Oracle基于时间点的恢复

getfuture

Oracle基于时间点的恢复能够精确到什么样的精度?
这是一个需要关心的问题。

以下测试用于进行一点说明。

1.首先做好冷备份
2.创建测试数据

D:\>sqlplus "/ as sysdba"

SQL*Plus: Release 9.2.0.6.0 - Production on Mon Jan 17 11:56:43 2005

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

Connected to an idle instance.

11:56:44 SQL> startup
ORACLE instance started.

Total System Global Area 101785428 bytes
Fixed Size 454484 bytes
Variable Size 75497472 bytes
Database Buffers 25165824 bytes
Redo Buffers 667648 bytes
Database mounted.
Database opened.

11:57:01 SQL> create table test (name varchar2(20));

Table created.

Elapsed: 00:00:00.04
11:57:23 SQL> insert into test values('aaaaaaaaaaaaaaaaaaaa');

1 row created.

Elapsed: 00:00:00.00
11:57:23 SQL> insert into test values('bbbbbbbbbbbbbbbbbbbb');

1 row created.

Elapsed: 00:00:00.00
11:57:23 SQL> insert into test values('cccccccccccccccccccc');

1 row created.

Elapsed: 00:00:00.00
11:57:24 SQL> commit;

Commit complete.

Elapsed: 00:00:00.00
11:57:28 SQL>
--注意这个时间，是Commit完成时间

11:57:29 SQL> drop table test;

Table dropped.

Elapsed: 00:00:00.07
11:57:34 SQL> shutdown immediate;
Database closed.
Database dismounted.
ORACLE instance shut down.
11:57:45 SQL> exit
Disconnected from Oracle9i Enterprise Edition Release 9.2.0.6.0 - Production
With the Partitioning, OLAP and Oracle Data Mining options
JServer Release 9.2.0.6.0 - Production

3.恢复备份数据
保留当前日志

D:\>sqlplus "/ as sysdba"

SQL*Plus: Release 9.2.0.6.0 - Production on Mon Jan 17 11:58:04 2005

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

Connected to an idle instance.

11:58:04 SQL> startup mount;
ORACLE instance started.

Total System Global Area 101785428 bytes
Fixed Size 454484 bytes
Variable Size 75497472 bytes
Database Buffers 25165824 bytes
Redo Buffers 667648 bytes
Database mounted.

SQL> select to_char(sysdate,'yyyy-mm-dd hh24:mi:ss') from dual;

11:58:15 SQL> alter session set nls_date_format='yyyy-mm-dd hh24:mi:ss';

Session altered.

Elapsed: 00:00:00.00
11:58:17 SQL> recover database until time '2005-01-17 11:57:28';
Media recovery complete.

recover database until time '2010-10-19 18:25:03';


--恢复到提交完成时刻

11:58:33 SQL> alter database open resetlogs;

Database altered.

Elapsed: 00:00:05.08
11:58:46 SQL> select * from test;

no rows selected

Elapsed: 00:00:00.00

--注意此时数据没有被恢复。
--也就是说，落在了提交之前


4.第二个测试

D:\>sqlplus "/ as sysdba"

SQL*Plus: Release 9.2.0.6.0 - Production on Mon Jan 17 11:48:50 2005

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

Connected to an idle instance.

11:48:50 SQL> startup
ORACLE instance started.

Total System Global Area 101785428 bytes
Fixed Size 454484 bytes
Variable Size 75497472 bytes
Database Buffers 25165824 bytes
Redo Buffers 667648 bytes
Database mounted.
Database opened.
11:49:03 SQL> create table test (name varchar2(20));

Table created.

Elapsed: 00:00:00.04
11:49:32 SQL> insert into test values('aaaaaaaaaaaaaaaaaaaa');

1 row created.

Elapsed: 00:00:00.00
11:49:32 SQL> insert into test values('bbbbbbbbbbbbbbbbbbbb');

1 row created.

Elapsed: 00:00:00.00
11:49:32 SQL> insert into test values('cccccccccccccccccccc');

1 row created.

Elapsed: 00:00:00.00
11:49:32 SQL> commit;

Commit complete.

Elapsed: 00:00:00.00
11:49:34 SQL>
--注意这里是提交时间
11:49:34 SQL>
11:49:35 SQL>
--等待时间流逝一秒
11:49:36 SQL>
11:49:37 SQL> drop table test;

Table dropped.

Elapsed: 00:00:00.06
11:49:44 SQL> shutdown immediate;
Database closed.
Database dismounted.
ORACLE instance shut down.
11:49:54 SQL> exit
Disconnected from Oracle9i Enterprise Edition Release 9.2.0.6.0 - Production
With the Partitioning, OLAP and Oracle Data Mining options
JServer Release 9.2.0.6.0 - Production

D:\>sqlplus "/ as sysdba"

SQL*Plus: Release 9.2.0.6.0 - Production on Mon Jan 17 11:50:42 2005

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

Connected to an idle instance.

11:50:42 SQL> startup mount;
ORACLE instance started.

Total System Global Area 101785428 bytes
Fixed Size 454484 bytes
Variable Size 75497472 bytes
Database Buffers 25165824 bytes
Redo Buffers 667648 bytes
Database mounted.
11:50:59 SQL> alter session set nls_date_format='yyyy-mm-dd hh24:mi:ss';

Session altered.

Elapsed: 00:00:00.00
11:51:20 SQL> recover database until time '2005-01-17 11:49:35';
Media recovery complete.

--此时恢复到提交一秒之后

11:51:22 SQL> alter database open resetlogs;

Database altered.

Elapsed: 00:00:03.09
11:51:32 SQL> select * from test;

NAME
--------------------
aaaaaaaaaaaaaaaaaaaa
bbbbbbbbbbbbbbbbbbbb
cccccccccccccccccccc

Elapsed: 00:00:00.00

--数据得以恢复

结论:
Oracle能够恢复的时间精度为1秒，但是在Oracle数据库内部，用以产生SCN的时间点有更为精确的精度。
所以，如果你指定秒级恢复，如11:57:28，那么秒后的精度被置00，反而就落在了提交之前。（猜测）
而等待下一秒来到时，这种情况就不会出现了。

renxiao2003

还没有做过这样的恢复，用RMAN会不会被这个更好。

duolanshizhe

恢复的程度  取决于  备份时间+归档的完整性