1:请从表EMP中查找工资低于2000的雇员的姓名、工作、工资,并按工资降序排列。
SQL> select ename,job,sal from emp where sal<2000 order by sal desc;
2:查询所有雇员的姓名、SAL与之和(显示字段为SC)
SQL> select ename,sal+nvl(comm,0) as sc from emp;
3:查询各部门中81年2月1日以后入职的员工数量
SQL> SELECT deptno, COUNT(*) AS cnt
2 FROM emp
3 WHERE hiredate > to_date('19810201', 'yyyymmdd')
4 GROUP BY deptno
5 ORDER BY 1 ASC;
4:查询列出各部门的部门名和部门经理的名字?
scott用户下:
SQL> select d.ename,t.dname from emp d,dept t
2 where d.job=upper('MANAGER') and d.deptno = t.deptno;
列出所有人的部门经理的名称及部门名字
select d.root,d.ename,t.dname from
(select connect_by_root(ename) as root,
deptno,
ename
from emp
start with job=upper('manager')
connect by prior empno = mgr) d,dept t where d.deptno=t.deptno;
5:统计emp中每个部门的工资总和及每个部门的人数
SQL> select deptno,sum(sal),count(*) from emp group by deptno order by 1;
6:查询部门平均工资最高的部门名称和最低的部门名称
select t.dname,avg(d.sal) avg_sal from emp d,dept t
一:
SQL> with q1 as
2 (select e.deptno, d.dname, avg(sal) avg
3 from emp e, dept d
4 where e.deptno = d.deptno
5 group by e.deptno, d.dname)
6 select dname, avg
7 from q1
8 where avg = (select max(avg) from q1)
9 union all
10 select dname, avg from q1 where avg = (select min(avg) from q1);
DNAME AVG
-------------- ----------
ACCOUNTING 2916.66667
SALES 1566.66667
二:
with t1 as
(select dname deptno, avg(sal) as avgsal
from emp d, dept t
where d.deptno = t.deptno
group by dname),
t2 as
(select max(avgsal) as sal1, min(avgsal) sal2 from t1)
select d1.deptno, d2.sal1, d3.deptno, d4.sal2
from t1 d1, t2 d2, t1 d3, t2 d4
where d1.avgsal = d2.sal1
and d3.avgsal = d4.sal2;
7:统计2008年8月8日至今总共有多少天?
SQL> select sysdate-to_date('20080808','yyyymmdd') from dual;
SYSDATE-TO_DATE('20080808','YYYYMMDD')
--------------------------------------
2957.4211
8:在HR用户利用CTAS方式创建表EMP,仅创建表结构。
SQL> create table emp as select * from employees where 1=2;
Table created.
SQL> select * from emp;
no rows selected
9:在HR用户,将EMPLOYEES表中,除SALARY字段外,其他字段的值都插入到EMP表中。
sql> insert into emp(employee_id,first_name,last_name,email,phone_number,hire_date,job_id,commission_pct,manager_id,department_id)
2 select employee_id,first_name,last_name,email,phone_number,hire_date,job_id,commission_pct,manager_id,department_id
3 from employees;
107 rows created.
10:在HR用户,通过查询EMPLOYEES表的SALARY字段,更新至EMP表的SALARY字段。
SQL> update emp d set salary=(select salary from employees t where d.employee_id = t.employee_id);
107 rows updated.
11:在HR用户,EMP表中,所有工资小于10000的,增加10%工资。
SQL> update emp set salary=salary*1.1 where salary <10000;
64 rows updated.
12:在HR用户,查询EMPLOYEES表,显示人员ID、姓名、电话号码,要求:电话号码不满18位的,电话号码前面用‘0’补齐,人员ID升序排列。
select employee_id,first_name,lpad(phone_number,18,'0') ph from employees order by 1;
13:在SCOTT用户,更新COMM字段,只更新COMM字段为空的记录,更新的值为薪金的20%。
update emp set comm=SAL*0.2 where comm is null;
14:在SCOTT用户,查询EMP,SALGRADE这两张表,显示每个人的级别。显示字段:人员id,姓名,级别。
SQL> select e.empno,e.ename,s.grade from emp e,salgrade s where e.sal in(s.hisal,s.losal);
EMPNO ENAME GRADE
---------- ---------- ----------
7788 SCOTT 4
7902 FORD 4
select d.empno,d.ename,t.grade from emp d,salgrade t where sal between losal and hisal
15:在HR用户,查询没有任何职员的部门
select d.department_name,d.department_id from departments d
where d.department_id not in (select e.department_id from employees e
where e.department_id is not null);
select department_id, department_name
from departments
where department_id not in
(select distinct (department_id)
from employees
where department_id is not null);
select department_id, department_name
from departments d
where not exists (select null
from employees
where department_id = d.department_id);
16:打印2016-06-15~2016-07-14之间的连续日期
select to_date('2016-06-15','yyyy-mm-dd')+rownum-1 from dual connect by rownum<=30;
17:在SCOTT用户,利用下面语句建表:
create table emp_enmo as select * from emp where 1=2;
insert into emp_enmo select empno,ename,null as job,mgr,hiredate,sal,comm,deptno from emp where sal>=3000;
要求:使用merge完成对表emp_enmo数据的补全。
语法:
MERGE INTO [schema .] table [t_alias]
USING [schema .] { table | view | subquery } [t_alias]
ON ( condition )--------------------匹配必须为主键或唯一键
WHEN MATCHED THEN merge_update_clause
WHEN NOT MATCHED THEN merge_insert_clause;
SQL> merge into emp_enmo d
2 using emp t
3 on (d.empno = t.empno)
4 when matched then
5 update
6 set d.ename = t.ename,
7 d.job = t.job,
8 d.mgr = t.mgr,
9 d.hiredate = t.hiredate,
10 d.sal = t.sal,
11 d.comm = t.comm,
12 d.deptno = t.deptno
13 when not matched then
14 insert
15 values
16 (t.empno, t.ename, t.job, t.mgr, t.hiredate, t.sal, t.comm, t.deptno);
14 rows merged.
18:在SCOTT用户,只查EMP表,显示:人员id,姓名,及部门名称,要求:部门名称利用decode函数来完成显示部门名称,按照人员id升序排列。
SQL> select empno,ename,decode(deptno,10,'ACCOUNTING',20,
2 'RESEARCH',30,'SALES',40,'OPERATIONS',deptno) from emp
3 order by 1;
字段,10(如果),,'ACCOUNTING'(那么),....deptno(否则)。
19:在HR用户,查询employees表,要求:大于5人的job_id的最大薪金,最小薪金,人数。
SQL> select job_id,max(salary),min(salary),count(*) from employees
2 group by job_id having count(*) > 5 ;
20:TOP N
1)获取所有成绩的top-3
2)每门课程成绩的前三名;(注意:如果遇到了相同成绩的情况,改如何考虑?)
测试表如下:
--创建表
create TABLE sc
(
sno NUMBER(5),
cno NUMBER(5),
grade NUMBER(4,1)
);
SQL> create table sc(sno number(5),cno number(5),grade number(4,1));
Table created.
--表中插入测试数据
insert into sc values(1,1,91);
insert into sc values(1,2,92);
insert into sc values(1,3,93);
insert into sc values(2,1,88);
insert into sc values(2,2,92);
insert into sc values(2,3,99);
insert into sc values(3,1,65);
insert into sc values(3,2,75);
insert into sc values(3,3,85);
insert into sc values(4,1,80);
insert into sc values(4,2,88);
insert into sc values(4,3,93);
commit;
SQL> select * from sc;
答:
1):
select sno,cno,grade from (select sno,cno,grade from sc order by grade desc) where rownum<=3;
2):
HR@ORA11GR2>
要求2:遇到相同成绩,则并列,并列后,占用排名名次
HR@ORA11GR2>select * from
(SELECT sno,cno,grade,rank() over(partition by cno
order by grade desc) as rn FROM sc)
where rn<=3;
SNO CNO GRADE RN
---- -------- --------- --------
1 1 91 1
2 1 88 2
4 1 80 3
1 2 92 1
2 2 92 1
4 2 88 3
2 3 99 1
1 3 93 2
4 3 93 2
9 rows selected.
HR@ORA11GR2>
要求3:遇到相同成绩,则并列,并列后,并列不多占用名次
HR@ORA11GR2>select * from
(SELECT sno,cno,grade,dense_rank() over(partition by
cno order by grade desc) as rn FROM sc)
where rn<=3;
SNO CNO GRADE RN
---- -------- --------- --------
1 1 91 1
2 1 88 2
4 1 80 3
1 2 92 1
2 2 92 1
4 2 88 2
3 2 75 3
2 3 99 1
1 3 93 2
4 3 93 2
3 3 85 3
11 rows selected.
HR@ORA11GR2>
附加题:请查询出scott.emp表中根据工资(SAL字段)从多到少排名,工资位于第5-第10的记录。
select * from(
select ename,sal,rownum as rn from(select ename,sal from emp order by sal desc ) t)
where rn >=5 and rn<=10;
分析函数:
select d.*,row_number() over(order by sal) from emp d
select d.*,dense_rank() over(order by sal) from emp d
select d.*,rank() over(order by sal) from emp d
SQL> select ename,job,sal from emp where sal<2000 order by sal desc;
2:查询所有雇员的姓名、SAL与之和(显示字段为SC)
SQL> select ename,sal+nvl(comm,0) as sc from emp;
3:查询各部门中81年2月1日以后入职的员工数量
SQL> SELECT deptno, COUNT(*) AS cnt
2 FROM emp
3 WHERE hiredate > to_date('19810201', 'yyyymmdd')
4 GROUP BY deptno
5 ORDER BY 1 ASC;
4:查询列出各部门的部门名和部门经理的名字?
scott用户下:
SQL> select d.ename,t.dname from emp d,dept t
2 where d.job=upper('MANAGER') and d.deptno = t.deptno;
列出所有人的部门经理的名称及部门名字
select d.root,d.ename,t.dname from
(select connect_by_root(ename) as root,
deptno,
ename
from emp
start with job=upper('manager')
connect by prior empno = mgr) d,dept t where d.deptno=t.deptno;
5:统计emp中每个部门的工资总和及每个部门的人数
SQL> select deptno,sum(sal),count(*) from emp group by deptno order by 1;
6:查询部门平均工资最高的部门名称和最低的部门名称
select t.dname,avg(d.sal) avg_sal from emp d,dept t
一:
SQL> with q1 as
2 (select e.deptno, d.dname, avg(sal) avg
3 from emp e, dept d
4 where e.deptno = d.deptno
5 group by e.deptno, d.dname)
6 select dname, avg
7 from q1
8 where avg = (select max(avg) from q1)
9 union all
10 select dname, avg from q1 where avg = (select min(avg) from q1);
DNAME AVG
-------------- ----------
ACCOUNTING 2916.66667
SALES 1566.66667
二:
with t1 as
(select dname deptno, avg(sal) as avgsal
from emp d, dept t
where d.deptno = t.deptno
group by dname),
t2 as
(select max(avgsal) as sal1, min(avgsal) sal2 from t1)
select d1.deptno, d2.sal1, d3.deptno, d4.sal2
from t1 d1, t2 d2, t1 d3, t2 d4
where d1.avgsal = d2.sal1
and d3.avgsal = d4.sal2;
7:统计2008年8月8日至今总共有多少天?
SQL> select sysdate-to_date('20080808','yyyymmdd') from dual;
SYSDATE-TO_DATE('20080808','YYYYMMDD')
--------------------------------------
2957.4211
8:在HR用户利用CTAS方式创建表EMP,仅创建表结构。
SQL> create table emp as select * from employees where 1=2;
Table created.
SQL> select * from emp;
no rows selected
9:在HR用户,将EMPLOYEES表中,除SALARY字段外,其他字段的值都插入到EMP表中。
sql> insert into emp(employee_id,first_name,last_name,email,phone_number,hire_date,job_id,commission_pct,manager_id,department_id)
2 select employee_id,first_name,last_name,email,phone_number,hire_date,job_id,commission_pct,manager_id,department_id
3 from employees;
107 rows created.
10:在HR用户,通过查询EMPLOYEES表的SALARY字段,更新至EMP表的SALARY字段。
SQL> update emp d set salary=(select salary from employees t where d.employee_id = t.employee_id);
107 rows updated.
11:在HR用户,EMP表中,所有工资小于10000的,增加10%工资。
SQL> update emp set salary=salary*1.1 where salary <10000;
64 rows updated.
12:在HR用户,查询EMPLOYEES表,显示人员ID、姓名、电话号码,要求:电话号码不满18位的,电话号码前面用‘0’补齐,人员ID升序排列。
select employee_id,first_name,lpad(phone_number,18,'0') ph from employees order by 1;
13:在SCOTT用户,更新COMM字段,只更新COMM字段为空的记录,更新的值为薪金的20%。
update emp set comm=SAL*0.2 where comm is null;
14:在SCOTT用户,查询EMP,SALGRADE这两张表,显示每个人的级别。显示字段:人员id,姓名,级别。
SQL> select e.empno,e.ename,s.grade from emp e,salgrade s where e.sal in(s.hisal,s.losal);
EMPNO ENAME GRADE
---------- ---------- ----------
7788 SCOTT 4
7902 FORD 4
select d.empno,d.ename,t.grade from emp d,salgrade t where sal between losal and hisal
15:在HR用户,查询没有任何职员的部门
select d.department_name,d.department_id from departments d
where d.department_id not in (select e.department_id from employees e
where e.department_id is not null);
select department_id, department_name
from departments
where department_id not in
(select distinct (department_id)
from employees
where department_id is not null);
select department_id, department_name
from departments d
where not exists (select null
from employees
where department_id = d.department_id);
16:打印2016-06-15~2016-07-14之间的连续日期
select to_date('2016-06-15','yyyy-mm-dd')+rownum-1 from dual connect by rownum<=30;
17:在SCOTT用户,利用下面语句建表:
create table emp_enmo as select * from emp where 1=2;
insert into emp_enmo select empno,ename,null as job,mgr,hiredate,sal,comm,deptno from emp where sal>=3000;
要求:使用merge完成对表emp_enmo数据的补全。
语法:
MERGE INTO [schema .] table [t_alias]
USING [schema .] { table | view | subquery } [t_alias]
ON ( condition )--------------------匹配必须为主键或唯一键
WHEN MATCHED THEN merge_update_clause
WHEN NOT MATCHED THEN merge_insert_clause;
SQL> merge into emp_enmo d
2 using emp t
3 on (d.empno = t.empno)
4 when matched then
5 update
6 set d.ename = t.ename,
7 d.job = t.job,
8 d.mgr = t.mgr,
9 d.hiredate = t.hiredate,
10 d.sal = t.sal,
11 d.comm = t.comm,
12 d.deptno = t.deptno
13 when not matched then
14 insert
15 values
16 (t.empno, t.ename, t.job, t.mgr, t.hiredate, t.sal, t.comm, t.deptno);
14 rows merged.
18:在SCOTT用户,只查EMP表,显示:人员id,姓名,及部门名称,要求:部门名称利用decode函数来完成显示部门名称,按照人员id升序排列。
SQL> select empno,ename,decode(deptno,10,'ACCOUNTING',20,
2 'RESEARCH',30,'SALES',40,'OPERATIONS',deptno) from emp
3 order by 1;
字段,10(如果),,'ACCOUNTING'(那么),....deptno(否则)。
19:在HR用户,查询employees表,要求:大于5人的job_id的最大薪金,最小薪金,人数。
SQL> select job_id,max(salary),min(salary),count(*) from employees
2 group by job_id having count(*) > 5 ;
20:TOP N
1)获取所有成绩的top-3
2)每门课程成绩的前三名;(注意:如果遇到了相同成绩的情况,改如何考虑?)
测试表如下:
--创建表
create TABLE sc
(
sno NUMBER(5),
cno NUMBER(5),
grade NUMBER(4,1)
);
SQL> create table sc(sno number(5),cno number(5),grade number(4,1));
Table created.
--表中插入测试数据
insert into sc values(1,1,91);
insert into sc values(1,2,92);
insert into sc values(1,3,93);
insert into sc values(2,1,88);
insert into sc values(2,2,92);
insert into sc values(2,3,99);
insert into sc values(3,1,65);
insert into sc values(3,2,75);
insert into sc values(3,3,85);
insert into sc values(4,1,80);
insert into sc values(4,2,88);
insert into sc values(4,3,93);
commit;
SQL> select * from sc;
答:
1):
select sno,cno,grade from (select sno,cno,grade from sc order by grade desc) where rownum<=3;
2):
HR@ORA11GR2>
要求2:遇到相同成绩,则并列,并列后,占用排名名次
HR@ORA11GR2>select * from
(SELECT sno,cno,grade,rank() over(partition by cno
order by grade desc) as rn FROM sc)
where rn<=3;
SNO CNO GRADE RN
---- -------- --------- --------
1 1 91 1
2 1 88 2
4 1 80 3
1 2 92 1
2 2 92 1
4 2 88 3
2 3 99 1
1 3 93 2
4 3 93 2
9 rows selected.
HR@ORA11GR2>
要求3:遇到相同成绩,则并列,并列后,并列不多占用名次
HR@ORA11GR2>select * from
(SELECT sno,cno,grade,dense_rank() over(partition by
cno order by grade desc) as rn FROM sc)
where rn<=3;
SNO CNO GRADE RN
---- -------- --------- --------
1 1 91 1
2 1 88 2
4 1 80 3
1 2 92 1
2 2 92 1
4 2 88 2
3 2 75 3
2 3 99 1
1 3 93 2
4 3 93 2
3 3 85 3
11 rows selected.
HR@ORA11GR2>
附加题:请查询出scott.emp表中根据工资(SAL字段)从多到少排名,工资位于第5-第10的记录。
select * from(
select ename,sal,rownum as rn from(select ename,sal from emp order by sal desc ) t)
where rn >=5 and rn<=10;
分析函数:
select d.*,row_number() over(order by sal) from emp d
select d.*,dense_rank() over(order by sal) from emp d
select d.*,rank() over(order by sal) from emp d
来自 “ ITPUB博客 ” ,链接:http://blog.itpub.net/31397003/viewspace-2127037/,如需转载,请注明出处,否则将追究法律责任。
转载于:http://blog.itpub.net/31397003/viewspace-2127037/