MAA recommends to enable Flashback Database on the primary prior to FailOver:
Enable Flashback Database to reinstate failed production databases after a
failover operation has completed. Flashback Database provides a second
very significant function, enabling fast point in time recovery if needed.
See: MAA_WP_10gR2_SwitchoverFailoverBestPractices.pdf
Enabling Flashback Database involves setting up flash recovery area and setting a flashback retention target which specifies how far back you want to be able to restore your database using the Flashback Database feature.
Once Flashback Database is setup, the database starts to copy images of each altered block into the flashback logs — this works for all datafiles. When it’s time to Flashback the database, the copies of the blocks from flashback logs are used to reconstruct the datafiles to a state just prior to the desired flashback time, the redo/arch logs are then used to bring a datafiles to a consistent state.
WARNING::
Redo logs must be available for the entire time period spanned by the
flashback logs, whether on tape or on disk. (In practice, however, redo
logs are generally needed much longer than the flashback retention target
to support point-in-time recovery.)There are also a number of operations you can perform on your database, such
as dropping a tablespace or shrinking a datafile, which cannot be reversed
with Flashback Database. After such an operation, the flashback database
window begins at the time immediately following that operation.
One thing to consider is that the only way to guarantee a database can be returned to a specific point in time is to use guaranteed restore points. In other words, don’t use “normal” restore points for this purpose and don’t rely on Flashback Database alone. The only constraint to how far you can go back, is the size of your disk space in the flash recovery area.
Again, WARNING:
Limitations that apply to Flashback Database also apply to guaranteed
restore points. For example, shrinking a datafile or dropping a tablespace
can prevent flashing back the affected datafiles to the guaranteed restore
point.
Creating a guaranteed restore point without having enough sufficient free space in the flash recovery area (FRA) will cause the FRA to fill completely, that’s because “No file in the flash recovery area is eligible for deletion if it is required to satisfy the guarantee [restore point]“. In many circumstances, this will cause your database to halt.
To save on space in FRA you can disable the Flashback Database and still create a guaranteed restore point. In this case, the first time a datafile block is modified, an image of this block before the modification is stored in the flashback logs. This saves on space because only one-time copy of every changed data block is stored there, but subsequent modifications to the same block do not cause the block contents to be logged again. This method works really well and it’s more efficient as long as your primary only needs to be able to return to the specific point in time at which the guaranteed restore point was created, such as to a before-state of a failed application upgrade that might have made changes to a database.
See: 5.1.1 About Flashback Database
That’s theory, how about some practice time? Follow along. First, we verify our setup (in this case no FRA [db_recovery_file_dest] is setup:
rac1.XRACP1-> sqlplus /nolog
SQL*Plus: Release 10.2.0.4.0 - Production on Fri Jun 12 19:25:07 2009
Copyright (c) 1982, 2007, Oracle. All Rights Reserved.
SQL> connect / as sysdba
Connected.
SQL> select * from v$recovery_file_dest;
no rows selected
SQL> show parameter recovery
NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
db_recovery_file_dest string
db_recovery_file_dest_size big integer 0
recovery_parallelism integer 0
SQL> select * from v$flash_recovery_area_usage;
no rows selected
SQL>
Lets create FRA (NOTE: since this is a RAC database, I am creating FRA on a clustered FS (OCFS2) /u02):
[root@rac1 log]# mkdir -p /u02/oradata/rcv_area
[root@rac1 log]# chown -R oracle:dba /u02/oradata
rac1.XRACP1-> df -k /u02
Filesystem 1K-blocks Used Available Use% Mounted on
/dev/mapper/2000b080023002235p1
37899136 1971200 35927936 6% /u02
rac1.XRACP1->
SQL> alter system set db_recovery_file_dest_size=28g scope=both sid='*';
System altered.
SQL> alter system set db_recovery_file_dest='/u02/oradata/rcv_area' scope=both sid='*';
System altered.
SQL>
SQL> set lines 132
SQL> col name format a35
SQL> set trims on
SQL> select * from v$recovery_file_dest;
NAME SPACE_LIMIT SPACE_USED SPACE_RECLAIMABLE NUMBER_OF_FILES
----------------------------------- ----------- ---------- ----------------- ---------------
/u02/oradata/rcv_area 3.0065E+10 0 0 0
SQL> select * from v$flash_recovery_area_usage;
FILE_TYPE PERCENT_SPACE_USED PERCENT_SPACE_RECLAIMABLE NUMBER_OF_FILES
------------ ------------------ ------------------------- ---------------
CONTROLFILE 0 0 0
ONLINELOG 0 0 0
ARCHIVELOG 0 0 0
BACKUPPIECE 0 0 0
IMAGECOPY 0 0 0
FLASHBACKLOG 0 0 0
6 rows selected.
SQL>
So far we setup the FRA, now lets try to create guaranteed restore point. One thing to remember here though is this — if flashback database is not enabled (to save on space), then the database must be mounted, not open, when creating the first guaranteed restore point (or if all previously created guaranteed restore points have been dropped). If you attempt to create a guaranteed restore point when the database is opened you get ORA-38787 error:
SQL> CREATE RESTORE POINT gr_01 GUARANTEE FLASHBACK DATABASE;
CREATE RESTORE POINT gr_01 GUARANTEE FLASHBACK DATABASE
*
ERROR at line 1:
ORA-38784: Cannot create restore point 'GR_01'.
ORA-38787: Creating the first guaranteed restore point requires mount mode when flashback database is off.
A normal restore point however, works, but it will not do anything since flashback database is off, here’s an example:
SQL> !ls -l /u02/oradata/rcv_area
total 0
SQL> CREATE RESTORE POINT nr_01;
Restore point created.
SQL> !ls -l /u02/oradata/rcv_area
total 0
SQL> create table xyz(t number) tablespace tools;
Table created.
SQL> !ls -l /u02/oradata/rcv_area
total 0
SQL> insert into xyz values (1000);
1 row created.
SQL> commit;
Commit complete.
SQL> !ls -l /u02/oradata/rcv_area
total 0
SQL>
As you can see nothing happened … even though a normal restore point is created. Lets drop it (this syntax works for both NORMAL and GUARANTEED RPs):
SQL> drop restore point nr_01;
Restore point dropped.
SQL> SELECT NAME, SCN, TIME, DATABASE_INCARNATION#,
GUARANTEE_FLASHBACK_DATABASE,STORAGE_SIZE
FROM V$RESTORE_POINT;
2 3
no rows selected
Now lets create GUARANTEED restore point WITHOUT enabling FLASHBACK DATABASE which requires database to be mounted only, and since this is RAC database all other instances must be shutdown first as well:
rac1.XRACP1-> srvctl stop database -d XRACP
rac1.XRACP1-> sqlplus /nolog
SQL*Plus: Release 10.2.0.4.0 - Production on Fri Jun 12 20:01:53 2009
Copyright (c) 1982, 2007, Oracle. All Rights Reserved.
SQL> connect / as sysdba
Connected to an idle instance.
SQL> startup mount;
ORACLE instance started.
Total System Global Area 1610612736 bytes
Fixed Size 2084296 bytes
Variable Size 822084152 bytes
Database Buffers 771751936 bytes
Redo Buffers 14692352 bytes
Database mounted.
SQL> CREATE RESTORE POINT gr_01 GUARANTEE FLASHBACK DATABASE;
Restore point created.
SQL> shutdown immediate;
ORA-01109: database not open
Database dismounted.
ORACLE instance shut down.
SQL> exit
Disconnected from Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - 64bit Production
With the Partitioning, Real Application Clusters, OLAP, Data Mining
and Real Application Testing options
rac1.XRACP1-> srvctl start database -d XRACP
rac1.XRACP1-> srvctl start service -d XRACP
rac1.XRACP1->
Now lets see what we can gather from V$RESTORE_POINT:
SQL> col name format a15
col time format a35
set lines 132
set trims on
SQL> SQL> SQL> SQL>
SQL>
SQL>
SQL>
SQL> SELECT NAME, SCN, TIME, DATABASE_INCARNATION#,
GUARANTEE_FLASHBACK_DATABASE,STORAGE_SIZE
FROM V$RESTORE_POINT;
2 3
NAME SCN TIME DATABASE_INCARNATION# GUA STORAGE_SIZE
--------------- ---------- ----------------------------------- --------------------- --- ------------
GR_01 9.0317E+12 12-JUN-09 08.02.19.000000000 PM 1 YES 63766528
SQL>
rac1.XRACP1-> find /u02/oradata/rcv_area -ls
5161314 4 drwxr-xr-x 3 oracle dba 4096 Jun 12 20:02 /u02/oradata/rcv_area
5161316 4 drwxr-x--- 3 oracle oinstall 4096 Jun 12 20:02 /u02/oradata/rcv_area/XRACP
5161317 4 drwxr-x--- 2 oracle oinstall 4096 Jun 12 20:03 /u02/oradata/rcv_area/XRACP/flashback
5161318 15576 -rw-r----- 1 oracle oinstall 15949824 Jun 12 20:04 /u02/oradata/rcv_area/XRACP/flashback/o1_mf_5365ov16_.flb
237846 15576 -rw-rw---- 1 oracle oinstall 15949824 Jun 12 20:03 /u02/oradata/rcv_area/XRACP/flashback/o1_mf_5365qm48_.flb
4128929 15576 -rw-rw---- 1 oracle oinstall 15949824 Jun 12 20:03 /u02/oradata/rcv_area/XRACP/flashback/o1_mf_5365qtxy_.flb
7227425 15576 -rw-rw---- 1 oracle oinstall 15949824 Jun 12 20:03 /u02/oradata/rcv_area/XRACP/flashback/o1_mf_5365qtyd_.flb
rac1.XRACP1->
Much better now — it is working, and a few days later:
SQL> !date;ls -l /u02/oradata/rcv_area/XRACP/flashback
Mon Jun 15 15:02:17 PDT 2009
total 358248
-rw-r----- 1 oracle oinstall 15949824 Jun 13 04:00 o1_mf_5365ov16_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 12 20:03 o1_mf_5365qm48_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 12 20:03 o1_mf_5365qtxy_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 12 20:03 o1_mf_5365qtyd_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 12 22:00 o1_mf_536dm0jm_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 12 22:01 o1_mf_536dmnxj_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 13 02:00 o1_mf_536dntxs_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 12 23:03 o1_mf_536j9b8b_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 12 23:28 o1_mf_536j9jvj_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 13 06:00 o1_mf_536kqnmf_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 13 15:10 o1_mf_536tnpky_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 13 21:00 o1_mf_5371or32_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 13 23:03 o1_mf_5378qrx2_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 14 12:00 o1_mf_537m8tox_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 14 07:33 o1_mf_5388yvsd_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 14 19:00 o1_mf_538xgm1v_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 14 02:00 o1_mf_5394pdx7_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 14 23:04 o1_mf_539h0tcs_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 15 04:00 o1_mf_53b2lorr_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 15 15:00 o1_mf_53bl5v8y_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 15 15:00 o1_mf_53cbs5hs_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 15 15:00 o1_mf_53cs3y00_.flb
-rw-rw---- 1 oracle oinstall 15949824 Jun 15 15:00 o1_mf_53dbfmx5_.flb
SQL> SELECT NAME, SCN, TIME, DATABASE_INCARNATION#,
GUARANTEE_FLASHBACK_DATABASE,STORAGE_SIZE
FROM V$RESTORE_POINT; 2 3
NAME SCN TIME DATABASE_INCARNATION# GUA STORAGE_SIZE
--------------- ---------- ----------------------------------- --------------------- --- ------------
GR_01 9.0317E+12 12-JUN-09 08.02.19.000000000 PM 1 YES 366657536
SQL> select * from v$flash_recovery_area_usage;
FILE_TYPE PERCENT_SPACE_USED PERCENT_SPACE_RECLAIMABLE NUMBER_OF_FILES
------------ ------------------ ------------------------- ---------------
CONTROLFILE 0 0 0
ONLINELOG 0 0 0
ARCHIVELOG 0 0 0
BACKUPPIECE 0 0 0
IMAGECOPY 0 0 0
FLASHBACKLOG 1.22 0 23
6 rows selected.
SQL>
And that’s how you create a guaranteed restore point. But what do you use it for, with regards to STANDBY FAILOVER? You really shouldn’t, because you don’t know when a FAILOVER occurs so you can’t create a point right before it. You should however, use a guaranteed restore point during a SWITCHOVER because during a switchover you are in full control of both primary and standby databases. In a FAILOVER scenario, enabling Flashback Database [ ALTER DATABASE FLASHBACK ON; ] is more effective than trying to create guaranteed restore points.
Still have questions? Here are pointers to additional reading with excerpts:
From Note:565535.1 titled “Flashback Database Best Practices & Performance”:
Manual primary database reinstate- if Data Guard is being used without the
fast-start failover feature and a Data Guard failover is necessary, then
flashback database can be used to manually reinstate the failed primary
database. This is documented in the Data Guard Administration and Concepts
guide – see “12.4 Using Flashback Database After a Failover” here:
And from 12.4 Using Flashback Database After a Failover:
After a failover occurs, the original primary database can no longer
participate in the Data Guard configuration until it is repaired and
established as a standby database in the new configuration. To do
this, you can use the Flashback Database feature to recover the
failed primary database to a point in time before the failover
occurred, and then convert it into a physical or logical standby
database in the new configuration.
Vitaliy Mogilevskiy August 22, 2009
Posted In: Data Guard, RAC
Tags: Failover, Flashback Database, FRA, guaranteed restore point, MAA, Switchover