oracle 常用command

Lunatic 整理

1. 删除表的注意事项
在删除一个表中的全部数据时，须使用TRUNCATE TABLE 表名;因为用DROP TABLE，DELETE * FROM 表名时，TABLESPACE表空间该表的占用空间并未释放，反复几次DROP，DELETE操作后，该TABLESPACE上百兆的空间就被耗光了。

2.having 子句的用法
　　having 子句对 group by 子句所确定的行组进行控制,having 子句条件中只允许涉及常量,聚组函数或group by 子句中的列.

3.外部联接"+"的用法
　　外部联接"+"按其在"="的左边或右边分左联接和右联接.若不带"+"运算符的表中的一个行不直接匹配于带"+"预算符的表中的任何行,则前者的行与后者中的一个空行相匹配并被返回.若二者均不带’+’,则二者中无法匹配的均被返回.利用外部联接"+",可以替代效率十分低下的 not in 运算,大大提高运行速度.例如,下面这条命令执行起来很慢

用外联接提高表连接的查询速度
在作表连接(常用于视图)时，常使用以下方法来查询数据:
SELECT PAY_NO, PROJECT_NAME
FROM A
WHERE A.PAY_NO NOT IN (SELECT PAY_
NO FROM B WHERE VALUE >=120000);
---- 但是若表A有10000条记录，表B有10000条记录，则要用掉30分钟才能查完，主要因为NOT IN要进行一条一条的比较，共需要10000*10000次比较后，才能得到结果。该用外联接后，可以缩短到1分左右的时间:
SELECT PAY_NO,PROJECT_NAME
FROM A,B
WHERE A.PAY_NO=B.PAY_NO(+)
AND B.PAY_NO IS NULL
AND B.VALUE >=12000;


4.set transaction 命令的用法
在执行大事务时,有时oracle会报出如下的错误:
ORA-01555:snapshot too old (rollback segment too small)
　　这说明oracle给此事务随机分配的回滚段太小了,这时可以为它指定一个足够大的回滚段,以确保这个事务的成功执行.例如

set transaction use rollback segment roll_abc;
delete from table_name where ...
commit;
　　回滚段roll_abc被指定给这个delete事务,commit命令则在事务结束之后取消了回滚段的指定.


5.数据库重建应注意的问题

　　在利用import进行数据库重建过程中,有些视图可能会带来问题,因为结构输入的顺序可能造成视图的输入先于它低层次表的输入,这样建立视图就会失败.要解决这一问题,可采取分两步走的方法:首先输入结构,然后输入数据.命令举例如下 (uesrname:jfcl,password:hfjf,host sting:ora1,数据文件:expdata.dmp):

imp jfcl/hfjf@ora1 file=empdata.dmp rows=N

imp jfcl/hfjf@ora1 file=empdata.dmp full=Y buffer=64000

commit=Y ignore=Y

　　第一条命令输入所有数据库结构,但无记录.第二次输入结构和数据,64000字节提交一次.ignore=Y选项保证第二次输入既使对象存在的情况下也能成功.

select a.empno from emp a where a.empno not in

(select empno from emp1 where job=’SALE’);

　　倘若利用外部联接,改写命令如下:

select a.empno from emp a ,emp1 b

where a.empno=b.empno(+)

and b.empno is null

and b.job=’SALE’;

　　可以发现,运行速度明显提高.


6.从已知表新建另一个表：
CREATE TABLE b
AS SELECT * (可以是表a中的几列)
FROM a
WHERE a.column = ...;


7.查找、删除重复记录:

法一: 用Group by语句 此查找很快的
select count(num), max(name) from student --查找表中num列重复的，列出重复的记录数，并列出他的name属性
group by num
having count(num) >1 --按num分组后找出表中num列重复，即出现次数大于一次

delete from student(上面Select的)
这样的话就把所有重复的都删除了。-----慎重

法二:当表比较大(例如10万条以上)时,这个方法的效率之差令人无法忍受,需要另想办法:
---- 执行下面SQL语句后就可以显示所有DRAWING和DSNO相同且重复的记录
SELECT * FROM EM5_PIPE_PREFAB
WHERE ROWID!=(SELECT MAX(ROWID) FROM EM5_PIPE_PREFAB D --D相当于First,Second
WHERE EM5_PIPE_PREFAB.DRAWING=D.DRAWING AND
EM5_PIPE_PREFAB.DSNO=D.DSNO);

---- 执行下面SQL语句后就可以刪除所有DRAWING和DSNO相同且重复的记录
DELETE FROM EM5_PIPE_PREFAB
WHERE ROWID!=(SELECT MAX(ROWID) FROM EM5_PIPE_PREFAB D
WHERE EM5_PIPE_PREFAB.DRAWING=D.DRAWING AND
EM5_PIPE_PREFAB.DSNO=D.DSNO);

8.返回表中[N，M]条记录：

取得某列中第N大的行
select column_name from
(select table_name.*,dense_rank() over (order by column desc) rank from table_name)
where rank = &N；

　假如要返回前5条记录：
　　select * from tablename where rownum<6;(或是rownum <= 5 或是rownum != 6)
假如要返回第5-9条记录：
select * from tablename
where …
and rownum<10
minus
select * from tablename
where …
and rownum<５
order by name
　选出结果后用name排序显示结果。(先选再排序)

注意：只能用以上符号(<、<=、!=)。
select * from tablename where rownum != 10;返回的是前９条记录。
不能用：>,>=,=,Between...and。由于rownum是一个总是从1开始的伪列，Oracle 认为这种条件 不成立，查不到记录.

另外，这个方法更快：
select * from (
select rownum r,a from yourtable
where rownum <= 20
order by name )
where r > 10
这样取出第11-20条记录!(先选再排序再选)
要先排序再选则须用select嵌套：内层排序外层选。

rownum是随着结果集生成的，一旦生成，就不会变化了；同时,生成的结果是依次递加的，没有1就永远不会有2!
rownum 是在 查询集合产生的过程中产生的伪列，并且如果where条件中存在 rownum 条件的话，则:
1： 假如 判定条件是常量，则：
只能 rownum = 1, <= 大于1 的自然数， = 大于1 的数是没有结果的， 大于一个数也是没有结果的
即 当出现一个 rownum 不满足条件的时候则 查询结束 　　this is stop key!
2: 当判定值不是常量的时候
若条件是 = var , 则只有当 var 为1 的时候才满足条件，这个时候不存在 stop key ,必须进行 full scan ,对每个满足其他where条件的数据进行判定
选出一行后才能去选rownum=2的行……


9.快速编译所有视图

---- 当在把数据库倒入到新的服务器上后(数据库重建)，需要将视图重新编译一遍，因为该表空间视图到其它表空间的表的连接会出现问题，可以利用PL/SQL的语言特性，快速编译。
SQL >SPOOL ON.SQL
SQL >SELECT ‘ALTER VIEW ‘||TNAME||’
COMPILE;’ FROM TAB;
SQL >SPOOL OFF
然后执行ON.SQL即可。
SQL >@ON.SQL
当然，授权和创建同义词也可以快速进行，如：
SQL >SELECT ‘GRANT SELECT ON ’
||TNAME||’ TO USERNAME;’ FROM TAB;
SQL >SELECT ‘CREATE SYNONYM
‘||TNAME||’ FOR USERNAME.’||TNAME||’;’ FROM TAB;

让你的天空只有甜和美

遗忘———该怎么流泪

10.读写文本型操作系统文件
---- 在PL/SQL 3.3以上的版本中，UTL_FILE包允许用户通过PL/SQL读写操作系统文件。如下：

DECALRE
FILE_HANDLE UTL_FILE.FILE_TYPE;
BEGIN
FILE_HANDLE:=UTL_FILE.FOPEN(
‘C:\’,’TEST.TXT’,’A’);
UTL_FILE.PUT_LINE(FILE_HANDLE,’
HELLO,IT’S A TEST TXT FILE’);
UTL_FILE.FCLOSE(FILE_HANDLE);
END;


11.在数据库触发器中使用列的新值与旧值
---- 在数据库触发器中几乎总是要使用触发器基表的列值，如果某条语句需要某列修改前的值，使用:OLD就可以了，使用某列修改后的新值，用:NEW就可以了。如:OLD.DEPT_NO,:NEW.DEPT_NO。


12.数据库文件的移动方法
当想将数据库文件移动到另外一个目录下时，可以用ALTER DATABASE命令来移动(比ALTER TABLESPACE适用性强)：
1. 使用SERVER MANAGER关闭实例.
SVRMGR > connect internal;
SVRMGR > shutdown;
SVRMGR >exit;
2. 使用操作系统命令来移动数据库文件位置(假设这里操作系统为SOLARIS 2.6). 在UNIX中用 mv命令可以把文件移动到新的位置，
#mv /ora13/orarun/document.dbf /ora12/orarun
3. 装载数据库并用alter database命令来改变数据库中的文件名.
SVRMGR > connect internal;
SVRMGR > startup mount RUN73;
SVRMGR > alter database rename file
> ‘/ ora13/orarun/document.dbf’
> ‘/ ora12/orarun/document.dbf’;
4. 启动实例.
SVRMGR > alter database open;


13.连接查询结果：
表a 列 a1 a2
记录 1 a
1 b
2 x
2 y
2 z
用select能选成以下结果:
1 ab
2 xyz

下面有两个例子：
1.使用pl/sql代码实现，但要求你组合后的长度不能超出oracle varchar2长度的限制
create or replace type strings_table is table of varchar2(20);
/
create or replace function merge (pv in strings_table) return varchar2
is
ls varchar2(4000);
begin
for i in 1..pv.count loop
ls := ls || pv(i);
end loop;
return ls;
end;
/
create table t (id number,name varchar2(10));
insert into t values(1,'Joan');
insert into t values(1,'Jack');
insert into t values(1,'Tom');
insert into t values(2,'Rose');
insert into t values(2,'Jenny');

column names format a80;
select t0.id,merge(cast(multiset(select name from t where t.id = t0.id) as strings_table)) names
from (select distinct id from t) t0;

drop type strings_table;
drop function merge;
drop table t;


2.纯粹用sql：
表dept, emp
要得到如下结果
deptno, dname, employees
---------------------------------
10, accounting, clark;king;miller
20, research, smith;adams;ford;scott;jones
30, sales, allen;blake;martin;james;turners
每个dept的employee串起来作为一条记录返回

This example uses a max of 6, and would need more cut n pasting to do more than that：

SQL> select deptno, dname, emps
2 from (
3 select d.deptno, d.dname, rtrim(e.ename ||', '||
4 lead(e.ename,1) over (partition by d.deptno
5 order by e.ename) ||', '||
6 lead(e.ename,2) over (partition by d.deptno
7 order by e.ename) ||', '||
8 lead(e.ename,3) over (partition by d.deptno
9 order by e.ename) ||', '||
10 lead(e.ename,4) over (partition by d.deptno
11 order by e.ename) ||', '||
12 lead(e.ename,5) over (partition by d.deptno
13 order by e.ename),', ') emps,
14 row_number () over (partition by d.deptno
15 order by e.ename) x
16 from emp e, dept d
17 where d.deptno = e.deptno
18 )
19 where x = 1
20 /

DEPTNO DNAME EMPS
------- ----------- ------------------------------------------
10 ACCOUNTING CLARK, KING, MILLER
20 RESEARCH ADAMS, FORD, JONES, ROONEY, SCOTT, SMITH
30 SALES ALLEN, BLAKE, JAMES, MARTIN, TURNER, WARD



14.在Oracle中建一个编号会自动增加的字段,以利于查询

1、建立序列：
CREATE SEQUENCE checkup_no_seq
NOCYCLE
MAXVALUE 9999999999
START WITH 2;

2、建立触发器：
CREATE OR REPLACE TRIGGER set_checkup_no
BEFORE INSERT ON checkup_history
FOR EACH ROW
DECLARE
next_checkup_no NUMBER;
BEGIN
--Get the next checkup number from the sequence
SELECT checkup_no_seq.NEXTVAL
INTO next_checkup_no
FROM dual;

--use the sequence number as the primary key
--for the record being inserted
:new.checkup_no := next_checkup_no;
END;


15.查看对象的依赖关系(比如视图与表的引用)

查看视图：dba_dependencies 记录了相关的依赖关系
查东西不知道要查看哪个视图时，可以在DBA_Objects里看，
select object_name from dba_objects where object_name like '%ROLE%'(假如查看ROLE相关)
然后DESC一下就大体上知道了。


16.要找到某月中所有周五的具体日期
select to_char(t.d,'YY-MM-DD') from (
select trunc(sysdate, 'MM')+rownum-1 as d
from dba_objects
where rownum < 32) t
where to_char(t.d, 'MM') = to_char(sysdate, 'MM') --找出当前月份的周五的日期
and trim(to_char(t.d, 'Day')) = '星期五'
--------
03-05-02
03-05-09
03-05-16
03-05-23
03-05-30

如果把where to_char(t.d, 'MM') = to_char(sysdate, 'MM')改成sysdate-90，即为查找当前
月份的前三个月中的每周五的日期。


让你的天空只有甜和美

遗忘———该怎么流泪

17 How to find the tablespace of a table?
SELECT tablespace_name
FROM all_tables
WHERE table_name = 'YOURTABLENAME';

18 How to remove duplicate rows from a table

If the unique/primary keys can be identified from the table, it is easier to remove the records from the table using the following query:
DELETE FROM tablename
WHERE rowid not in (SELECT MIN(rowid)
FROM tablename
GROUP BY column1, column2, column3...);
--Here column1, column2, column3 constitute the identifying key for each record.

If the keys cannot be identified for the table, you may create a temporary table using the query
CREATE TABLE temptablename
AS SELECT DISTINCT *
FROM tablename;
Then drop the original table and rename the temp table to original tablename.

19 How to identify and remove bad 'child' records to enable / create a foreign key ("Parent Keys Not Found" error when you try to enable/create relation from child table to parent table!)
The records can be identified and removed using the query

DELETE FROM childtablename ct
WHERE NOT EXISTS (SELECT 'x' FROM parenttablename pt
WHERE ct.keycolumn1 = pt.keycolumn1 AND ct.keycolumn2 = pt.keycolumn2...)
Or if you need to provide the user with bad records you may change the DELETE to SELECT with column list.

20 What is my current session id?

The username, program, machine, terminal, session id, serial # and more can be found from the v$session view. This view has a column audsid. When you join this coulum to your userenv('sessionid') value, you get the session information for your current session. The query could be

SELECT USERNAME, SID, SERIAL#, PROGRAM FROM V$SESSION
WHERE AUDSID = USERENV('SESSOINID');

21 How to terminate a session?
Using the above method you find the SID and SERIAL# for the session you wish to terminate. Then issue the command
ALTER SYSTEM KILL SESSION 'sid, serial#';

Please note that the sid and serial# should be in quotes separated by a comma.

22 Which database am I connected to? As which user?

The database name can be found out from different views. The view which everyone has access is GLOBAL_NAME. The query is
SELECT GLOBAL_NAME FROM GLOBAL_NAME;
To find the user, from sqlplus you can do "SHOW USER".

23 What is the SGA size?

There are two simple ways to find this. The first is to invoke server manager (svrmgrl) and connect ineternal (or as any user). Issue the command "SHOW SGA". Or you can run the query "SELECT * FROM V$SGA;" from svrmgrl or sqlplus.

24 Where are my alert log and dump files written to?

The alert file is written to your BACKGROUND_DUMP_DEST. This variable is set in the config.ora (init.ora) file. You can find the current values of the dump directories from the database. Invoke svrmgrl and connect. Issue command "SHOW PARAMETER DUMP". The SHOW PARAMETER command can be used to find the value of any database parameter value. For example if you want to find the block size and block buffers, issue command "SHOW PARAMETER BLOCK".

25 How to create structure (no data) of a table from another table?

If you need to duplicate a table you can do a "create table newtablename as select * from tablename;" This will create the new table will all the data. If you need to create only the structure, add where condition "1=2" but in the new table every relation is false.
CREATE TABLE NEWTABLENAME AS
SELECT * FROM OLDTABLENAME WHERE 1=2;

26 How to increase the size of a tablespace?

The size of the tablespace is increased by changing the size of the underlying physical files. You can either add more space to the existing file by
ALTER DATABASE DATAFILE 'filename' RESIZE nn M;

OR you can add more physical datafiles to the tablespace by
ALTER TABLESPACE tablespacename
ADD DATAFILE 'filename' SIZE nn M;
Make sure you specify the full path name for the filename. Use the script tsinfo.sql to find the size and related physical files of a tablespace.

27 Is my database running in Archivelog mode? Where are the archived files written?
This can be found by invoking server manager (svrmgrl) and issuing the command "ARCHIVE LOG LIST".



To be continued ......


让你的天空只有甜和美

遗忘———该怎么流泪