 | sean | male | 48 | 2010-11-11 | teacher | NULL | 10000.00 | 401 | 1 | +----+--------+--------+-----+------------+---------+--------------+----------+--------+-----------+ 4 rows in set (0.00 sec) # 3、查询工资最高的人的详细信息 select * from emp order by salary limit 1; mysql> select * from emp order by salary limit 1; +----+------+------+-----+------------+------+--------------+--------+--------+-----------+ | id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id | +----+------+------+-----+------------+------+--------------+--------+--------+-----------+ | 19 | o | male | 28 | 0000-00-00 | NULL | NULL | NULL | NULL | NULL | +----+------+------+-----+------------+------+--------------+--------+--------+-----------+ 1 row in set (0.00 sec)
- 연관 쿼리
- 하위 쿼리
테이블 생성 및 데이터 삽입 준비 # 在编程中,凡是看到reg开头的,基本上都是跟正则有关
# *: 代表 0 或 多个
select * from emp where name regexp '^程.*(金|银|铜|铁)$';
mysql> select * from emp where name regexp '^程.*(金|银|铜|铁)$';
+----+-----------+--------+-----+------------+-----------+--------------+----------+--------+-----------+
| id | name | sex | age | hire_date | post | post_comment | salary | office | depart_id |
+----+-----------+--------+-----+------------+-----------+--------------+----------+--------+-----------+
| 15 | 程咬金 | male | 18 | 1997-03-12 | operation | NULL | 20000.00 | 403 | 3 |
| 16 | 程咬银 | female | 18 | 2013-03-11 | operation | NULL | 19000.00 | 403 | 3 |
| 17 | 程咬铜 | male | 18 | 2015-04-11 | operation | NULL | 18000.00 | 403 | 3 |
| 18 | 程咬铁 | female | 18 | 2014-05-12 | operation | NULL | 17000.00 | 403 | 3 |
+----+-----------+--------+-----+------------+-----------+--------------+----------+--------+-----------+
4 rows in set (0.00 sec)
#建表 create table dep2( id int, name varchar(20) ); create table emp2( id int primary key auto_increment, name varchar(20), sex enum('male','female') not null default 'male', age int, dep_id int ); #插入数据 insert into dep2 values (200,'技术'), (201,'人力资源'), (202,'销售'), (203,'运营'); insert into emp2(name,sex,age,dep_id) values ('tank','male',17,200), ('egon','female',48,201), ('kevin','male',38,201), ('jason','female',28,202), ('owen','male',18,200), ('sean','female',18,204); mysql> select * from dep2; +------+--------------+ | id | name | +------+--------------+ | 200 | 技术 | | 201 | 人力资源 | | 202 | 销售 | | 203 | 运营 | +------+--------------+ 4 rows in set (0.00 sec) mysql> select * from emp2; +----+-------+--------+------+--------+ | id | name | sex | age | dep_id | +----+-------+--------+------+--------+ | 1 | tank | male | 17 | 200 | | 2 | egon | female | 48 | 201 | | 3 | kevin | male | 38 | 201 | | 4 | jason | female | 28 | 202 | | 5 | owen | male | 18 | 200 | | 6 | sean | female | 18 | 204 | +----+-------+--------+------+--------+ # 指到了如何根据表关系对字段进行拆分,目的是为了更好的管理,表数据都存放在硬盘中,存不是目的,目的是为了取,所以我们将数据从硬盘读到内存中,接下来我们因应该将他们拼成一张表来查询更加合理; # 注意: 将拆分的表,再拼接到一起进行查询, 可以通过一张表查另一张表的数据;
# 左表的一条记录与右表的一条记录都对应一遍称之为 --> "笛卡尔积" PS: 百度科普 # 将所有的数据都对应了一遍,虽然不合理但是其中有合理的数据,现在我们需要做的就是找出合理的数据
举例:通过之前准备的数据进行多表操作 1、查询 员工 以及所在 部门 的信息; # 通过where 约束条件 select * from emp2, dep2 where emp2.dep_id = dep2.id; 2、查询 部门 为 技术部 的 员工 及 部门信息 select * from emp2, dep2 where emp2.dep_id = dep2.id and dep2.name = '技术';
위 내용은 MySQL 쿼리 문의 기본 작업을 마스터하세요.의 상세 내용입니다. 자세한 내용은 PHP 중국어 웹사이트의 기타 관련 기사를 참조하세요!