delete 操作不會改變表的高水標記，因此如果我們對一個表插入1000萬條數據，然后再回滾（對insert操作做回滾相當於相應地做delete操作），會使表的高水標記增長得很高，這時雖然我們操作的表依然是一個空表，但是查詢它卻會讀驚人數量的內存塊，實驗如下：
ETL@RACTEST> create table test_table (a number);
Table created.
Elapsed: 00:00:00.01
ETL@RACTEST> set autotrace traceonly statistics;
ETL@RACTEST> select * from test_table;
no rows selected
Elapsed: 00:00:00.00
Statistics
----------------------------------------------------------
24 recursive calls
0 db block gets
7 consistent gets
0 physical reads
0 redo size
318 bytes sent via SQL*Net to client
453 bytes received via SQL*Net from client
1 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
0 rows processed
可以看到查詢這個空表讀了7個內存塊。然后我們向表中插入1000萬條數據后再回滾：
insert into test_table select level from dual connect by level<=10000000;
10000000 rows created.
Elapsed: 00:00:58.38
ETL@RACTEST> rollback;
Rollback complete.
Elapsed: 00:00:01.15
ETL@RACTEST> set autotrace traceonly statistics;
ETL@RACTEST> select * from test_table;
no rows selected
Elapsed: 00:00:00.14
Statistics
----------------------------------------------------------
283 recursive calls
1 db block gets
15463 consistent gets
0 physical reads
176 redo size
318 bytes sent via SQL*Net to client
453 bytes received via SQL*Net from client
1 SQL*Net roundtrips to/from client
4 sorts (memory)
0 sorts (disk)
0 rows processed
可以看到，同樣是讀一個空表，但是卻做了15463(consistent gets)+1(db block gets)個邏輯讀，可見delete操作不會修改表的高水標記。並且delete操作也很耗費時間，因此我們通常在想清空一個表的數據時用truncate來替代delete。truncate會以一種快速的方式清空表，只產生很少的 日志信息，並且會將高水標記清零。例如：
ETL@RACTEST> insert into test_table select level from dual connect by level<=10000000;
10000000 rows created.
Elapsed: 00:00:32.45
ETL@RACTEST> commit;
Commit complete.
Elapsed: 00:00:00.02
ETL@RACTEST> truncate table test_table;
Table truncated.
Elapsed: 00:00:29.52
ETL@RACTEST> set autotrace traceonly statistics;
ETL@RACTEST> select * from test_table;
no rows selected
Elapsed: 00:00:00.00
Statistics
----------------------------------------------------------
1 recursive calls
1 db block gets
6 consistent gets
0 physical reads
96 redo size
318 bytes sent via SQL*Net to client
453 bytes received via SQL*Net from client
1 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
0 rows processed
truncate后再查詢只做了7個邏輯讀，也就是讀了7個內存塊。
但是truncate操作有一個限制，例如B表有一個外鍵參照A表的一個字段，那么truncate A表Oracle就會報錯，實驗如下：
create table ttt1 (a number,primary key(a));
Table created.
Elapsed: 00:00:00.02
ETL@RACTEST> insert into ttt1 select level from dual connect by level<=10;
10 rows created.
Elapsed: 00:00:00.02
ETL@RACTEST> commit;
Commit complete.
Elapsed: 00:00:00.00
ETL@RACTEST> select * from ttt1;
A
----------
1
2
3
4
5
6
7
8
9
10
10 rows selected.
Elapsed: 00:00:00.00
ETL@RACTEST> create table ttt2 (a number,constraint fk_ttt2_a foreign key(a) references ttt1(a));
Table created.
這時ttt2表的a字段參照了ttt1表的a字段，我如果用delete操作清空ttt1表的值是可以的，因為ttt2表沒有數據，刪除ttt1的數據不會違背外鍵參照完整性。
ETL@RACTEST> delete from ttt1;
10 rows deleted.
但是如果我們使用truncate,oracle則要報錯：
ETL@RACTEST> truncate table ttt1;
truncate table ttt1
*
ERROR at line 1:
ORA-02266: unique/primary keys in table referenced by enabled foreign keys
因為truncate實現的是一種快速刪除，因此它無法檢驗參照完整性，即使刪除的數據不違反參照完整性，Oracle也不允許做truncate操作，也就是我們說的寧可錯殺一千，不可放過一個，呵呵。
我在培訓的時候有人說外鍵如果設置了級聯刪除就可以truncate，但實際上是不行的，實驗如下：
ETL@RACTEST> drop table ttt2 purge;
Table dropped.
Elapsed: 00:00:00.03
ETL@RACTEST> create table ttt2 (a number,constraint fk_ttt2_a foreign key(a) references ttt1(a) on delete cascade);
Table created.
Elapsed: 00:00:00.01
ETL@RACTEST> truncate table ttt1;
truncate table ttt1
*
ERROR at line 1:
ORA-02266: unique/primary keys in table referenced by enabled foreign keys
Elapsed: 00:00:00.00
這是一種方式的級聯刪除，trucate仍然報錯，下面是另一種級聯：
ETL@RACTEST> drop table ttt2 purge;
Table dropped.
Elapsed: 00:00:00.01
ETL@RACTEST> create table ttt2 (a number,constraint fk_ttt2_a foreign key(a) references ttt1(a) on delete set null);
Table created.
Elapsed: 00:00:00.01
ETL@RACTEST> truncate table ttt1;
truncate table ttt1
*
ERROR at line 1:
ORA-02266: unique/primary keys in table referenced by enabled foreign keys
同樣不行，因此得出結論，只要有外鍵參照A表的字段，那么就不能對A表使用truncate操作！
0 db block gets
7 consistent gets
0 physical reads
0 redo size
318 bytes sent via SQL*Net to client
453 bytes received via SQL*Net from client
1 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
0 rows processed
可以看到查詢這個空表讀了7個內存塊。然后我們向表中插入1000萬條數據后再回滾：
insert into test_table select level from dual connect by level<=10000000;
10000000 rows created.
Elapsed: 00:00:58.38
ETL@RACTEST> rollback;
Rollback complete.
Elapsed: 00:00:01.15
ETL@RACTEST> set autotrace traceonly statistics;
ETL@RACTEST> select * from test_table;
no rows selected
Elapsed: 00:00:00.14
Statistics
----------------------------------------------------------
283 recursive calls
1 db block gets
15463 consistent gets
0 physical reads
176 redo size
318 bytes sent via SQL*Net to client
453 bytes received via SQL*Net from client
1 SQL*Net roundtrips to/from client
4 sorts (memory)
0 sorts (disk)
0 rows processed
可以看到，同樣是讀一個空表，但是卻做了15463(consistent gets)+1(db block gets)個邏輯讀，可見delete操作不會修改表的高水標記。並且delete操作也很耗費時間，因此我們通常在想清空一個表的數據時用truncate來替代delete。truncate會以一種快速的方式清空表，只產生很少的 日志信息，並且會將高水標記清零。例如：
ETL@RACTEST> insert into test_table select level from dual connect by level<=10000000;
10000000 rows created.
Elapsed: 00:00:32.45
ETL@RACTEST> commit;
Commit complete.
Elapsed: 00:00:00.02
ETL@RACTEST> truncate table test_table;
Table truncated.
Elapsed: 00:00:29.52
ETL@RACTEST> set autotrace traceonly statistics;
ETL@RACTEST> select * from test_table;
no rows selected
Elapsed: 00:00:00.00
Statistics
----------------------------------------------------------
1 recursive calls
1 db block gets
6 consistent gets
0 physical reads
96 redo size
318 bytes sent via SQL*Net to client
453 bytes received via SQL*Net from client
1 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
0 rows processed
truncate后再查詢只做了7個邏輯讀，也就是讀了7個內存塊。
但是truncate操作有一個限制，例如B表有一個外鍵參照A表的一個字段，那么truncate A表Oracle就會報錯，實驗如下：
create table ttt1 (a number,primary key(a));
Table created.
Elapsed: 00:00:00.02
ETL@RACTEST> insert into ttt1 select level from dual connect by level<=10;
10 rows created.
Elapsed: 00:00:00.02
ETL@RACTEST> commit;
Commit complete.
Elapsed: 00:00:00.00
ETL@RACTEST> select * from ttt1;
A
----------
1
2
3
4
5
6
7
8
9
10
10 rows selected.
Elapsed: 00:00:00.00
ETL@RACTEST> create table ttt2 (a number,constraint fk_ttt2_a foreign key(a) references ttt1(a));
Table created.
這時ttt2表的a字段參照了ttt1表的a字段，我如果用delete操作清空ttt1表的值是可以的，因為ttt2表沒有數據，刪除ttt1的數據不會違背外鍵參照完整性。
ETL@RACTEST> delete from ttt1;
10 rows deleted.
但是如果我們使用truncate,oracle則要報錯：
ETL@RACTEST> truncate table ttt1;
truncate table ttt1
*
ERROR at line 1:
ORA-02266: unique/primary keys in table referenced by enabled foreign keys
因為truncate實現的是一種快速刪除，因此它無法檢驗參照完整性，即使刪除的數據不違反參照完整性，Oracle也不允許做truncate操作，也就是我們說的寧可錯殺一千，不可放過一個，呵呵。
我在培訓的時候有人說外鍵如果設置了級聯刪除就可以truncate，但實際上是不行的，實驗如下：
ETL@RACTEST> drop table ttt2 purge;
Table dropped.
Elapsed: 00:00:00.03
ETL@RACTEST> create table ttt2 (a number,constraint fk_ttt2_a foreign key(a) references ttt1(a) on delete cascade);
Table created.
Elapsed: 00:00:00.01
ETL@RACTEST> truncate table ttt1;
truncate table ttt1
*
ERROR at line 1:
ORA-02266: unique/primary keys in table referenced by enabled foreign keys
Elapsed: 00:00:00.00
這是一種方式的級聯刪除，trucate仍然報錯，下面是另一種級聯：
ETL@RACTEST> drop table ttt2 purge;
Table dropped.
Elapsed: 00:00:00.01
ETL@RACTEST> create table ttt2 (a number,constraint fk_ttt2_a foreign key(a) references ttt1(a) on delete set null);
Table created.
Elapsed: 00:00:00.01
ETL@RACTEST> truncate table ttt1;
truncate table ttt1
*
ERROR at line 1:
ORA-02266: unique/primary keys in table referenced by enabled foreign keys
同樣不行，因此得出結論，只要有外鍵參照A表的字段，那么就不能對A表使用truncate操作！
轉自：http://blog.sina.com.cn/s/blog_6ff05a2c0100mira.html