keep是Oracle下的另一个分析函数，他的用法不同于通过over关键字指定的分析函数。使用KEEP 时和DENSE_RANK FIRST /DENSE_RANK LAST一起使用，获取一组中排名第一或者排名最后的记录。必须有order by 子句用来排序。后面也可以接over()分析函数部分。
SELECT deptno, MIN(t.mgr) KEEP (DENSE_RANK FIRST ORDER BY t.sal) a
from emp t
group by deptno;
例如上面这条语句a字段的含义就是：
1、按deptno分组，
2、分组内按sal排序，
3、DENSE_RANK FIRST表示HOLD住sal排在前面的一组数据（当排在前面的sal有重复值时，多条被HOLD），
4、然后在这组记录中，执行前面的聚合函数，这里是min(t.mgr)。
例子：
oracle；
建表：
CREATE TABLE emp
(
emp_id NUMBER(6),
ename VARCHAR2(45),
dept_id NUMBER(4),
hire_date DATE,
sal NUMBER(8,2)
);
插入数据：
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (101, 'Tom', 20, TO_DATE('21-09-1989', 'DD-MM-YYYY'), 2000);
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (102, 'Mike', 20, TO_DATE('13-01-1993', 'DD-MM-YYYY'), 8000);
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (120, 'John', 50, TO_DATE('18-07-1996', 'DD-MM-YYYY'), 1000);
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (121, 'Joy', 50, TO_DATE('10-04-1997', 'DD-MM-YYYY'), 1000);
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (122, 'Rich', 50, TO_DATE('01-05-1995', 'DD-MM-YYYY'), 3000);
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (123, 'Kate', 50, TO_DATE('10-10-1997', 'DD-MM-YYYY'), 5000);
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (124, 'Jess', 50, TO_DATE('16-11-1999', 'DD-MM-YYYY'), 6000);
INSERT INTO emp (emp_id, ename, dept_id, hire_date, sal) VALUES (100, 'Stev', 10, TO_DATE('01-01-1990', 'DD-MM-YYYY'), 7000);
COMMIT;
oracle中keep查询：
SQL> SELECT emp_id,ename,dept_id,hire_date,sal,
2 DENSE_RANK() OVER(PARTITION BY dept_id ORDER BY sal) DENSE_RANK,
3 MIN(hire_date) KEEP (DENSE_RANK FIRST ORDER BY sal) OVER(PARTITION BY dept_id) min_first,
4 MIN(hire_date) KEEP (DENSE_RANK LAST ORDER BY sal) OVER(PARTITION BY dept_id) min_last,
5 MAX(hire_date) KEEP (DENSE_RANK FIRST ORDER BY sal) OVER(PARTITION BY dept_id) max_first,
MAX(hire_date) KEEP (DENSE_RANK LAST ORDER BY sal) OVER(PARTITION BY dept_id) max_last
6 FROM emp; 7
EMP_ID ENAME DEPT_ID HIRE_DATE SAL DENSE_RANK MIN_FIRST MIN_LAST MAX_FIRST MAX_LAST
---------- --------------------------------------------- ---------- --------- ---------- ---------- --------- --------- --------- ---------
100 Stev 10 01-JAN-90 7000 1 01-JAN-90 01-JAN-90 01-JAN-90 01-JAN-90
101 Tom 20 21-SEP-89 2000 1 21-SEP-89 13-JAN-93 21-SEP-89 13-JAN-93
102 Mike 20 13-JAN-93 8000 2 21-SEP-89 13-JAN-93 21-SEP-89 13-JAN-93
120 John 50 18-JUL-96 1000 1 18-JUL-96 16-NOV-99 10-APR-97 16-NOV-99
121 Joy 50 10-APR-97 1000 1 18-JUL-96 16-NOV-99 10-APR-97 16-NOV-99
122 Rich 50 01-MAY-95 3000 2 18-JUL-96 16-NOV-99 10-APR-97 16-NOV-99
123 Kate 50 10-OCT-97 5000 3 18-JUL-96 16-NOV-99 10-APR-97 16-NOV-99
124 Jess 50 16-NOV-99 6000 4 18-JUL-96 16-NOV-99 10-APR-97 16-NOV-99
8 rows selected.
PostgreSQL兼容写法：
建表：
bill=# create table emp (empno int, ename text, mgr int, sal int, deptno int);
CREATE TABLE
插入数据：
insert into emp values (7369, 'SMITH', 7902, 800, 20);
insert into emp values (7900, 'JAMES', 7698, 950, 30);
insert into emp values (7876, 'ADAMS', 7788 , 1100, 20);
insert into emp values (7521, 'WARD' , 7698 , 1250, 30);
insert into emp values (7654, 'MARTIN', 7698 , 1250, 30);
insert into emp values (7934, 'MILLER', 7782 , 1300, 10);
insert into emp values (7844, 'TURNER', 7698 , 1500, 30);
insert into emp values (7499, 'ALLEN', 7698, 1600, 30);
insert into emp values (7782, 'CLARK', 7839 , 2450, 10);
insert into emp values (7698, 'BLAKE', 7839 , 2850, 30);
insert into emp values (7566, 'JONES', 7839 , 2975, 20);
insert into emp values (7788, 'SCOTT', 7566 , 3000, 20);
insert into emp values (7902, 'FORD' , 7555 , 3000, 20);
insert into emp values (7839, 'KING' , 7567, 5000, 10);
分开查询：
bill=# select deptno,min(mgr),max(mgr) from (
bill(# select *, dense_rank() over w1 from emp window w1 as (partition by deptno order by sal) -- 得到dense_rank的值 , order by sal 对应 FIRST
bill(# ) t
bill-# where dense_rank=1
bill-# group by deptno;
deptno | min | max
--------+------+------
10 | 7782 | 7782
20 | 7902 | 7902
30 | 7698 | 7698
(3 rows)
bill=# select deptno,min(mgr),max(mgr) from (
bill(# select *, dense_rank() over w1 from emp window w1 as (partition by deptno order by sal desc) -- 得到dense_rank的值 , order by sal desc 对应 LAST
bill(# ) t
bill-# where dense_rank=1
bill-# group by deptno;
deptno | min | max
--------+------+------
10 | 7567 | 7567
20 | 7555 | 7566
30 | 7839 | 7839
(3 rows)
合并查询:
bill=# select t1.deptno, t1.min, t1.max, t2.min, t2.max from
bill-# (select deptno,min(mgr),max(mgr) from (select *, dense_rank() over w1 from emp window w1 as (partition by deptno order by sal)) t where dense_rank=1 group by deptno) t1
bill-# join
bill-# (select deptno,min(mgr),max(mgr) from (select *, dense_rank() over w1 from emp window w1 as (partition by deptno order by sal desc)) t where dense_rank=1 group by deptno) t2
bill-# using (deptno);
deptno | min | max | min | max
--------+------+------+------+------
10 | 7782 | 7782 | 7567 | 7567
20 | 7902 | 7902 | 7555 | 7566
30 | 7698 | 7698 | 7839 | 7839
(3 rows)
