关于Oracle的Dual表

有朋友问到关于Oracle的Dual表问题，并且提到了Tom的一个链接：
http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:1562813956388
很多人关心的是Dual的内部实现，这Oracle自然是不会披露的，不过我们可以从一些有限的资料获得关于Dual的印象：

There is internalized code that makes this happen. Code checks that ensure that a table scan of SYS.DUAL only returns one row. Svrmgrl behaviour is incorrect but this is now an obsolete product.
The base issue you should always remember and keep is: DUAL table should always have 1 ROW. Dual is a normal table with one dummy column of varchar2(1).
This is basically used from several applications as a pseudo table for getting results from a select statement that use functions like sysdate or other prebuilt or application functions.
If DUAL has no rows at all some applications (that use DUAL) may fail with NO_DATA_FOUND exception. If DUAL has more than 1 row then applications (that use DUAL) may fail with TOO_MANY_ROWS exception.
So DUAL should ALWAYS have 1 and only 1 row.

前两句话最为关键，实际上我们也容易猜到，Oracle通过内部代码来实现对于DUAL的访问和控制，并且通过Internal Code使得这个表与众不同。
Tom提到在Close了Database之后，可以看到这个表的内存地址及展现，这说明这个表的结构并不单纯：

SQL> select * from dual;
D
-
X
SQL> alter database close;
Database altered.
SQL> select * from dual;
ADDR          INDX    INST_ID D
-------- ---------- ---------- -
0A4F8858          0          1 X

注意不要更改Dual表的内容，否则可能引起数据库的问题。
如果该表意外删除，可以通过设置初始化参数replication_dependency_tracking = FALSE，重启数据库来重建该表。