一 介绍
本节主题
- 多表连接查询
- 复合条件连接查询
- 子查询
准备表
#建表 create table department( id int, name varchar(20) ); create table employee( 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 department values (200,‘技术‘), (201,‘人力资源‘), (202,‘销售‘), (203,‘运营‘); insert into employee(name,sex,age,dep_id) values (‘egon‘,‘male‘,18,200), (‘alex‘,‘female‘,48,201), (‘wupeiqi‘,‘male‘,38,201), (‘yuanhao‘,‘female‘,28,202), (‘liwenzhou‘,‘male‘,18,200), (‘jingliyang‘,‘female‘,18,204) ; #查看表结构和数据 mysql> desc department; +-------+-------------+------+-----+---------+-------+ | Field | Type | Null | Key | Default | Extra | +-------+-------------+------+-----+---------+-------+ | id | int(11) | YES | | NULL | | | name | varchar(20) | YES | | NULL | | +-------+-------------+------+-----+---------+-------+ mysql> desc employee; +--------+-----------------------+------+-----+---------+----------------+ | Field | Type | Null | Key | Default | Extra | +--------+-----------------------+------+-----+---------+----------------+ | id | int(11) | NO | PRI | NULL | auto_increment | | name | varchar(20) | YES | | NULL | | | sex | enum(‘male‘,‘female‘) | NO | | male | | | age | int(11) | YES | | NULL | | | dep_id | int(11) | YES | | NULL | | +--------+-----------------------+------+-----+---------+----------------+ mysql> select * from department; +------+--------------+ | id | name | +------+--------------+ | 200 | 技术 | | 201 | 人力资源 | | 202 | 销售 | | 203 | 运营 | +------+--------------+ mysql> select * from employee; +----+------------+--------+------+--------+ | id | name | sex | age | dep_id | +----+------------+--------+------+--------+ | 1 | egon | male | 18 | 200 | | 2 | alex | female | 48 | 201 | | 3 | wupeiqi | male | 38 | 201 | | 4 | yuanhao | female | 28 | 202 | | 5 | liwenzhou | male | 18 | 200 | | 6 | jingliyang | female | 18 | 204 | +----+------------+--------+------+--------+
二 多表连接查询
#重点:外链接语法 SELECT 字段列表 FROM 表1 INNER|LEFT|RIGHT JOIN 表2 ON 表1.字段 = 表2.字段;
1 交叉连接:不适用任何匹配条件。生成笛卡尔积
mysql> select * from employee,department; +----+------------+--------+------+--------+------+--------------+ | id | name | sex | age | dep_id | id | name | +----+------------+--------+------+--------+------+--------------+ | 1 | egon | male | 18 | 200 | 200 | 技术 | | 1 | egon | male | 18 | 200 | 201 | 人力资源 | | 1 | egon | male | 18 | 200 | 202 | 销售 | | 1 | egon | male | 18 | 200 | 203 | 运营 | | 2 | alex | female | 48 | 201 | 200 | 技术 | | 2 | alex | female | 48 | 201 | 201 | 人力资源 | | 2 | alex | female | 48 | 201 | 202 | 销售 | | 2 | alex | female | 48 | 201 | 203 | 运营 | | 3 | wupeiqi | male | 38 | 201 | 200 | 技术 | | 3 | wupeiqi | male | 38 | 201 | 201 | 人力资源 | | 3 | wupeiqi | male | 38 | 201 | 202 | 销售 | | 3 | wupeiqi | male | 38 | 201 | 203 | 运营 | | 4 | yuanhao | female | 28 | 202 | 200 | 技术 | | 4 | yuanhao | female | 28 | 202 | 201 | 人力资源 | | 4 | yuanhao | female | 28 | 202 | 202 | 销售 | | 4 | yuanhao | female | 28 | 202 | 203 | 运营 | | 5 | liwenzhou | male | 18 | 200 | 200 | 技术 | | 5 | liwenzhou | male | 18 | 200 | 201 | 人力资源 | | 5 | liwenzhou | male | 18 | 200 | 202 | 销售 | | 5 | liwenzhou | male | 18 | 200 | 203 | 运营 | | 6 | jingliyang | female | 18 | 204 | 200 | 技术 | | 6 | jingliyang | female | 18 | 204 | 201 | 人力资源 | | 6 | jingliyang | female | 18 | 204 | 202 | 销售 | | 6 | jingliyang | female | 18 | 204 | 203 | 运营 | +----+------------+--------+------+--------+------+--------------+
2 内连接:只连接匹配的行
#找两张表共有的部分,相当于利用条件从笛卡尔积结果中筛选出了正确的结果 #department没有204这个部门,因而employee表中关于204这条员工信息没有匹配出来 mysql> select employee.id,employee.name,employee.age,employee.sex,department.name from employee inner join department on employee.dep_id=department.id; +----+-----------+------+--------+--------------+ | id | name | age | sex | name | +----+-----------+------+--------+--------------+ | 1 | egon | 18 | male | 技术 | | 2 | alex | 48 | female | 人力资源 | | 3 | wupeiqi | 38 | male | 人力资源 | | 4 | yuanhao | 28 | female | 销售 | | 5 | liwenzhou | 18 | male | 技术 | +----+-----------+------+--------+--------------+ #上述sql等同于 mysql> select employee.id,employee.name,employee.age,employee.sex,department.name from employee,department where employee.dep_id=department.id;
3 外链接之左连接:优先显示左表全部记录
#以左表为准,即找出所有员工信息,当然包括没有部门的员工 #本质就是:在内连接的基础上增加左边有右边没有的结果 mysql> select employee.id,employee.name,department.name as depart_name from employee left join department on employee.dep_id=department.id; +----+------------+--------------+ | id | name | depart_name | +----+------------+--------------+ | 1 | egon | 技术 | | 5 | liwenzhou | 技术 | | 2 | alex | 人力资源 | | 3 | wupeiqi | 人力资源 | | 4 | yuanhao | 销售 | | 6 | jingliyang | NULL | +----+------------+--------------+
4 外链接之右连接:优先显示右表全部记录
#以右表为准,即找出所有部门信息,包括没有员工的部门 #本质就是:在内连接的基础上增加右边有左边没有的结果 mysql> select employee.id,employee.name,department.name as depart_name from employee right join department on employee.dep_id=department.id; +------+-----------+--------------+ | id | name | depart_name | +------+-----------+--------------+ | 1 | egon | 技术 | | 2 | alex | 人力资源 | | 3 | wupeiqi | 人力资源 | | 4 | yuanhao | 销售 | | 5 | liwenzhou | 技术 | | NULL | NULL | 运营 | +------+-----------+--------------+
5 全外连接:显示左右两个表全部记录
全外连接:在内连接的基础上增加左边有右边没有的和右边有左边没有的结果 #注意:mysql不支持全外连接 full JOIN #强调:mysql可以使用此种方式间接实现全外连接 select * from employee left join department on employee.dep_id = department.id union select * from employee right join department on employee.dep_id = department.id ; #查看结果 +------+------------+--------+------+--------+------+--------------+ | id | name | sex | age | dep_id | id | name | +------+------------+--------+------+--------+------+--------------+ | 1 | egon | male | 18 | 200 | 200 | 技术 | | 5 | liwenzhou | male | 18 | 200 | 200 | 技术 | | 2 | alex | female | 48 | 201 | 201 | 人力资源 | | 3 | wupeiqi | male | 38 | 201 | 201 | 人力资源 | | 4 | yuanhao | female | 28 | 202 | 202 | 销售 | | 6 | jingliyang | female | 18 | 204 | NULL | NULL | | NULL | NULL | NULL | NULL | NULL | 203 | 运营 | +------+------------+--------+------+--------+------+--------------+ #注意 union与union all的区别:union会去掉相同的纪录
三 符合条件连接查询
#示例1:以内连接的方式查询employee和department表,并且employee表中的age字段值必须大于25,即找出年龄大于25岁的员工以及员工所在的部门 select employee.name,department.name from employee inner join department on employee.dep_id = department.id where age > 25; #示例2:以内连接的方式查询employee和department表,并且以age字段的升序方式显示 select employee.id,employee.name,employee.age,department.name from employee,department where employee.dep_id = department.id and age > 25 order by age asc;
四 子查询
#1:子查询是将一个查询语句嵌套在另一个查询语句中。 #2:内层查询语句的查询结果,可以为外层查询语句提供查询条件。 #3:子查询中可以包含:IN、NOT IN、ANY、ALL、EXISTS 和 NOT EXISTS等关键字 #4:还可以包含比较运算符:= 、 !=、> 、<等
1 带IN关键字的子查询
#查询平均年龄在25岁以上的部门名 select id,name from department where id in (select dep_id from employee group by dep_id having avg(age) > 25); #查看技术部员工姓名 select name from employee where dep_id in (select id from department where name=‘技术‘); #查看不足1人的部门名(子查询得到的是有人的部门id)
select name from department where id not in (select distinct dep_id from employee);
2 带比较运算符的子查询
#比较运算符:=、!=、>、>=、<、<=、<> #查询大于所有人平均年龄的员工名与年龄 mysql> select name,age from emp where age > (select avg(age) from emp); +---------+------+ | name | age | +---------+------+ | alex | 48 | | wupeiqi | 38 | +---------+------+ 2 rows in set (0.00 sec) #查询大于部门内平均年龄的员工名、年龄 select t1.name,t1.age from emp t1 inner join (select dep_id,avg(age) avg_age from emp group by dep_id) t2 on t1.dep_id = t2.dep_id where t1.age > t2.avg_age;
3 带EXISTS关键字的子查询
EXISTS关字键字表示存在。在使用EXISTS关键字时,内层查询语句不返回查询的记录。
而是返回一个真假值。True或False
当返回True时,外层查询语句将进行查询;当返回值为False时,外层查询语句不进行查询
#department表中存在dept_id=203,Ture mysql> select * from employee -> where exists -> (select id from department where id=200); +----+------------+--------+------+--------+ | id | name | sex | age | dep_id | +----+------------+--------+------+--------+ | 1 | egon | male | 18 | 200 | | 2 | alex | female | 48 | 201 | | 3 | wupeiqi | male | 38 | 201 | | 4 | yuanhao | female | 28 | 202 | | 5 | liwenzhou | male | 18 | 200 | | 6 | jingliyang | female | 18 | 204 | +----+------------+--------+------+--------+ #department表中存在dept_id=205,False mysql> select * from employee -> where exists -> (select id from department where id=204); Empty set (0.00 sec)
练习:查询每个部门最新入职的那位员工
company.employee 员工id id int 姓名 emp_name varchar 性别 sex enum 年龄 age int 入职日期 hire_date date 岗位 post varchar 职位描述 post_comment varchar 薪水 salary double 办公室 office int 部门编号 depart_id int #创建表 create table employee( 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 ); #查看表结构 mysql> desc employee; +--------------+-----------------------+------+-----+---------+----------------+ | Field | Type | Null | Key | Default | Extra | +--------------+-----------------------+------+-----+---------+----------------+ | id | int(11) | NO | PRI | NULL | auto_increment | | name | varchar(20) | NO | | NULL | | | sex | enum(‘male‘,‘female‘) | NO | | male | | | age | int(3) unsigned | NO | | 28 | | | hire_date | date | NO | | NULL | | | post | varchar(50) | YES | | NULL | | | post_comment | varchar(100) | YES | | NULL | | | salary | double(15,2) | YES | | NULL | | | office | int(11) | YES | | NULL | | | depart_id | int(11) | YES | | NULL | | +--------------+-----------------------+------+-----+---------+----------------+ #插入记录 #三个部门:教学,销售,运营 insert into employee(name,sex,age,hire_date,post,salary,office,depart_id) values (‘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 (‘alex‘,‘male‘,78,‘20150302‘,‘teacher‘,1000000.31,401,1), (‘wupeiqi‘,‘male‘,81,‘20130305‘,‘teacher‘,8300,401,1), (‘yuanhao‘,‘male‘,73,‘20140701‘,‘teacher‘,3500,401,1), (‘liwenzhou‘,‘male‘,28,‘20121101‘,‘teacher‘,2100,401,1), (‘jingliyang‘,‘female‘,18,‘20110211‘,‘teacher‘,9000,401,1), (‘jinxin‘,‘male‘,18,‘19000301‘,‘teacher‘,30000,401,1), (‘成龙‘,‘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 AS t1 INNER JOIN ( SELECT post, max(hire_date) max_date FROM emp GROUP BY post ) AS t2 ON t1.post = t2.post WHERE t1.hire_date = t2.max_date;
mysql> select (select t2.name from emp as t2 where t2.post=t1.post order by hire_date desc limit 1) from emp as t1 group by post; +---------------------------------------------------------------------------------------+ | (select t2.name from emp as t2 where t2.post=t1.post order by hire_date desc limit 1) | +---------------------------------------------------------------------------------------+ | 张野 | | 格格 | | alex | | egon | +---------------------------------------------------------------------------------------+ rows in set (0.00 sec) mysql> select (select t2.id from emp as t2 where t2.post=t1.post order by hire_date desc limit 1) from emp as t1 group by post; +-------------------------------------------------------------------------------------+ | (select t2.id from emp as t2 where t2.post=t1.post order by hire_date desc limit 1) | +-------------------------------------------------------------------------------------+ | 14 | | 13 | | 2 | | 1 | +-------------------------------------------------------------------------------------+ rows in set (0.00 sec) #正确答案 mysql> select t3.name,t3.post,t3.hire_date from emp as t3 where id in (select (select id from emp as t2 where t2.post=t1.post order by hire_date desc limit 1) from emp as t1 group by post); +--------+-----------------------------------------+------------+ | name | post | hire_date | +--------+-----------------------------------------+------------+ | egon | 老男孩驻沙河办事处外交大使 | 2017-03-01 | | alex | teacher | 2015-03-02 | | 格格 | sale | 2017-01-27 | | 张野 | operation | 2016-03-11 | +--------+-----------------------------------------+------------+ rows in set (0.00 sec)
答案一为正确答案,答案二中的limit 1有问题(每个部门可能有>1个为同一时间入职的新员工),我只是想用该例子来说明可以在select后使用子查询
可以基于上述方法解决:比如某网站在全国各个市都有站点,每个站点一条数据,想取每个省下最新的那一条市的网站质量信息
五 综合练习
init.sql文件内容
/* 数据导入: Navicat Premium Data Transfer Source Server : localhost Source Server Type : MySQL Source Server Version : 50624 Source Host : localhost Source Database : sqlexam Target Server Type : MySQL Target Server Version : 50624 File Encoding : utf-8 Date: 10/21/2016 06:46:46 AM */ SET NAMES utf8; SET FOREIGN_KEY_CHECKS = 0; -- ---------------------------- -- Table structure for `class` -- ---------------------------- DROP TABLE IF EXISTS `class`; CREATE TABLE `class` ( `cid` int(11) NOT NULL AUTO_INCREMENT, `caption` varchar(32) NOT NULL, PRIMARY KEY (`cid`) ) ENGINE=InnoDB AUTO_INCREMENT=5 DEFAULT CHARSET=utf8; -- ---------------------------- -- Records of `class` -- ---------------------------- BEGIN; INSERT INTO `class` VALUES (‘1‘, ‘三年二班‘), (‘2‘, ‘三年三班‘), (‘3‘, ‘一年二班‘), (‘4‘, ‘二年九班‘); COMMIT; -- ---------------------------- -- Table structure for `course` -- ---------------------------- DROP TABLE IF EXISTS `course`; CREATE TABLE `course` ( `cid` int(11) NOT NULL AUTO_INCREMENT, `cname` varchar(32) NOT NULL, `teacher_id` int(11) NOT NULL, PRIMARY KEY (`cid`), KEY `fk_course_teacher` (`teacher_id`), CONSTRAINT `fk_course_teacher` FOREIGN KEY (`teacher_id`) REFERENCES `teacher` (`tid`) ) ENGINE=InnoDB AUTO_INCREMENT=5 DEFAULT CHARSET=utf8; -- ---------------------------- -- Records of `course` -- ---------------------------- BEGIN; INSERT INTO `course` VALUES (‘1‘, ‘生物‘, ‘1‘), (‘2‘, ‘物理‘, ‘2‘), (‘3‘, ‘体育‘, ‘3‘), (‘4‘, ‘美术‘, ‘2‘); COMMIT; -- ---------------------------- -- Table structure for `score` -- ---------------------------- DROP TABLE IF EXISTS `score`; CREATE TABLE `score` ( `sid` int(11) NOT NULL AUTO_INCREMENT, `student_id` int(11) NOT NULL, `course_id` int(11) NOT NULL, `num` int(11) NOT NULL, PRIMARY KEY (`sid`), KEY `fk_score_student` (`student_id`), KEY `fk_score_course` (`course_id`), CONSTRAINT `fk_score_course` FOREIGN KEY (`course_id`) REFERENCES `course` (`cid`), CONSTRAINT `fk_score_student` FOREIGN KEY (`student_id`) REFERENCES `student` (`sid`) ) ENGINE=InnoDB AUTO_INCREMENT=53 DEFAULT CHARSET=utf8; -- ---------------------------- -- Records of `score` -- ---------------------------- BEGIN; INSERT INTO `score` VALUES (‘1‘, ‘1‘, ‘1‘, ‘10‘), (‘2‘, ‘1‘, ‘2‘, ‘9‘), (‘5‘, ‘1‘, ‘4‘, ‘66‘), (‘6‘, ‘2‘, ‘1‘, ‘8‘), (‘8‘, ‘2‘, ‘3‘, ‘68‘), (‘9‘, ‘2‘, ‘4‘, ‘99‘), (‘10‘, ‘3‘, ‘1‘, ‘77‘), (‘11‘, ‘3‘, ‘2‘, ‘66‘), (‘12‘, ‘3‘, ‘3‘, ‘87‘), (‘13‘, ‘3‘, ‘4‘, ‘99‘), (‘14‘, ‘4‘, ‘1‘, ‘79‘), (‘15‘, ‘4‘, ‘2‘, ‘11‘), (‘16‘, ‘4‘, ‘3‘, ‘67‘), (‘17‘, ‘4‘, ‘4‘, ‘100‘), (‘18‘, ‘5‘, ‘1‘, ‘79‘), (‘19‘, ‘5‘, ‘2‘, ‘11‘), (‘20‘, ‘5‘, ‘3‘, ‘67‘), (‘21‘, ‘5‘, ‘4‘, ‘100‘), (‘22‘, ‘6‘, ‘1‘, ‘9‘), (‘23‘, ‘6‘, ‘2‘, ‘100‘), (‘24‘, ‘6‘, ‘3‘, ‘67‘), (‘25‘, ‘6‘, ‘4‘, ‘100‘), (‘26‘, ‘7‘, ‘1‘, ‘9‘), (‘27‘, ‘7‘, ‘2‘, ‘100‘), (‘28‘, ‘7‘, ‘3‘, ‘67‘), (‘29‘, ‘7‘, ‘4‘, ‘88‘), (‘30‘, ‘8‘, ‘1‘, ‘9‘), (‘31‘, ‘8‘, ‘2‘, ‘100‘), (‘32‘, ‘8‘, ‘3‘, ‘67‘), (‘33‘, ‘8‘, ‘4‘, ‘88‘), (‘34‘, ‘9‘, ‘1‘, ‘91‘), (‘35‘, ‘9‘, ‘2‘, ‘88‘), (‘36‘, ‘9‘, ‘3‘, ‘67‘), (‘37‘, ‘9‘, ‘4‘, ‘22‘), (‘38‘, ‘10‘, ‘1‘, ‘90‘), (‘39‘, ‘10‘, ‘2‘, ‘77‘), (‘40‘, ‘10‘, ‘3‘, ‘43‘), (‘41‘, ‘10‘, ‘4‘, ‘87‘), (‘42‘, ‘11‘, ‘1‘, ‘90‘), (‘43‘, ‘11‘, ‘2‘, ‘77‘), (‘44‘, ‘11‘, ‘3‘, ‘43‘), (‘45‘, ‘11‘, ‘4‘, ‘87‘), (‘46‘, ‘12‘, ‘1‘, ‘90‘), (‘47‘, ‘12‘, ‘2‘, ‘77‘), (‘48‘, ‘12‘, ‘3‘, ‘43‘), (‘49‘, ‘12‘, ‘4‘, ‘87‘), (‘52‘, ‘13‘, ‘3‘, ‘87‘); COMMIT; -- ---------------------------- -- Table structure for `student` -- ---------------------------- DROP TABLE IF EXISTS `student`; CREATE TABLE `student` ( `sid` int(11) NOT NULL AUTO_INCREMENT, `gender` char(1) NOT NULL, `class_id` int(11) NOT NULL, `sname` varchar(32) NOT NULL, PRIMARY KEY (`sid`), KEY `fk_class` (`class_id`), CONSTRAINT `fk_class` FOREIGN KEY (`class_id`) REFERENCES `class` (`cid`) ) ENGINE=InnoDB AUTO_INCREMENT=17 DEFAULT CHARSET=utf8; -- ---------------------------- -- Records of `student` -- ---------------------------- BEGIN; INSERT INTO `student` VALUES (‘1‘, ‘男‘, ‘1‘, ‘理解‘), (‘2‘, ‘女‘, ‘1‘, ‘钢蛋‘), (‘3‘, ‘男‘, ‘1‘, ‘张三‘), (‘4‘, ‘男‘, ‘1‘, ‘张一‘), (‘5‘, ‘女‘, ‘1‘, ‘张二‘), (‘6‘, ‘男‘, ‘1‘, ‘张四‘), (‘7‘, ‘女‘, ‘2‘, ‘铁锤‘), (‘8‘, ‘男‘, ‘2‘, ‘李三‘), (‘9‘, ‘男‘, ‘2‘, ‘李一‘), (‘10‘, ‘女‘, ‘2‘, ‘李二‘), (‘11‘, ‘男‘, ‘2‘, ‘李四‘), (‘12‘, ‘女‘, ‘3‘, ‘如花‘), (‘13‘, ‘男‘, ‘3‘, ‘刘三‘), (‘14‘, ‘男‘, ‘3‘, ‘刘一‘), (‘15‘, ‘女‘, ‘3‘, ‘刘二‘), (‘16‘, ‘男‘, ‘3‘, ‘刘四‘); COMMIT; -- ---------------------------- -- Table structure for `teacher` -- ---------------------------- DROP TABLE IF EXISTS `teacher`; CREATE TABLE `teacher` ( `tid` int(11) NOT NULL AUTO_INCREMENT, `tname` varchar(32) NOT NULL, PRIMARY KEY (`tid`) ) ENGINE=InnoDB AUTO_INCREMENT=6 DEFAULT CHARSET=utf8; -- ---------------------------- -- Records of `teacher` -- ---------------------------- BEGIN; INSERT INTO `teacher` VALUES (‘1‘, ‘张磊老师‘), (‘2‘, ‘李平老师‘), (‘3‘, ‘刘海燕老师‘), (‘4‘, ‘朱云海老师‘), (‘5‘, ‘李杰老师‘); COMMIT; SET FOREIGN_KEY_CHECKS = 1;
从init.sql文件中导入数据
#准备表、记录 mysql> create database db1; mysql> use db1; mysql> source /root/init.sql
表结构为
!!!重中之重:练习之前务必搞清楚sql逻辑查询语句的执行顺序
链接:http://www.cnblogs.com/6324TV/p/8481156.html
标签:www. full join 入职 查询条件 nod 列表 teacher 例子 splay
原文地址:https://www.cnblogs.com/6324TV/p/8481147.html
踩
(0)
赞
(0)
举报
评论 一句话评论(0)