05 November, 2011

Tablespace Recovery in a NOARCHIVELOG database

Following up on my previous blog, I show a Tablespace Restore and Recovery for a NOARCHIVELOG mode database.
This is possible only if all the redo is still available in the online redo logs. !
Do not attempt this if the online redo logs have suffered enough switches that the log that was current during backup and the subsequent transaction(s) has been overwritten.

First, I create a new tablespace :

SQL> connect / as sysdba
Connected.
SQL> archive log list
Database log mode No Archive Mode
Automatic archival Disabled
Archive destination USE_DB_RECOVERY_FILE_DEST
Oldest online log sequence 295
Current log sequence 297
SQL>
SQL> create tablespace TBS_RECOVERY ;

Tablespace created.

SQL> select file_name from dba_data_files where tablespace_name = 'TBS_RECOVERY';

FILE_NAME
--------------------------------------------------------------------------------------------------------------
/home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf

SQL>
SQL> -- startup in MOUNT mode for an RMAN Backup
SQL> shutdown immediate
Database closed.
Database dismounted.
ORACLE instance shut down.
SQL> startup mount
ORACLE instance started.

Total System Global Area 535662592 bytes
Fixed Size 1337720 bytes
Variable Size 213911176 bytes
Database Buffers 314572800 bytes
Redo Buffers 5840896 bytes
Database mounted.
SQL> archive log list
Database log mode No Archive Mode
Automatic archival Disabled
Archive destination USE_DB_RECOVERY_FILE_DEST
Oldest online log sequence 295
Current log sequence 297
SQL> pause Run an RMAN Backup of the tablespace
Run an RMAN Backup of the tablespace

Then I run an RMAN backup :
RMAN> backup tablespace TBS_RECOVERY;

Starting backup at 05-NOV-11
using target database control file instead of recovery catalog
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=20 device type=DISK
allocated channel: ORA_DISK_2
channel ORA_DISK_2: SID=21 device type=DISK
channel ORA_DISK_1: starting full datafile backup set
channel ORA_DISK_1: specifying datafile(s) in backup set
input datafile file number=00018 name=/home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf
channel ORA_DISK_1: starting piece 1 at 05-NOV-11
channel ORA_DISK_1: finished piece 1 at 05-NOV-11
piece handle=/home/oracle/app/oracle/flash_recovery_area/ORCL/backupset/2011_11_05/o1_mf_nnndf_TAG20111105T113924_7c9czg54_.bkp tag=TAG20111105T113924 comment=NONE
channel ORA_DISK_1: backup set complete, elapsed time: 00:00:02
Finished backup at 05-NOV-11

Starting Control File and SPFILE Autobackup at 05-NOV-11
piece handle=/home/oracle/app/oracle/flash_recovery_area/ORCL/autobackup/2011_11_05/o1_mf_s_766409773_7c9czltz_.bkp comment=NONE
Finished Control File and SPFILE Autobackup at 05-NOV-11

RMAN>


Now, I proceed to put some data into the tablespace :
SQL> alter database open;

Database altered.

SQL>
SQL> connect hemant/hemant
Connected.
SQL> -- get current SCN
SQL> select current_scn from v$database;

CURRENT_SCN
-----------
5482811

SQL> select to_char(sysdate,'DD-MON HH24:MI') from dual;

TO_CHAR(SYSDATE,'DD-M
---------------------
05-NOV 11:41

SQL>
SQL> -- run regular insert
SQL> create table INSERT_TBL tablespace TBS_RECOVERY
2 as select * from dba_objects where 1=2;

Table created.

SQL> insert into INSERT_TBL select * from dba_objects;

76684 rows created.

SQL> -- run CTAS
SQL> create table CTAS_TBL tablespace TBS_RECOVERY as select * from dba_objects;

Table created.

SQL>
SQL> select segment_name, segment_type, bytes from user_segments
2 where tablespace_name = 'TBS_RECOVERY';

SEGMENT_NAME SEGMENT_TYPE BYTES
------------------------------ ------------------ ----------
INSERT_TBL TABLE 9437184
CTAS_TBL TABLE 9437184

SQL>
SQL> -- check unrecoverable stamps on the datafile
SQL> select unrecoverable_change#, to_char(unrecoverable_time,'DD-MON HH24:MI')
2 from v$datafile, dba_data_files
3 where file#=file_id
4 and tablespace_name = 'TBS_RECOVERY'
5 /

UNRECOVERABLE_CHANGE# TO_CHAR(UNRECOVERABLE
--------------------- ---------------------
0

SQL>
SQL> -- get current SCN
SQL> select current_scn from v$database;

CURRENT_SCN
-----------
5482969

SQL> select to_char(sysdate,'DD-MON HH24:MI') from dual;

TO_CHAR(SYSDATE,'DD-M
---------------------
05-NOV 11:41

SQL>
SQL> connect / as sysdba
Connected.
SQL> -- set tablespace offline
SQL> alter tablespace TBS_RECOVERY offline;

Tablespace altered.

SQL>
SQL> pause Remove datafile and restore and recover datafile
Remove datafile and restore and recover datafile

I remove the datafile :
[root@localhost ~]# ls -l /home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf
-rw-rw---- 1 oracle oracle 104865792 Nov 5 11:41 /home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf
[root@localhost ~]# rm /home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf
rm: remove regular file `/home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf'? yes
[root@localhost ~]# ls -l /home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf
ls: /home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf: No such file or directory
[root@localhost ~]#

I then RESTORE and RECOVER the tablespace :
RMAN> restore tablespace TBS_RECOVERY;

Starting restore at 05-NOV-11
using target database control file instead of recovery catalog
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=21 device type=DISK
allocated channel: ORA_DISK_2
channel ORA_DISK_2: SID=20 device type=DISK

channel ORA_DISK_1: starting datafile backup set restore
channel ORA_DISK_1: specifying datafile(s) to restore from backup set
channel ORA_DISK_1: restoring datafile 00018 to /home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9crk23_.dbf
channel ORA_DISK_1: reading from backup piece /home/oracle/app/oracle/flash_recovery_area/ORCL/backupset/2011_11_05/o1_mf_nnndf_TAG20111105T113924_7c9czg54_.bkp
channel ORA_DISK_1: piece handle=/home/oracle/app/oracle/flash_recovery_area/ORCL/backupset/2011_11_05/o1_mf_nnndf_TAG20111105T113924_7c9czg54_.bkp tag=TAG20111105T113924
channel ORA_DISK_1: restored backup piece 1
channel ORA_DISK_1: restore complete, elapsed time: 00:00:07
Finished restore at 05-NOV-11

RMAN> recover tablespace TBS_RECOVERY;

Starting recover at 05-NOV-11
using channel ORA_DISK_1
using channel ORA_DISK_2

starting media recovery
media recovery complete, elapsed time: 00:00:04

Finished recover at 05-NOV-11

RMAN> sql 'alter tablespace TBS_RECOVERY online';

sql statement: alter tablespace TBS_RECOVERY online

RMAN>


I now verify the data in the two tables :
SQL>
SQL> -- get current SCN
SQL> select current_scn from v$database;

CURRENT_SCN
-----------
5483260

SQL> select to_char(sysdate,'DD-MON HH24:MI') from dual;

TO_CHAR(SYSDATE,'DD-M
---------------------
05-NOV 11:47

SQL>
SQL> connect hemant/hemant
Connected.
SQL> -- query the table
SQL> select count(*) from CTAS_TBL;
select count(*) from CTAS_TBL
*
ERROR at line 1:
ORA-01578: ORACLE data block corrupted (file # 18, block # 1283)
ORA-01110: data file 18: '/home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9dc5mw_.dbf'
ORA-26040: Data block was loaded using the NOLOGGING option


SQL>
SQL> select count(*) from INSERT_TBL;

COUNT(*)
----------
76684

SQL>


The table that was created as a CTAS operation is not readable as the rows were inserted as a NOLOGGING operation. (See my previous blog post : "CTAS in a NOARCHIVELOG database is a NOLOGGING operation").
However, the table that has data inserted by regular DML, using a normal INSERT has been recovered !
The RECOVER was able to read from the Online Redo Logs.

This case also shows how I did not have to do a Full Database Backup or a Full Database Restore+Recover. The single Tablespace was recovered to a consistent point in time with the rest of the database as evidenced below :

SQL> select current_scn from v$database;

CURRENT_SCN
-----------
5483376

SQL> select to_char(sysdate,'DD-MON HH24:MI') from dual;

TO_CHAR(SYSDATE,'DD-M
---------------------
05-NOV 11:51

SQL> alter system checkpoint;

System altered.

SQL> select file#, last_change#, checkpoint_change#, to_char(checkpoint_time,'DD-MON HH24:MI')
2 from v$datafile order by 1;

FILE# LAST_CHANGE# CHECKPOINT_CHANGE# TO_CHAR(CHECKPOINT_TI
---------- ------------ ------------------ ---------------------
1 5483383 05-NOV 11:51
2 5483383 05-NOV 11:51
3 5483383 05-NOV 11:51
4 5483383 05-NOV 11:51
5 5483383 05-NOV 11:51
6 5483383 05-NOV 11:51
7 5483383 05-NOV 11:51
8 5483383 05-NOV 11:51
9 5483383 05-NOV 11:51
10 5483383 05-NOV 11:51
11 5483383 05-NOV 11:51
12 5483383 05-NOV 11:51
13 5483383 05-NOV 11:51
14 5483383 05-NOV 11:51
15 5483383 05-NOV 11:51
16 5483383 05-NOV 11:51
17 5483383 05-NOV 11:51
18 5483383 05-NOV 11:51

18 rows selected.

SQL>


These are the Log Sequence Numbers (selected lines from the alert.log):
Sat Nov 05 11:22:02 2011
ALTER DATABASE OPEN
Thread 1 opened at log sequence 297
Current log# 3 seq# 297 mem# 0: /home/oracle/app/oracle/oradata/orcl/redo03.log
Successful open of redo thread 1
...
Sat Nov 05 11:22:14 2011
QMNC started with pid=20, OS id=2378
Completed: ALTER DATABASE OPEN
...
Sat Nov 05 11:35:45 2011
create tablespace TBS_RECOVERY
Sat Nov 05 11:35:55 2011
Completed: create tablespace TBS_RECOVERY
...
Sat Nov 05 11:37:12 2011
ALTER DATABASE MOUNT
Successful mount of redo thread 1, with mount id 1294202344
Database mounted in Exclusive Mode
Lost write protection disabled
Completed: ALTER DATABASE MOUNT
...
Sat Nov 05 11:41:13 2011
alter database open
Sat Nov 05 11:41:13 2011
Thread 1 opened at log sequence 297
Current log# 3 seq# 297 mem# 0: /home/oracle/app/oracle/oradata/orcl/redo03.log
...
Sat Nov 05 11:41:59 2011
alter tablespace TBS_RECOVERY offline
Completed: alter tablespace TBS_RECOVERY offline
...
Sat Nov 05 11:45:46 2011
Full restore complete of datafile 18 /home/oracle/app/oracle/oradata/ORCL/datafile/o1_mf_tbs_reco_7c9dc5mw_.dbf. Elapsed time: 0:00:05
checkpoint is 5482691
Sat Nov 05 11:46:04 2011
alter database recover datafile list clear
Completed: alter database recover datafile list clear
alter database recover if needed
tablespace TBS_RECOVERY
Media Recovery Start
Serial Media Recovery started
Recovery of Online Redo Log: Thread 1 Group 3 Seq 297 Reading mem 0
Mem# 0: /home/oracle/app/oracle/oradata/orcl/redo03.log
Media Recovery Complete (orcl)
Completed: alter database recover if needed
tablespace TBS_RECOVERY
Sat Nov 05 11:46:23 2011
alter tablespace TBS_RECOVERY online
Completed: alter tablespace TBS_RECOVERY online


Therefore it is possible to recover a Tablespace in a NOARCHIVELOG database in a very limited frame -- enough redo being available in the Online Redo Logs.
.
.
.

5 comments:

Joel Garry said...

Maybe I'm just not getting it, but this looks to me like it is showing that the ctas is not converted to nologging. It just simply isn't recoverable if you go beyond what is in redo, since you can't dependably do media recovery in noarchivelog mode. Even though the log says you are doing media recovery and you are actually doing media recovery. What you are really doing is simulating a device going offline and online, then crash recovery. This is because db writing is asynchronous, and has little to do with what the db thinks is committed - that is entirely redo, and the datafile is always fuzzy unless shutdown. Copying in a non-fuzzy data file just hides the issue, as though you could write to a data file that has never come up.

It is interesting and trick, though.

word: fupplort

Hemant K Chitale said...

I took the Tablespace OFFLINE. That Checkpoints the file, closes it and leaves it non-fuzzy. I then used RMAN RESTORE -- which does a clean restore (and should raise an error if the target file isn't closed). BTW, that CTAS in NOARCHIVELOG is a NOLOGGING operation has been documented and noted. Once I get those references, I will post them. Also note how the normal table is readable (although I understand that your observations about async, fuzzy and file copy might explain that table ; yet these don't apply because I used OFFLINE and RESTORE).

Hemant K Chitale said...

See the "Followup April 18, 2006 - 2pm Central time zone" entry at http://asktom.oracle.com/pls/asktom/f?p=100:11:3325444209131519::::P11_QUESTION_ID:485221567528

"create table as select - skips redo".
The exceptions are :
1. An Index would generate redo
2. FORCE LOGGING would generate redo even in NOARCHIVELOG mode


In NOARCHIVELOG mode, a CTAS is a Direct Path Insert and that is how it avoids generating redo for the table.

Benjamin Bell said...

Marvelous work.Just wanted to drop a comment and say I am new to your blog and really like what I am reading.Its very helpful for me you did a great work i will keep visiting.

Anonymous said...

Hi,
it is possible to recover a table in a NOARCHIVELOG database? Any explanations would be appreciated