使用外部表可以很容易的实现网站的访问日志分析。
虽然使用Awstats等工具也可以实现，可是使用Oracle来分析我们更应该得心应手。
而且这一切还是有那么一点点Cool的。

好了，闲言少叙，让我们来看一下我分析的过程。
首先创建路径指向日志存放目录:

[oracle@jumper elog]$ pwd
/opt/oracle/elog
[oracle@jumper elog]$ ls
eygle_access_log.20061016
[oracle@jumper elog]$ sqlplus "/ as sysdba"

SQL*Plus: Release 9.2.0.4.0 - Production on Wed Oct 18 08:59:35 2006

Copyright (c) 1982, 2002, Oracle Corporation. All rights reserved.

Connected to:
Oracle9i Enterprise Edition Release 9.2.0.4.0 - Production
With the Partitioning option
JServer Release 9.2.0.4.0 - Production

SQL> create or replace directory elog
2 as '/opt/oracle/elog';

Directory created.

然后我将这个路径的访问权限授予eygle用户来进行具体操作:

SQL> grant read,write on directory elog to eygle;

Grant succeeded.

选择合适的分隔符创建外部表:

SQL> connect eygle/eygle
Connected.

SQL> create table eygle_access_log_20061016
2 ( ip_address_date varchar2(100),
3 acc_file varchar2(400),
4 acc_cdsz varchar2(20),
5 acc_url varchar2(400),
6 left_blank varchar2(10),
7 acc_agent varchar2(400))
8 organization external (
9 type oracle_loader
10 default directory ELOG
11 access parameters (
12 records delimited by newline
13 nobadfile
14 nodiscardfile
15 nologfile
16 fields terminated by '"'
17 missing field values are null
18 )
19 location('eygle_access_log.20061016')
20 ) reject limit unlimited
21 /

Table created.

此时我们就可以对eygle.com的2006年10月16日的访问日志进行分析了。

我们可以先看一下各个字段的分界结果，示例如下:

SQL> select ip_address_date from eygle_access_log_20061016
2 where rownum <11;

IP_ADDRESS_DATE
-------------------------------------------------------------
38.102.128.140 - - [16/Oct/2006:00:00:17 +0800]
66.249.65.113 - - [16/Oct/2006:00:00:19 +0800]
202.160.178.221 - - [16/Oct/2006:00:00:35 +0800]
59.36.78.100 - - [16/Oct/2006:00:00:37 +0800]
59.36.78.100 - - [16/Oct/2006:00:00:38 +0800]
72.30.61.8 - - [16/Oct/2006:00:00:38 +0800]
221.217.84.230 - - [16/Oct/2006:00:00:42 +0800]
221.217.84.230 - - [16/Oct/2006:00:00:42 +0800]
74.6.65.236 - - [16/Oct/2006:00:01:07 +0800]
74.6.73.36 - - [16/Oct/2006:00:01:09 +0800]

10 rows selected.

通过SQL析取出访问的ip地址:

SQL> select substr(ip_address_date,1,instr(ip_address_date,' ')) ip_address
2 from eygle_access_log_20061016 where rownum <11;

IP_ADDRESS
---------------------------------------------------------------------------
38.102.128.140
66.249.65.113
202.160.178.221
59.36.78.100
59.36.78.100
72.30.61.8
221.217.84.230
221.217.84.230
74.6.65.236
74.6.73.36

10 rows selected.

接下来我们就可以很容易的获得当日访问我站点的独立IP数量了:

SQL> set timing on
SQL> select count(distinct(substr(ip_address_date,1,instr(ip_address_date,' ')))) uip
2 from eygle_access_log_20061016;

UIP
----------
7534

Elapsed: 00:00:06.86

我们可以对比一下数据库表的性能。
首先将日志加载到数据库表中:

SQL> create table ealog as
2 select * from eygle_access_log_20061016;

Table created.

SQL> desc ealog;
Name Null? Type
----------------------------------------------------------------- -------- --------------------------------------------
IP_ADDRESS_DATE VARCHAR2(100)
ACC_FILE VARCHAR2(400)
ACC_CDSZ VARCHAR2(20)
ACC_URL VARCHAR2(400)
LEFT_BLANK VARCHAR2(10)
ACC_AGENT VARCHAR2(400)

SQL> select count(*) from ealog;

COUNT(*)
----------
165443

然后我们强制刷新Buffer Cache，消除Cache的影响，再次执行查询：

SQL> alter session set events = 'immediate trace name flush_cache';

Session altered.

Elapsed: 00:00:00.03
SQL> select count(distinct(substr(ip_address_date,1,instr(ip_address_date,' ')))) uip
2 from ealog;

UIP
----------
7528

Elapsed: 00:00:02.15

此时用了大约2秒的时间，也就是说，外部表的性能较数据库表大约慢了3倍左右。

继续，我们可以查询当日网站中，哪些网页是被最频繁访问的:

SQL> select replace((replace(acc_file,'GET ','http://www.eygle.com')),'HTTP/1.1') accfile,ct from (
2 select ACC_FILE,count(*) ct from eygle_access_log_20061016
3 where acc_file like '%htm%'
4 group by acc_file order by ct desc)
5 where rownum <21;

ACCFILE CT
-------------------------------------------------------------------------------- ----------
http://www.eygle.com/index-tech.htm 110
http://www.eygle.com/archives/2006/10/wish_home.html 103
http://www.eygle.com/index-ha.htm 79
http://www.eygle.com/me/fairy_tale_leaf.htm 77
http://www.eygle.com/archives/2006/11/use_oracle_external_table.html 73
http://www.eygle.com/index-sql.htm 69
http://www.eygle.com/archives/2006/10/tom_oracle_9i10g.html 68
http://www.eygle.com/archives/2008/08/my_book_services.html 63
http://www.eygle.com/archives/2006/11/welcome_friend.html 62
http://www.eygle.com/archives/2006/10/veritas_vcs_simulator.html 61
http://www.eygle.com/index-case.htm 60
http://www.eygle.com/archives/2004/08/aoaouiiciona.html 59
http://www.eygle.com/archives/2006/08/oracle_fundbook_recommand.html 52
http://www.eygle.com/archives/2006/08/5460_8174.html 49
http://www.eygle.com/archives/2004/12/gmailaeaeoa.html 48
http://www.eygle.com/archives/2005/06/howlsmovingcast.html 48
http://www.eygle.com/gbook/index.html 48
http://www.eygle.com/index-hist.htm 44
http://www.eygle.com/index-special.htm 41
http://www.eygle.com/index-f&l.htm 37
20 rows selected.

Elapsed: 00:00:06.31
SQL>

通过外部表及SQL查询，只要日志文件中存在的信息，都可以很容易的被获取和分析.

-The End-

我们看一下使用外部表访问跟踪文件的例子(我的例子是用SYS用户来完成的)。

首先创建一个指向跟踪文件的Directory：

[oracle@jumper oracle]$ sqlplus "/ as sysdba"

SQL*Plus: Release 9.2.0.4.0 - Production on Tue Oct 17 22:10:05 2006

Copyright (c) 1982, 2002, Oracle Corporation. All rights reserved.

Connected to:
Oracle9i Enterprise Edition Release 9.2.0.4.0 - Production
With the Partitioning option
JServer Release 9.2.0.4.0 - Production

SQL> show parameter user_dump

NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
user_dump_dest string /opt/oracle/admin/eygle/udump
SQL> create or replace directory udump
2 as '/opt/oracle/admin/eygle/udump';

Directory created.

SQL> col DIRECTORY_PATH for a30
SQL> select * from dba_directories;

OWNER DIRECTORY_NAME DIRECTORY_PATH
---------- ------------------------------ ------------------------------
SYS BDUMP /opt/oracle/admin/eygle/bdump
SYS UDUMP /opt/oracle/admin/eygle/udump

如果我们的某些操作需要生成跟踪文件:

SQL> alter session set sql_trace=true;

Session altered.

SQL> select count(*) from dba_users;

COUNT(*)
----------
7

SQL> alter session set sql_trace=false;

Session altered.

通过简单的脚本我们可以获得跟踪文件的名称:

SQL> set echo on
SQL> @gettrcname
SQL> SELECT d.VALUE
2 || '/'
3 || LOWER (RTRIM (i.INSTANCE, CHR (0)))
4 || '_ora_'
5 || p.spid
6 || '.trc' trace_file_name
7 FROM (SELECT p.spid
8 FROM SYS.v$mystat m, SYS.v$session s, SYS.v$process p
9 WHERE m.statistic# = 1 AND s.SID = m.SID AND p.addr = s.paddr) p,
10 (SELECT t.INSTANCE
11 FROM SYS.v$thread t, SYS.v$parameter v
12 WHERE v.NAME = 'thread'
13 AND (v.VALUE = 0 OR t.thread# = TO_NUMBER (v.VALUE))) i,
14 (SELECT VALUE
15 FROM SYS.v$parameter
16 WHERE NAME = 'user_dump_dest') d
17 /

TRACE_FILE_NAME
------------------------------------------------------------------------------------------------------------------------
/opt/oracle/admin/eygle/udump/eygle_ora_29731.trc

SQL>

然后我们可以创建外部表用以访问这个跟踪文件:

SQL> create table eygle_ora_29731 ( text varchar2(400) )
2 organization external (
3 type oracle_loader
4 default directory UDUMP
5 access parameters (
6 records delimited by newline
7 nobadfile
8 nodiscardfile
9 nologfile
10 )
11 location('eygle_ora_29731.trc')
12 ) reject limit unlimited
13 /

Table created.

最后，我们可以很容易的使用外部表访问和查询这个跟踪文件的内容:

SQL> set pagesize 100
SQL> select * from eygle_ora_29731;

TEXT
-----------------------------------------------------------------
/opt/oracle/admin/eygle/udump/eygle_ora_29731.trc
Oracle9i Enterprise Edition Release 9.2.0.4.0 - Production
With the Partitioning option
JServer Release 9.2.0.4.0 - Production
ORACLE_HOME = /opt/oracle/product/9.2.0
System name: Linux
Node name: jumper.hurray.com.cn
Release: 2.4.21-15.EL
Version: #1 Thu Apr 22 00:27:41 EDT 2004
Machine: i686
Instance name: eygle
Redo thread mounted by this instance: 1
Oracle process number: 8
Unix process pid: 29731
*** 2006-10-17 22:15:50.310
*** SESSION ID:(7.141) 2006-10-17 22:15:50.310
APPNAME mod='sqlplus@jumper.hurray.com.cn (TNS V1-V3)' mh=0 act='' ah=0
=====================
PARSING IN CURSOR #1 len=32 dep=0 uid=0 oct=42 lid=0 tim=1133881396787772 hv=3943786303 ad='5ac609d4'
alter session set sql_trace=true
END OF STMT
EXEC #1:c=0
=====================
PARSING IN CURSOR #3 len=37 dep=1 uid=0 oct=3 lid=0 tim=1133881402205250 hv=3468666020 ad='5adae2d8'
select text from view$ where rowid=:1
END OF STMT
PARSE #3:c=0
EXEC #3:c=0
FETCH #3:c=0
=====================
PARSING IN CURSOR #4 len=116 dep=2 uid=0 oct=3 lid=0 tim=1133881402238727 hv=431456802 ad='5af20410'
select o.owner#
END OF STMT
PARSE #4:c=10000
EXEC #4:c=0
FETCH #4:c=0
STAT #3 id=1 cnt=1 pid=0 pos=1 obj=62 op='TABLE ACCESS BY USER ROWID VIEW$ (cr=1 r=0 w=0 time=79 us)'
=====================
PARSING IN CURSOR #1 len=30 dep=0 uid=0 oct=3 lid=0 tim=1133881402246452 hv=3198913505 ad='5adaa4ac'
select count(*) from dba_users
END OF STMT
PARSE #1:c=10000
EXEC #1:c=0
FETCH #1:c=10000
FETCH #1:c=0
STAT #1 id=1 cnt=1 pid=0 pos=1 obj=0 op='SORT AGGREGATE (cr=48 r=0 w=0 time=5276 us)'
STAT #1 id=2 cnt=7 pid=1 pos=1 obj=0 op='MERGE JOIN (cr=48 r=0 w=0 time=5250 us)'
STAT #1 id=3 cnt=7 pid=2 pos=1 obj=0 op='SORT JOIN (cr=45 r=0 w=0 time=5051 us)'
=====================
PARSING IN CURSOR #3 len=116 dep=1 uid=0 oct=3 lid=0 tim=1133881410335382 hv=431456802 ad='5af20410'
select o.owner#
END OF STMT
PARSE #3:c=0
EXEC #3:c=0
FETCH #3:c=0
STAT #1 id=4 cnt=7 pid=3 pos=1 obj=94 op='TABLE ACCESS BY INDEX ROWID PROFILE$ (cr=45 r=0 w=0 time=4881 us)'
STAT #1 id=5 cnt=127 pid=4 pos=1 obj=0 op='NESTED LOOPS (cr=44 r=0 w=0 time=3951 us)'
STAT #1 id=6 cnt=7 pid=5 pos=1 obj=0 op='NESTED LOOPS (cr=42 r=0 w=0 time=3235 us)'
STAT #1 id=7 cnt=7 pid=6 pos=1 obj=0 op='NESTED LOOPS (cr=26 r=0 w=0 time=3021 us)'
STAT #1 id=8 cnt=7 pid=7 pos=1 obj=0 op='MERGE JOIN (cr=10 r=0 w=0 time=2760 us)'
STAT #1 id=9 cnt=9 pid=8 pos=1 obj=0 op='SORT JOIN (cr=6 r=0 w=0 time=2479 us)'
STAT #1 id=10 cnt=9 pid=9 pos=1 obj=94 op='TABLE ACCESS BY INDEX ROWID PROFILE$ (cr=6 r=0 w=0 time=2284 us)'
STAT #1 id=11 cnt=163 pid=10 pos=1 obj=0 op='NESTED LOOPS (cr=5 r=0 w=0 time=1303 us)'
EXEC #3:c=0
FETCH #3:c=0
STAT #1 id=12 cnt=9 pid=11 pos=1 obj=280 op='TABLE ACCESS FULL USER_ASTATUS_MAP (cr=3 r=0 w=0 time=244 us)'
STAT #1 id=13 cnt=153 pid=11 pos=2 obj=139 op='INDEX RANGE SCAN I_PROFILE (cr=2 r=0 w=0 time=518 us)'
STAT #1 id=14 cnt=7 pid=8 pos=2 obj=0 op='SORT JOIN (cr=4 r=0 w=0 time=200 us)'
STAT #1 id=15 cnt=7 pid=14 pos=1 obj=22 op='TABLE ACCESS FULL USER$ (cr=4 r=0 w=0 time=117 us)'
STAT #1 id=16 cnt=7 pid=7 pos=2 obj=16 op='TABLE ACCESS CLUSTER TS$ (cr=16 r=0 w=0 time=199 us)'
STAT #1 id=17 cnt=7 pid=16 pos=1 obj=7 op='INDEX UNIQUE SCAN I_TS# (cr=2 r=0 w=0 time=69 us)'
STAT #1 id=18 cnt=7 pid=6 pos=2 obj=16 op='TABLE ACCESS CLUSTER TS$ (cr=16 r=0 w=0 time=166 us)'
STAT #1 id=19 cnt=7 pid=18 pos=1 obj=7 op='INDEX UNIQUE SCAN I_TS# (cr=2 r=0 w=0 time=37 us)'
STAT #1 id=20 cnt=119 pid=5 pos=2 obj=139 op='INDEX RANGE SCAN I_PROFILE (cr=2 r=0 w=0 time=342 us)'
STAT #1 id=21 cnt=7 pid=2 pos=2 obj=0 op='SORT JOIN (cr=3 r=0 w=0 time=121 us)'
EXEC #3:c=0
FETCH #3:c=0
STAT #1 id=22 cnt=1 pid=21 pos=1 obj=95 op='TABLE ACCESS FULL PROFNAME$ (cr=3 r=0 w=0 time=66 us)'
=====================
PARSING IN CURSOR #1 len=33 dep=0 uid=0 oct=42 lid=0 tim=1133881410339660 hv=4238949625 ad='5ac49f60'
alter session set sql_trace=false
END OF STMT
PARSE #1:c=0
EXEC #1:c=0

84 rows selected.

SQL>

外部表极大的简化了我们的很多管理工作。

-The End-

对于DBA来说，最常见一个例子是可以使用外部表来访问警告日志文件或其他跟踪文件.
以下一个例子用来说明外部表的用途。

首先需要创建一个Directory：

[oracle@jumper oracle]$ sqlplus "/ as sysdba"

SQL*Plus: Release 9.2.0.4.0 - Production on Sun Oct 15 21:42:28 2006

Copyright (c) 1982, 2002, Oracle Corporation. All rights reserved.

Connected to:
Oracle9i Enterprise Edition Release 9.2.0.4.0 - Production
With the Partitioning option
JServer Release 9.2.0.4.0 - Production

SQL> create or replace directory bdump
2 as '/opt/oracle/admin/eygle/bdump';

Directory created.

SQL> col DIRECTORY_PATH for a30
SQL> col owner for a10
SQL> select * from dba_directories;

OWNER DIRECTORY_NAME DIRECTORY_PATH
---------- ------------------------------ ------------------------------
SYS BDUMP /opt/oracle/admin/eygle/bdump

然后创建一个外部表:

SQL> create table alert_log ( text varchar2(400) )
2 organization external (
3 type oracle_loader
4 default directory BDUMP
5 access parameters (
6 records delimited by newline
7 nobadfile
8 nodiscardfile
9 nologfile
10 )
11 location('alert_eygle.log')
12 )
13 reject limit unlimited
14 /

Table created.

SQL> select * from alert_log where rownum < 51;

TEXT
-----------------------------------------------------------------------------------------
Mon Jun 26 12:00:24 2006
Starting ORACLE instance (normal)
Mon Jun 26 12:00:25 2006
WARNING: EINVAL creating segment of size 0x0000000008c00000
fix shm parameters in /etc/system or equivalent
LICENSE_MAX_SESSION = 0
LICENSE_SESSIONS_WARNING = 0
SCN scheme 2
Using log_archive_dest parameter default value
LICENSE_MAX_USERS = 0
SYS auditing is disabled
Starting up ORACLE RDBMS Version: 9.2.0.4.0.
System parameters with non-default values:
processes = 150
timed_statistics = TRUE
shared_pool_size = 104857600
large_pool_size = 0
java_pool_size = 0
control_files = /opt/oracle/oradata/eygle/control01.ctl
db_block_size = 8192
db_cache_size = 16777216
db_cache_advice = ON
compatible = 9.2.0.0.0
db_file_multiblock_read_count= 16
fast_start_mttr_target = 300
log_checkpoints_to_alert = TRUE
undo_management = AUTO
undo_tablespace = UNDOTBS1
undo_retention = 10800
remote_login_passwordfile= EXCLUSIVE
db_domain =
instance_name = eygle
job_queue_processes = 10
hash_join_enabled = TRUE
background_dump_dest = /opt/oracle/admin/eygle/bdump
user_dump_dest = /opt/oracle/admin/eygle/udump
core_dump_dest = /opt/oracle/admin/eygle/cdump
sort_area_size = 524288
db_name = eygle
open_cursors = 500
star_transformation_enabled= FALSE
query_rewrite_enabled = FALSE
pga_aggregate_target = 52428800
aq_tm_processes = 0
PMON started with pid=2
DBW0 started with pid=3
LGWR started with pid=4
CKPT started with pid=5
SMON started with pid=6
RECO started with pid=7

50 rows selected.

SQL>

SQL> SELECT deptno, ename,
  2         ROW_NUMBER () OVER (PARTITION BY deptno ORDER BY sal DESC) seq
  3    FROM emp;

    DEPTNO ENAME             SEQ
---------- ---------- ----------
        10 KING                1
        10 CLARK               2
        10 MILLER              3
        20 SCOTT               1
        20 FORD                2
        20 JONES               3
        20 ADAMS               4
        20 SMITH               5
        30 BLAKE               1
        30 ALLEN               2
        30 TURNER              3
        30 WARD                4
        30 MARTIN              5
        30 JAMES               6

14 rows selected.

SQL> select deptno,
  2  max(decode(seq,1,ename,null)) highest,
  3  max(decode(seq,2,ename,null)) second,
  4  max(decode(seq,3,ename,null)) third
  5  from (
  6  select deptno,ename,
  7  row_number() over
  8  (partition by deptno order by sal desc) seq
  9  from emp)
 10  where seq <=3 group by deptno
 11  /

    DEPTNO HIGHEST    SECOND     THIRD
---------- ---------- ---------- ----------
        10 KING       CLARK      MILLER
        20 SCOTT      FORD       JONES
        30 BLAKE      ALLEN      TURNER

最初的问题是这样的:

我有一个表结构，
fphm,kshm
2014,00000001
2014,00000002
2014,00000003
2014,00000004
2014,00000005
2014,00000007
2014,00000008
2014,00000009
2013,00000120
2013,00000121
2013,00000122
2013,00000124
2013,00000125

(第二个字段内可能是连续的数据，可能存在断点。)

怎样能查询出来这样的结果,查询出连续的记录来。
就像下面的这样？
2014,00000001,00000005
2014,00000009,00000007
2013,00000120,00000122
2013,00000124,00000125

ITPUB上的朋友给出了一个非常巧妙的答案:

SQL> SELECT b.fphm, MIN (b.kshm) Start_HM, MAX (b.kshm) End_HM
2 FROM (SELECT a.*, TO_NUMBER (a.kshm - ROWNUM) cc
3 FROM (SELECT *
4 FROM t
5 ORDER BY fphm, kshm) a) b
6 GROUP BY b.fphm, b.cc
7 /

FPHM START_HM END_HM
---------- -------- --------
2013 00000120 00000122
2013 00000124 00000125
2014 00000001 00000005
2014 00000007 00000009

巧思妙想，就在一念之间。
ITPUB其他参考链接如下:
http://blog.itpub.net/post/5042/27936

-The End-

以下是一个简单范例说明:

SQL> CREATE VIEW d10_emp
  2  ( empno, ename, job,deptno,
  3  CONSTRAINT pk_d10
  4  PRIMARY KEY (ename)
  5  RELY DISABLE NOVALIDATE
  6  ) AS
  7  SELECT empno, ename, job ,deptno
  8  FROM emp
  9  WHERE deptno = 10;  

View created.

SQL> select * from d10_emp;

     EMPNO ENAME      JOB           DEPTNO
---------- ---------- --------- ----------
      7782 CLARK      MANAGER           10
      7839 KING       PRESIDENT         10
      7934 MILLER     CLERK             10

SQL> select CONSTRAINT_NAME,TABLE_NAME,CONSTRAINT_TYPE from user_constraints
  2  where table_name='D10_EMP';

CONSTRAINT_NAME                TABLE_NAME                     C
------------------------------ ------------------------------ -
PK_D10                         D10_EMP                        P 

由于此约束仅仅为"约束声明",所以不具有实际约束力:

SQL> insert into d10_emp values(7777,'EYGLE','MANAGER',10);

1 row created.

SQL> insert into d10_emp values(8888,'EYGLE','MANAGER',10);

1 row created.

SQL> select * from d10_emp;

     EMPNO ENAME      JOB           DEPTNO
---------- ---------- --------- ----------
      7782 CLARK      MANAGER           10
      7839 KING       PRESIDENT         10
      7934 MILLER     CLERK             10
      7777 EYGLE      MANAGER           10
      8888 EYGLE      MANAGER           10 

SQL> CREATE TABLE employees
  2  (
  3    empno NUMBER(6),
  4    NAME VARCHAR2(30),
  5    dept_no NUMBER(2),
  6    CONSTRAINT emp_pk PRIMARY KEY(empno)
  7      USING INDEX
  8      (CREATE UNIQUE INDEX emp_pk_idx ON employees(empno))
  9  );

Table created.

SQL> select index_name,UNIQUENESS from user_indexes
  2  where table_name=upper('employees');

INDEX_NAME                     UNIQUENES
------------------------------ ---------
EMP_PK_IDX                     UNIQUE

SQL> select CONSTRAINT_NAME,CONSTRAINT_TYPE,INDEX_NAME from user_constraints
  2  where table_name=upper('employees');

CONSTRAINT_NAME                C INDEX_NAME
------------------------------ - ------------------------------
EMP_PK                         P EMP_PK_IDX 

Oracle9i中新增的命令可以用以DROP约束保留索引.

ALTER TABLE employees DROP PRIMARY KEY KEEP INDEX;

SQL> ALTER TABLE employees DROP PRIMARY KEY KEEP INDEX;

Table altered.

SQL> select index_name,UNIQUENESS from user_indexes
  2  where table_name=upper('employees');

INDEX_NAME                     UNIQUENES
------------------------------ ---------
EMP_PK_IDX                     UNIQUE 

这时候我们可以使用Oracle提供的另外一个Hints: Leading 提示.

这个Hints在Oracle9i中的含义为:

The LEADING hint causes Oracle to use the specified table as the first table in the join order.

If you specify two or more LEADING hints on different tables, then all of them are ignored. If you specify the ORDERED hint, then it overrides all LEADING hints.

通过Leading 和 use_hash 提示连用,我们可以巧妙的影响SQL中表和结果集的Join顺序.

我们通过如下示例看一下这个提示是如何影响SQL执行的:

SQL> SELECT /*+ leading(t_max) use_hash(t_max t_middle) */ COUNT (*)
  2    FROM t_small, t_max, t_middle
  3  WHERE t_small.object_id = t_middle.object_id
  4  AND t_middle.object_id = t_max.object_id
  5  /

Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=262 Card=1 Bytes=12)
   1    0   SORT (AGGREGATE)
   2    1     HASH JOIN (Cost=262 Card=400 Bytes=4800)
   3    2       HASH JOIN (Cost=225 Card=113776 Bytes=910208)
   4    3         TABLE ACCESS (FULL) OF 'T_MAX' (Cost=151 Card=113792 Bytes=455168)
   5    3         TABLE ACCESS (FULL) OF 'T_MIDDLE' (Cost=39 Card=28447 Bytes=113788)
   6    2       TABLE ACCESS (FULL) OF 'T_SMALL' (Cost=2 Card=100 Bytes=400) 

我们看到,通过这两个Hints的联合使用,该查询首先对T_MAX和T_MIDDLE表进行HASH JOIN,再以这个结果集同T_SMALL进行HASH JION.

通过ordered提示，可以避免CBO SQL解析过程中的表连接评估，从而避免Oracle产生错误的执行计划，或者强制Oracle按照我们指定的方式执行。

在很多时候，当我们清楚地了解数据结构和数据分布之后，就可以通过ORDERED提示来提高SQL性能。

通过以下例子我们来说明一下Ordered提示的作用.

1.不加Hints时SQL的执行计划

SQL> set autotrace trace explain
SQL>  SELECT COUNT (*)
  2    FROM t_small, t_max, t_middle
  3   WHERE t_small.object_id = t_middle.object_id
  4   AND t_middle.object_id = t_max.object_id;

Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=194 Card=1 Bytes=12)
   1    0   SORT (AGGREGATE)
   2    1     HASH JOIN (Cost=194 Card=400 Bytes=4800)
   3    2       HASH JOIN (Cost=42 Card=100 Bytes=800)
   4    3         TABLE ACCESS (FULL) OF 'T_SMALL' (Cost=2 Card=100 Bytes=400)
   5    3         TABLE ACCESS (FULL) OF 'T_MIDDLE' (Cost=39 Card=28447 Bytes=113788)
   6    2       TABLE ACCESS (FULL) OF 'T_MAX' (Cost=151 Card=113792 Bytes=455168)
 

我们可以通过10053事件跟踪一下该SQL的解析:

SQL> alter session set events='10053 trace name context forever,level 1';

Session altered.

SQL> explain plan for
  2  SELECT COUNT (*)
  3    FROM t_small, t_max, t_middle
  4  WHERE t_small.object_id = t_middle.object_id
  5  AND t_middle.object_id = t_max.object_id;    

Explained. 

查看Trace文件可以看到,Oracle需要进行3! (6)次表连接顺序的评估:

bash-2.03$ cat testora9_ora_10862.trc |grep "Join order"
Join order[1]: T_SMALL [T_SMALL] T_MIDDLE [T_MIDDLE] T_MAX [T_MAX] 
Join order[2]: T_SMALL [T_SMALL] T_MAX [T_MAX] T_MIDDLE [T_MIDDLE] 
Join order[3]: T_MIDDLE [T_MIDDLE] T_SMALL [T_SMALL] T_MAX [T_MAX] 
Join order[4]: T_MIDDLE [T_MIDDLE] T_MAX [T_MAX] T_SMALL [T_SMALL] 
Join order[5]: T_MAX [T_MAX] T_SMALL [T_SMALL] T_MIDDLE [T_MIDDLE] 
Join order[6]: T_MAX [T_MAX] T_MIDDLE [T_MIDDLE] T_SMALL [T_SMALL]  

很多时候,我们看似可以使用全索引扫描(Full Index Scan)的情况,可能Oracle就会因为Null值的存在而放弃索引.

在此情况下即使使用Hints,Oracle也不会使用索引,其根本原因就是因为Null值的存在.

我们看以下测试.

在username字段为Not Null时,Index Hints可以生效.

SQL> create table t as select username,password from dba_users;

Table created.

SQL> desc t
 Name                                      Null?    Type
 ----------------------------------------- -------- ----------------------------
 USERNAME                                  NOT NULL VARCHAR2(30)
 PASSWORD                                           VARCHAR2(30)

SQL> create index i_t on t(username);

Index created.

SQL> set autotrace trace explain
SQL> select * from t where username='EYGLE';

Execution Plan
----------------------------------------------------------
Plan hash value: 1601196873

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |     1 |    34 |     2   (0)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| T    |     1 |    34 |     2   (0)| 00:00:01 |
--------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("USERNAME"='EYGLE')

Note
-----
   - dynamic sampling used for this statement

SQL> set linesize 120
SQL> select /*+ index(t,i_t) */ * from t where username='EYGLE';

Execution Plan
----------------------------------------------------------
Plan hash value: 2928007915

------------------------------------------------------------------------------------
| Id  | Operation                   | Name | Rows  | Bytes | Cost (%CPU)| Time     |
------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |      |     1 |    34 |     2   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| T    |     1 |    34 |     2   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | I_T  |     1 |       |     1   (0)| 00:00:01 |
------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   2 - access("USERNAME"='EYGLE')

Note
-----
   - dynamic sampling used for this statement

Index Hints的格式如下:

/*+ INDEX ( table [index [index]...] ) */

我们简单看一下这个提示的用法(范例为Oracle10g数据库):

SQL> create table t as select username,password from dba_users;

Table created.

SQL> create index i_t on t(username);

Index created.

SQL> set autotrace trace explain

SQL> select /*+ index(t i_t) */ * from t where username='EYGLE';

Execution Plan
----------------------------------------------------------
Plan hash value: 2928007915

------------------------------------------------------------------------------------
| Id  | Operation                   | Name | Rows  | Bytes | Cost (%CPU)| Time     |
------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |      |     1 |    34 |     2   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| T    |     1 |    34 |     2   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | I_T  |     1 |       |     1   (0)| 00:00:01 |
------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   2 - access("USERNAME"='EYGLE')

Note
-----
   - dynamic sampling used for this statement 

这里的查询使用了索引.

需要注意的是使用CTAS方式创建数据表,新建表会继承原表的约束属性:

SQL> desc t
 Name                                      Null?    Type
 ----------------------------------------- -------- ----------------------------
 USERNAME                                  NOT NULL VARCHAR2(30)
 PASSWORD                                           VARCHAR2(30) 

在oralce中给自建函数创建索引，结果不成功。

source:Create Index IDX_T_SP_TWOTYPESTAT_0_f On T_SP_TWOTYPESTAT_0(f_dateadd(yearmonth,12,2));
err:the function is not deterministic.

我们看一下这是为什么?

随便一个测试可以再现这个问题,我门创建一个函数(本范例函数用于进行16进制向10进制转换):

CREATE OR REPLACE FUNCTION h2ten (
   p_str         IN   VARCHAR2,
   p_from_base   IN   NUMBER DEFAULT 16
)
   RETURN NUMBER 
IS
   l_num   NUMBER        DEFAULT 0;
   l_hex   VARCHAR2 (16) DEFAULT '0123456789ABCDEF';
BEGIN
   FOR i IN 1 .. LENGTH (p_str)
   LOOP
      l_num :=
         l_num * p_from_base + INSTR (l_hex, UPPER (SUBSTR (p_str, i, 1)))
         - 1;
   END LOOP;

   RETURN l_num;
END h2ten;
 

此时创建索引,获得如下错误信息:

SQL> create table t as select username,'a' hex from dba_users;

Table created

SQL> create index i_t on t (h2ten(hex));

create index i_t on t (h2ten(hex))

ORA-30553: The function is not deterministic
 

在Oracle8i以前，所有已编译存储对象(包括packages, procedures, functions, triggers, and views)只能以定义者(Definer)身份解析运行；从Oracle8i开始，Oracle引入调用者(invoker)权限，使得对象可以以调用者身份和权限执行。

定义者(Definer)指编译存储对象的所有者.
调用者(Invoker)指拥有当前会话权限的模式，这可能和当前登录用户相同或不同(alter session set current_schema 可以改变调用者Schema).

TOM在他的《Expert One on One》的第23章曾经详细介绍这一特性，本文引用Tom的一个例子用于说明Definer and Invoker权限。

1.以Eygle用户(definer)创建2个过程

 $ sqlplus eygle/eygle

SQL*Plus: Release 9.2.0.4.0 - Production on Sun Dec 11 11:39:27 2005

Copyright (c) 1982, 2002, Oracle Corporation.  All rights reserved.

Connected to:
Oracle9i Enterprise Edition Release 9.2.0.4.0 - 64bit Production
With the Partitioning option
JServer Release 9.2.0.4.0 - Production

SQL> create or replace procedure definer_proc
  2  as
  3  begin
  4      for x in
  5      ( select sys_context( 'userenv', 'current_user' ) current_user,
  6               sys_context( 'userenv', 'session_user' ) session_user,
  7               sys_context( 'userenv', 'current_schema' ) current_schema
  8          from dual )
  9      loop
 10          dbms_output.put_line( 'Current User:   ' || x.current_user );
 11          dbms_output.put_line( 'Session User:   ' || x.session_user );
 12          dbms_output.put_line( 'Current Schema: ' || x.current_schema );
 13      end loop;
 14  end;
 15  /

Procedure created.

SQL> 
SQL> grant execute on definer_proc to test;

Grant succeeded.

SQL> 
SQL> create or replace procedure invoker_proc
  2  AUTHID CURRENT_USER
  3  as
  4  begin
  5      for x in
  6      ( select sys_context( 'userenv', 'current_user' ) current_user,
  7               sys_context( 'userenv', 'session_user' ) session_user,
  8               sys_context( 'userenv', 'current_schema' ) current_schema
  9          from dual )
 10      loop
 11          dbms_output.put_line( 'Current User:   ' || x.current_user );
 12          dbms_output.put_line( 'Session User:   ' || x.session_user );
 13          dbms_output.put_line( 'Current Schema: ' || x.current_schema );
 14      end loop;
 15  end;
 16  /

Procedure created.

SQL> 
SQL> grant execute on invoker_proc to test;

Grant succeeded.

注意invoker权限的本质是引入了AUTHID CURRENT_USER子句，通过此句Oracle得以使用invoker身份编译执行对象。

2.以test用户(invoker)身份执行

SQL> connect test/test
Connected.
SQL> 
SQL> set serveroutput on
SQL> exec eygle.definer_proc
Current User:   EYGLE
Session User:   TEST
Current Schema: EYGLE

PL/SQL procedure successfully completed.

SQL> exec eygle.invoker_proc
Current User:   TEST
Session User:   TEST
Current Schema: TEST

PL/SQL procedure successfully completed.

注意只有使用invoker者权限执行时，Schema才转换为TEST.

FORALL index IN lower_bound..upper_bound sql_statement;

SQL> drop table blktest;

Table dropped.

Elapsed: 00:00:00.13
SQL> 
SQL> CREATE TABLE blktest (num NUMBER(20), name varchar2(50));

Table created.

Elapsed: 00:00:00.08
SQL> 
SQL> CREATE OR REPLACE PROCEDURE bulktest
  2  IS
  3     TYPE numtab IS TABLE OF NUMBER (20)
  4        INDEX BY BINARY_INTEGER;
  5  
  6     TYPE nametab IS TABLE OF VARCHAR2 (50)
  7        INDEX BY BINARY_INTEGER;
  8  
  9     pnums    numtab;
 10     pnames   nametab;
 11     t1       NUMBER;
 12     t2       NUMBER;
 13     t3       NUMBER;
 14  BEGIN
 15     FOR j IN 1 .. 1000000
 16     LOOP
 17        pnums (j)         := j;
 18        pnames (j)        := 'Seq No. ' || TO_CHAR (j);
 19     END LOOP;
 20  
 21     SELECT DBMS_UTILITY.get_time
 22       INTO t1
 23       FROM DUAL;
 24  
 25     FOR i IN 1 .. 1000000
 26     LOOP
 27        INSERT INTO blktest
 28             VALUES (pnums (i), pnames (i));
 29     END LOOP;
 30  
 31     SELECT DBMS_UTILITY.get_time
 32       INTO t2
 33       FROM DUAL;
 34  
 35     FORALL i IN 1 .. 1000000
 36        INSERT INTO blktest
 37             VALUES (pnums (i), pnames (i));
 38  
 39     SELECT DBMS_UTILITY.get_time
 40       INTO t3
 41       FROM DUAL;
 42  
 43     DBMS_OUTPUT.put_line ('Execution Time (hsecs)');
 44     DBMS_OUTPUT.put_line ('---------------------');
 45     DBMS_OUTPUT.put_line ('FOR loop: ' || TO_CHAR (t2 - t1));
 46     DBMS_OUTPUT.put_line ('FORALL:   ' || TO_CHAR (t3 - t2));
 47  END;
 48  /

Procedure created.

Elapsed: 00:00:01.46
SQL> exec bulktest;
Execution Time (hsecs)
---------------------
FOR loop: 30361
FORALL:   4792

PL/SQL procedure successfully completed.

Elapsed: 00:06:32.92

The syntax is:

DUMP(expr[,return_fmt[,start_position[,length]]])

The argument return_fmt specifies the format of the return value and can have any of the following values:

• 8 returns result in octal notation.
• 10 returns result in decimal notation.
• 16 returns result in hexadecimal notation.
• 17 returns result as single characters.

By default, the return value contains no character set information. To retrieve the character set name of expr, specify any of the preceding format values, plus 1000. For example, a return_fmt of 1008 returns the result in octal, plus provides the character set name of expr.

The arguments start_position and length combine to determine which portion of the internal representation to return. The default is to return the entire internal representation in decimal notation.

If expr is null, then this function returns a null.

Examples

The following examples show how to extract dump information from a string expression and a column:

SELECT DUMP('abc', 1016)
   FROM DUAL;

DUMP('ABC',1016)
------------------------------------------
Typ=96 Len=3 CharacterSet=WE8DEC: 61,62,63

SELECT DUMP(last_name, 8, 3, 2) "OCTAL"
   FROM employees
   WHERE last_name = 'Hunold';

OCTAL
-------------------------------------------------------------------
Typ=1 Len=6: 156,157

SELECT DUMP(last_name, 10, 3, 2) "ASCII"
   FROM employees
   WHERE last_name = 'Hunold';

ASCII
--------------------------------------------------------------------
Typ=1 Len=6: 110,111