主页 > 知识库 > oracle常用sql查询语句部分集合(图文)

oracle常用sql查询语句部分集合(图文)

热门标签:服务外包 网站排名优化 铁路电话系统 地方门户网站 呼叫中心市场需求 百度竞价排名 AI电销 Linux服务器

Oracle查询语句

select * from scott.emp ;

1.--dense_rank()分析函数(查找每个部门工资最高前三名员工信息)

select * from (select deptno,ename,sal,dense_rank() over(partition by deptno order by sal desc) a from scott.emp) where a=3 order by deptno asc,sal desc ;

结果:

--rank()分析函数(运行结果与上语句相同)

select * from (select deptno,ename,sal,rank() over(partition by deptno order by sal desc) a from scott.emp ) where a=3 order by deptno asc,sal desc ;

结果:

--row_number()分析函数(运行结果与上相同)

select * from(select deptno,ename,sal,row_number() over(partition by deptno order by sal desc) a from scott.emp) where a=3 order by deptno asc,sal desc ;

--rows unbounded preceding 分析函数(显示各部门的积累工资总和)

select deptno,sal,sum(sal) over(order by deptno asc rows unbounded preceding) 积累工资总和 from scott.emp ;

结果:

--rows 整数值 preceding(显示每最后4条记录的汇总值)

select deptno,sal,sum(sal) over(order by deptno rows 3 preceding) 每4汇总值 from scott.emp ;

结果:

--rows between 1 preceding and 1 following(统计3条记录的汇总值【当前记录居中】)

select deptno,ename,sal,sum(sal) over(order by deptno rows between 1 preceding and 1 following) 汇总值 from scott.emp ;

结果:

--ratio_to_report(显示员工工资及占该部门总工资的比例)

select deptno,sal,ratio_to_report(sal) over(partition by deptno) 比例 from scott.emp ;

结果:

--查看所有用户

select * from dba_users ;

select count(*) from dba_users ;

select * from all_users ;

select * from user_users ;

select * from dba_roles ;

--查看用户系统权限

select * from dba_sys_privs ;

select * from user_users ;

--查看用户对象或角色权限

select * from dba_tab_privs ;

select * from all_tab_privs ;

select * from user_tab_privs ;

--查看用户或角色所拥有的角色

select * from dba_role_privs ;

select * from user_role_privs ;

-- rownum:查询10至12信息

select * from scott.emp a where rownum=3 and a.empno not in(select b.empno from scott.emp b where rownum=9);

结果:

--not exists;查询emp表在dept表中没有的数据

select * from scott.emp a where not exists(select * from scott.dept b where a.empno=b.deptno) ;

结果:

--rowid;查询重复数据信息

select * from scott.emp a where a.rowid>(select min(x.rowid) from scott.emp x where x.empno=a.empno);

--根据rowid来分页(一万条数据,查询10000至9980时间大概在0.03秒左右)

select * from scott.emp where rowid in(select rid from(select rownum rn,rid from(select rowid rid,empno from scott.emp order by empno desc) where rownum10)where rn>=1)order by empno desc ;

结果:

--根据分析函数分页(一万条数据,查询10000至9980时间大概在1.01秒左右)

select * from(select a.*,row_number() over(order by empno desc) rk from scott.emp a ) where rk10 and rk>=1;

结果:

--rownum分页(一万条数据,查询10000至9980时间大概在0.01秒左右)

select * from(select t.*,rownum rn from(select * from scott.emp order by empno desc)t where rownum10) where rn>=1;

select * from(select a.*,rownum rn from (select * from scott.emp) a where rownum=10) where rn>=5 ;

--left outer join:左连接

select a.*,b.* from scott.emp a left outer join scott.dept b on a.deptno=b.deptno ;

--right outer join:右连接

select a.*,b.* from scott.emp a right outer join scott.dept b on a.deptno=b.deptno ;

--inner join

select a.*,b.* from scott.emp a inner  join scott.dept b on a.deptno=b.deptno ;

--full join

select a.*,b.* from scott.emp a full join scott.dept b on a.deptno=b.deptno ;

select a.*,b.* from scott.emp a,scott.dept b where a.deptno(+)=b.deptno ;

select distinct ename,sal from scott.emp a group by sal having ;

select * from scott.dept ;

select * from scott.emp ;

--case when then end (交叉报表)

select ename,sal,case deptno when 10 then '会计部' when 20 then '研究部' when 30 then '销售部' else '其他部门' end 部门 from scott.emp ;

结果:

select ename,sal,case when sal>0 and sal1500 then '一级工资' when sal>=1500 and sal3000 then '二级工资' when sal>=3000 and sal4500 then '三级工资' else '四级工资' end 工资等级 from scott.emp order by sal desc ;

结果:

--交叉报表是使用分组函数与case结构一起实现

select 姓名,sum(case 课程 when '数学' then 分数 end)数学,sum(case 课程 when '历史' then 分数 end)历史 from 学生 group by 姓名 ;

--decode 函数

select 姓名,sum(decode(课程,'数学',分数,null))数学,sum(decode(课程,'语文',分数,null))语文,sum(decode(课程,'历史','分数',null))历史 from 学生 group by 姓名 ;

--level。。。。connect by(层次查询)

select level,emp.* from scott.emp connect by prior empno = mgr order by level ;

结果:

--sys_connect_by_path函数

select ename,sys_connect_by_path(ename,'/') from scott.emp start with mgr is null connect by prior empno=mgr ;

结果:

--start with connect by prior 语法

select lpad(ename,3*(level),'')姓名,lpad(ename,3*(level),'')姓名 from scott.emp where job>'CLERK' start with mgr is null connect by prior mgr = empno ;

--level与prior关键字

select level,emp.* from scott.emp start with ename='SCOTT' connect by prior empno=mgr;

select level,emp.* from scott.emp start with ename='SCOTT' connect by empno = prior mgr ;

结果:

--等值连接

select empno,ename,job,sal,dname from scott.emp a,scott.dept b where a.deptno=b.deptno and (a.deptno=10 or sal>2500);

结果:

--非等值连接

select a.ename,a.sal,b.grade from scott.emp a,scott.salgrade b where a.sal between b.losal and b.hisal ;

结果:

--自连接

select a.ename,a.sal,b.ename from scott.emp a,scott.emp b where a.mgr=b.empno ;

结果:

--左外连接

select a.ename,a.sal,b.ename from scott.emp a,scott.emp b where a.mgr=b.empno(+);

结果:

--多表连接

select * from scott.emp ,scott.dept,scott.salgrade where scott.emp.deptno=scott.dept.deptno and scott.emp.sal between scott.salgrade.losal and scott.salgrade.hisal ;

结果:

select * from scott.emp a join scott.dept b on a.deptno=b.deptno join scott.salgrade s on a.sal between s.losal and s.hisal where a.sal>1000;

select * from(select * from scott.emp a join scott.dept b on a.deptno=b.deptno where a.sal>1000) c join scott.salgrade s on c.sal between s.losal and s.hisal ;

--单行子查询

select * from scott.emp a where a.deptno=(select deptno from scott.dept where loc='NEW YORK');

select * from scott.emp a where a.deptno in (select deptno from scott.dept where loc='NEW YORK');

结果:

--单行子查询在 from 后

select scott.emp.*,(select deptno from scott.dept where loc='NEW YORK') a from scott.emp ;

--使用 in ,all,any 多行子查询

--in:表示等于查询出来的对应数据

select ename,job,sal,deptno from scott.emp where job in(select distinct job from scott.emp where deptno=10);

--all:表示大于所有括号中查询出来的对应的数据信息

select ename,sal,deptno from scott.emp where sal>all(select sal from scott.emp where deptno=30);

--any:表示大于括号查询出来的其中任意一个即可(只随机一个)

select ename,sal,deptno from scott.emp where sal>any(select sal from scott.emp where deptno=30);

--多列子查询

select ename,job,sal,deptno from scott.emp where(deptno,job)=(select deptno,job from scott.emp where ename='SCOTT');

select ename,job,sal,deptno from scott.emp where(sal,nvl(comm,-1)) in(select sal,nvl(comm,-1) from scott.emp where deptno=30);

--非成对比较

select ename,job,sal,deptno from scott.emp where sal in(select sal from scott.emp where deptno=30) and nvl(comm,-1) in(select nvl(comm,-1) from scott.emp where deptno=30);

--其他子查询

select ename,job,sal,deptno from scott.emp where exists(select null from scott.dept where scott.dept.deptno=scott.emp.deptno and scott.dept.loc='NEW YORK');

select ename,job,sal from scott.emp join(select deptno,avg(sal) avgsal,null from scott.emp group by deptno) dept on emp.deptno=dept.deptno where sal>dept.avgsal ;

create table scott.test(

       ename varchar(20),

       job varchar(20)

);

--drop table test ;

select * from scott.test ;

--Insert与子查询(表间数据的拷贝)

insert into scott.test(ename,job) select ename,job from scott.emp ;

--Update与子查询

update scott.test set(ename,job)=(select ename,job from scott.emp where ename='SCOTT' and deptno ='10');

--创建表时,还可以指定列名

create table scott.test_1(ename,job) as select ename,job from scott.emp ;

select * from scott.test_1 ;

--delete与子查询

delete from scott.test where ename in('');

--合并查询

--union语法(合并且去除重复行,且排序)

select ename,sal,deptno from scott.emp where deptno>10 union select ename,sal,deptno from scott.emp where deptno30 ;

select a.deptno from scott.emp a union select b.deptno from scott.dept b ;

--union all(直接将两个结果集合并,不排序)

select ename,sal,deptno from scott.emp where deptno>10 union all select ename,sal,deptno from scott.emp where deptno30 ;

select a.deptno from scott.emp a union all select b.deptno from scott.dept b ;

--intersect:取交集

select ename,sal,deptno from scott.emp where deptno>10 intersect select ename,sal,deptno from scott.emp where deptno30;

--显示部门工资总和高于雇员工资总和三分之一的部门名及工资总和

select dname as 部门,sum(sal) as 工资总和 from scott.emp a,scott.dept b where a.deptno=b.deptno group by dname having sum(sal)>(select sum(sal)/3 from scott.emp c,scott.dept d where c.deptno=d.deptno);

结果:

--使用with得到以上同样的结果

with test as (select dname ,sum(sal) sumsal  from scott.emp ,scott.dept where scott.emp.deptno=scott.dept.deptno group by dname) select dname as 部门,sumsal as 工资总和 from scott.test where sumsal>(select sum(sumsal)/3 from scott.test);

结果:

--分析函数

select ename,sal,sum(sal) over(partition by deptno order by sal desc) from scott.emp ;

--rows n preceding(窗口子句一)

select deptno,sal,sum(sal) over(order by sal rows 5 preceding) from scott.emp ;

结果:

--rum(..) over(..)..

select sal,sum(1) over(order by sal) aa from scott.emp  ;

select deptno,ename,sal,sum(sal) over(order by ename) 连续求和,sum(sal) over() 总和,100*round(sal/sum(sal) over(),4) as 份额 from scott.emp;

结果:

select deptno,ename,sal,sum(sal) over(partition by deptno order by ename) 部门连续求和,sum(sal) over(partition by deptno) 部门总和,100*round(sal/sum(sal) over(),4) as 总份额 from scott.emp;

结果:

select deptno,sal,rank() over (partition by deptno order by sal),dense_rank() over(partition by deptno order by sal) from scott.emp order by deptno ;

结果;

select * from (select rank() over(partition by 课程 order by 分数 desc) rk,分析函数_rank.* from 分析函数_rank) where rk=3 ;

--dense_rank():有重复的数字不跳着排列

--row_number()

select deptno,sal,row_number() over(partition by deptno order by sal) rm from scott.emp ;

结果:

--lag()和lead()

select deptno,sal,lag(sal) over(partition by deptno order by sal) 上一个,lead(sal) over(partition by deptno order by sal) from scott.emp ;

结果:

--max(),min(),avg()

select deptno,sal,max(sal) over(partition by deptno order by sal)最大,min(sal) over(partition by deptno order by sal)最小,avg(sal) over(partition by deptno order by sal)平均 from scott.emp ;

结果:

--first_value(),last_value()

select deptno,sal,first_value(sal) over(partition by deptno)最前,last_value(sal) over(partition by deptno )最后 from scott.emp ;

结果:

--分组补充 group by grouping sets

select deptno ,sal,sum(sal) from scott.emp group by grouping sets(deptno,sal);

select null,sal,sum(sal) from scott.emp group by sal union all select deptno,null,sum(sal) from scott.emp group by deptno ;

结果:

--rollup

select deptno,job,avg(sal) from scott.emp group by rollup(deptno,job) ;

--理解rollup等价于

select deptno,job,avg(sal) from scott.emp group by deptno,job union select deptno ,null,avg(sal) from scott.emp group by deptno union select null,null,avg(sal) from scott.emp ;

结果:

select deptno,job,avg(sal) a from scott.emp group by cube(deptno,job) ;

--理解CUBE

select deptno,job,avg(sal) from scott.emp group by cube(deptno,job) ;

--等价于

select deptno,job,avg(sal) from scott.emp group by grouping sets((deptno,job),(deptno),(job),());

结果:

--查询工资不在1500至2850之间的所有雇员名及工资

select ename,sal from scott.emp where sal not in(select sal from scott.emp where sal between 1500 and 2850 );

--部门10和30中的工资超过1500的雇员名及工资

select deptno,ename,sal from scott.emp a where a.deptno in(10,30) and a.sal>1500 order by sal desc ;

结果:

--在1981年2月1日至1981年5月1日之间雇佣的雇员名,岗位及雇佣日期,并以雇佣日期先后顺序排序

select ename as 姓名,job as 岗位,hiredate as 雇佣日期 from scott.emp a where a.hiredate between to_date('1981-02-01','yyyy-mm-dd') and to_date('1981-05-01','yyyy-mm-dd') order by a.hiredate asc ;

结果:

select * from scott.emp where hiredate >to_date('1981-02-01','yyyy-MM-dd');

--查询获得补助的所有雇佣名,工资及补助额,并以工资和补助的降序排序

select ename,sal,comm from scott.emp a where a.comm > all(0) order by comm desc;

--工资低于1500的员工增加10%的工资,工资在1500及以上的增加5%的工资并按工资高低排序(降序)

select ename as 员工姓名,sal as 补发前的工资,case when sal1500 then (sal+sal*0.1) else (sal+sal*0.05) end 补助后的工资 from scott.emp order by sal desc ;

结果:

--查询公司每天,每月,每季度,每年的资金支出数额

select sum(sal/30) as 每天发的工资,sum(sal) as 每月发的工资,sum(sal)*3 as 每季度发的工资,sum(sal)*12 as 每年发的工资 from scott.emp;

结果:

--查询所有员工的平均工资,总计工资,最高工资和最低工资

select avg(sal) as 平均工资,sum(sal) as 总计工资,max(sal) as 最高工资,min(sal) as 最低工资 from scott.emp;

结果:

--每种岗位的雇员总数和平均工资

select job as 岗位,count(job) as 岗位雇员总数,avg(sal) as 平均工资 from scott.emp group by job order by 平均工资 desc;

结果:

--雇员总数以及获得补助的雇员数

select count(*) as 公司雇员总数,count(comm) as 获得补助的雇员人数 from scott.emp ;

--管理者的总人数

--雇员工资的最大差额

select max(sal),min(sal),(max(sal) - min(sal)) as 员工工资最大差额 from scott.emp ;

--每个部门的平均工资

select deptno,avg(sal) from scott.emp a group by a.deptno;

结果:

--查询每个岗位人数超过2人的所有职员信息

select * from scott.emp a,(select c.job,count(c.job) as sl from scott.emp c group by c.job ) b where b.sl>2 and a.job=b.job;

结果:

select * from scott.emp a where a.empno in(select mgr from scott.emp ) and (select count(mgr) from scott.emp)>2 ;

结果:

--处理重复行数据信息(删除,查找,修改)

select * from a1 a where not exists (select b.rd from (select rowid rd,row_number() over(partition by LOAN, BRANCH order by BEGIN_DATE desc) rn from a1) b where b.rn = 1 and a.rowid = b.rd);

--查询emp表数据信息重复问题

select * from scott.emp a where exists(select b.rd from(select rowid rd,row_number() over(partition by ename,job,mgr,hiredate,sal,comm,deptno order by empno asc) rn from scott.emp) b where b.rn=1 and a.rowid=b.rd);

--initcap:返回字符串,字符串第一个字母大写

select initcap(ename) Upp from scott.emp ;

结果:

--ascii:返回与指定的字符对应的十进制数

select ascii(a.empno) as 编号,ascii(a.ename) as 姓名,ascii(a.job) as 岗位 from scott.emp a ;

结果:

--chr:给出整数,返回对应的字符

select chr(ascii(ename)) as 姓名 from scott.emp ;

结果:

--concat:连接字符串

select concat(a.ename,a.job)|| a.empno as 字符连接 from scott.emp a;

结果:

--instr:在一个字符串中搜索指定的字符,返回发现指定的字符的位置

select instr(a.empno,a.mgr,1,1) from scott.emp a ;

--length:返回字符串的长度

select ename,length(a.ename) as 长度,a.job,length(a.job) as 长度 from scott.emp a ;

--lower:返回字符串,并将所返回的字符小写

select a.ename as 大写,lower(a.ename) as 小写 from scott.emp a ;

结果:

--upper:返回字符串,并将返回字符串都大写

select lower(a.ename) as 小写名字,upper(a.ename) as 大写名字 from scott.emp a ;

结果:

--rpad:在列的右边粘贴字符,lpad: 在列的左边粘贴字符(不够字符则用*来填满)

select lpad(rpad(a.ename,10,'*'),16,'*') as 粘贴 from scott.emp a ;

结果:

--like不同角度的使用

select * from scott.emp where ename like '%XXR%';

select * from scott.emp where ename like '%S';

select * from scott.emp where ename like 'J%';

select * from scott.emp where ename like 'S';

select * from scott.emp where ename like '%S_';

--每个部门的工资总和

select a.ename,sum(sal) from scott.emp a group by ename;

--每个部门的平均工资

select a.deptno,avg(sal) from scott.emp a group by deptno ;

--每个部门的最大工资

select a.deptno,max(sal) from scott.emp a group by deptno ;

--每个部门的最小工资

select a.deptno,min(sal) from scott.emp a group by deptno ;

--查询原工资占部门工资的比率

select deptno ,sal,ratio_to_report(sal) over(partition by deptno) sal_ratio from scott.emp ;

--查询成绩不及格的所有学生信息(提示:没有对应的表,只是意思意思。不及格人数大于等于三才能查)

select * from scott.emp where empno in(select distinct empno from scott.emp where 3(select count(sal) from scott.emp where sal3000) and empno in(select empno from scott.emp where sal3000));

结果:

--查询每个部门的平均工资

select distinct deptno,avg(sal) from scott.emp group by deptno  order by deptno desc;

--union组合查出的结果,但要求查出来的数据类型必须相同

select sal from scott.emp where sal >=all(select sal from scott.emp ) union select sal from scott.emp ;

select * from scott.emp a where a.empno between 7227 and 7369 ;--只能从小到大

---------创建表空间  要用拥有create tablespace权限的用户,比如sys

create tablespace tbs_dat datafile 'c:\oradata\tbs_dat.dbf' size 2000M;

---------添加数据文件

alter tablespace tbs_dat add datafile 'c:\oradata\tbs_dat2.dbf' size 100M;

---------改变数据文件大小

alter database datafile 'c:\oradata\tbs_dat.dbf' resize 250M;

---------数据文件自动扩展大小

alter database datafile 'c:\oradata\tbs_dat.dbf' autoextend on next 1m maxsize 20m;

---------修改表空间名称

alter tablespace tbs_dat rename to tbs_dat1;

---------删除表空间  and datafiles 表示同时删除物理文件

drop tablespace tbs_dat including contents and datafiles;

--substr(s1,s2,s3):截取s1字符串,从s2开始,结束s3

select substr(job,3,length(job)) from scott.emp ;

--replace:替换字符串

select replace(ename,'LL','aa') from scott.emp;

select * from scott.test;

insert into scott.test(ename,job) values('weather','好');

insert into scott.test(ename,job) values('wether','差');

--soundex:返回一个与给定的字符串读音相同的字符串

select ename from scott.test where soundex(ename)=soundex('wether');

--floor:取整数

select sal,floor(sal) as 整数 from scott.emp ;

--log(n,s):返回一个以n为低,s的对数

select empno,log(empno,2) as 对数 from scott.emp ;

--mod(n1,n2):返回一个n1除以n2的余数

select empno,mod(empno,2) as 余数 from scott.emp ;

结果:

--power(n1,n2):返回n1的n2次方根

select empno,power(empno,2) as 方根 from scott.emp ;

--round和trunc:按照指定的精度进行舍入

select round(41.5),round(-41.8),trunc(41.6),trunc(-41.9) from scott.emp ;

--sign:取数字n的符号,大于0返回1,小于0返回-1,等于0返回0

select sign(45),sign(-21),sign(0) from scott.emp ;

结果:

select * from scott.emp;

oracle相关的数据库SQL查询语句:

1.  在职员表中查询出基本工资比平均基本工资高的职工编号。

2.  查询一个或者多个部门的所有员工信息,该部门的所有员工工资都高于公司的平均工资。

3.  现有张三的出生日期:1985-01-15 01:27:36,请各自新建表,将此日期时间插入表中,并计算出张三的年龄,显示张三的生日。

4.  生日的输出格式要求为MM-DD(未满两位的用0不全),张三的生日为01-15。

5.  算年龄要求用三个方式实现。

6.  生日要求用两个方式实现。

7.  在数据库表中有以下字符数据,如:

13-1,14-2,13-15,13-2,13-108,13-3,13-10,13-200,13-18,100-11,14-1

现在希望通过一条SQL语句进行排序,并且首先要按照前半部分的数字进行排序,然后再按照后半部分的数字进行排序,输出要拍成如下所示:
13-1,13-2,13-3,13-10,13-15,13-18,13-108,13-200,14-1,14-2,100-11

数据库表名:SellRecord;字段ListNumber;

8.  显示所有雇员的姓名以及满10年服务年限后的日期。

9.  显示雇员姓名,根据其服务年限,将最老的雇员排在最前面。

10显示所有雇员的姓名和加入公司的年份和月份,按雇员受雇日期所在月排序,将最早年份的职员排在最前面。

10.             显示假设一个月为30天的情况下所有雇员的日薪金。

11.             找出在(任何年份的)2月受聘的所有雇员(用两种方式实现)。

12.             对于每个雇员,显示其加入公司的天数。

13.             以年,月和日的方式显示所有雇员的服务年限(入职多少年/入职了多少月/入职了多少天)。

14.             找出各月最后一天受雇的所有雇员。

15.             找出早于25年之前受雇的雇员(用两种方式实现)。

16.             工资最低1500的职员增加10%,1500以上的增加5%的工资,用一条update语句实现(用两种方式实现)。

17.             按照部门统计每种岗位的平均工资,要求输出的格式如下图所示:

18.

19.

20.

21,。

22.

本人声明:以上内容出现任何错误与不足,皆与本人无关。

您可能感兴趣的文章:
  • Oracle 11GR2的递归WITH子查询方法
  • Oracle基础学习之子查询
  • Oracle数据库中基本的查询优化与子查询优化讲解
  • Oracle通过递归查询父子兄弟节点方法示例
  • 一个oracle+PHP的查询的例子
  • oracle基本查询用法入门示例
  • oracle 查询表名以及表的列名
  • oracle查询语句大全(oracle 基本命令大全一)
  • oracle数据库常用的99条查询语句
  • ORACLE查询删除重复记录三种方法
  • oracle基本查询操作子查询用法实例分析

标签:湖南 兰州 黄山 崇左 湘潭 衡水 铜川 仙桃

巨人网络通讯声明:本文标题《oracle常用sql查询语句部分集合(图文)》,本文关键词  ;如发现本文内容存在版权问题,烦请提供相关信息告之我们,我们将及时沟通与处理。本站内容系统采集于网络,涉及言论、版权与本站无关。
  • 相关文章
  • 收缩
    • 微信客服
    • 微信二维码
    • 电话咨询

    • 400-1100-266