GROUP BY和HAVING子句
GROUP BY子句
用于将信息划分为更小的组
每一组行返回针对该组的单个结果
--统计每个部门的人数:
Select count(*) from emp group by deptno; --根据部门分组,并统计
Select deptno, count(*) form emp group by deptno; select deptno, avg(sal) from emp group by deptno; --每个部门的平均工资
HAVING子句
用于指定 GROUP BY 子句检索行的条件
Select deptno, avg(sal) from emp group by deptno having avg(sal) >2000; --找出平均工资大于2000的部门;
数据排序
Order by 列名1 asc | desc, 列2 asc | desc; --默认为asc升序
Select * from emp order by sal; --根据工资升序排序
Select * from emp order by sal desc; --工资降序
Select * from emp order by sal desc, hiredate asc; --按工资降序排,如果工资一样,则按日期升序排;
查询语句的执行顺序
Where ,group by , having
先根据where条件将符合要求的数据筛选出来,
再根据group by来进行分组
最后将分组之后的数据用having进行约束,得到select的结果
多表查询
Select e.ename, t.tid from emp e, test t; --笛卡尔集,交叉查询
连接查询
内连接: inner join返回多个表之间共同的数据
Select empno,ename,dname from emp,dept where emp.deptno = dept.deptno;
Select empno,ename,dname form emp inner join dept on emp.deptno = dept.deptno;
外连接:
左连接: left join 数据的结果以左表为主,右表如果在左表中不存在就显示是否为空!
select s.*, c.cid, c.grade from student s left join cs c on s.sid = c.sid
右连接: right join
全连接: full join
具体参看:IndexMan 文章 oracle连接查询详解
子查询
在一个查询中可以包含另一个查询,它可以出现在任何一个地方,外部查询得到结果,内部查询返回条件
--得到工资最高的人的信息
select * from emp where sal = (select max(sal) from emp);
--查询所有和SMITH在同一个部门的人
select ename from emp where deptno = (select deptno from emp where ename = 'SMITH');
--列出与smith在同一个部门的员工信息
Select * from emp where deptno = (select deptno form emp were ename = 'SMITH');
--与MARTIN是同一个领导的员工
Select * form emp where mrg = (select mgr from emp where ename = 'MARTIN');
--列出所有在New York办公的员工
--连接查询方式:
select * from emp inner join dept on emp.deptno = dept.deptno where LOC = 'NEW YORK';
--子查询方式:
select * from emp where deptno = (select deptno from dept where LOC = 'NEW YORK');
--列出所有在New York和DALLAS办公的员工
Select * from emp where deptno in (select deptno from dept where loc = 'NEW YORK' or loc = 'DALLAS');
Select * from emp where deptno = any (select deptno from dept where loc = 'NEW YORK' or loc = 'DALLAS');
数据的分页
--查询前5条
Select rownum, emp.* from emp where rownum <=5;
--查询第6条以后的数据
select * from emp where empno not in(select empno from emp where rownum <=5);
--查询第6-10条数据
○ Select * from (select * from emp where empno not in(select empno from emp where rownum <=5)) where rownum <=5;
○ select * from emp where empno not in(select empno from emp where rownum <=5) and rownum <= 5;
--总结
○ Select * from (select * from emp where empno not in(select empno from emp where rownum <= size * (page - 1))) where rownum <= size;
○ select * from emp where empno not in(select empno from emp where rownum <= size * (page - 1)) and rownum <= size; --排序后再分页查询
--将排序之后的数据编号(效率排第2)
select * from (select rownum rn, d.* from (select * from emp order by sal desc) d) where rn > 5 and rn < 11; --分页函数:row_number函数(效率排第1)
select row_number() over(order by sal desc) rn, emp.* from emp;
--分页
select * from (select row_number() over(order by sal desc) rn, emp.* from emp) where rn > 5 and rn < 11;