Hot Backup Mechanism

Question

Hi all,
I know that Oracle will not update the datafile headers with SCN when they are in backup mode. But, I have a question, if at all the committed transactions will be written to the datafile during backup? I thought that Oracle will hold all committed data in the DBBC (Entries will be made in Redo Logs though) until tablespaces are taken out of backup mode.
Please correct me if I am wrong.
Thanks,
Aswin.

Answer 1

Aswin,
You are ABSOLUTELY wrong. I am not sure how did you pick up the concept like this but this is plain wrong. Have a look here,
E:\Documents and Settings\aristadba>sqlplus / as sysdba
SQL*Plus: Release 10.2.0.1.0 - Production on Mon Jun 8 14:36:44 2009
Copyright (c) 1982, 2005, Oracle.  All rights reserved.
Connected to an idle instance.
SQL> startup mount
ORACLE instance started.
Total System Global Area  167772160 bytes
Fixed Size                  1247900 bytes
Variable Size              75498852 bytes
Database Buffers           88080384 bytes
Redo Buffers                2945024 bytes
Database mounted.
SQL> alter database archivelog;
alter database archivelog
ERROR at line 1:
ORA-00265: instance recovery required, cannot set ARCHIVELOG mode
SQL> alter database open;
Database altered.
SQL> shut imm
SP2-0717: illegal SHUTDOWN option
SQL> shut immediate
Database closed.
Database dismounted.
ORACLE instance shut down.
SQL> startup mount
ORACLE instance started.
Total System Global Area  167772160 bytes
Fixed Size                  1247900 bytes
Variable Size              75498852 bytes
Database Buffers           88080384 bytes
Redo Buffers                2945024 bytes
Database mounted.
SQL> alter database archivelog;
Database altered.
SQL> alter database open;
Database altered.
SQL> select checkpoint_change# from V$datafile;
CHECKPOINT_CHANGE#
           1762853
           1762853
           1762853
           1762853
           1762853
           1762853
6 rows selected.
SQL> select checkpoint_change#, last_change# from V$datafile;
CHECKPOINT_CHANGE# LAST_CHANGE#
           1762853
           1762853
           1762853
           1762853
           1762853
           1762853
6 rows selected.
SQL> select * from V$backup;
     FILE# STATUS                CHANGE# TIME
         1 NOT ACTIVE                  0
         2 NOT ACTIVE                  0
         3 NOT ACTIVE                  0
         4 NOT ACTIVE                  0
         5 NOT ACTIVE                  0
         6 NOT ACTIVE                  0
6 rows selected.
SQL> select name from V$tablespace';
ERROR:
ORA-01756: quoted string not properly terminated
SQL> select name from V$tablespace;
NAME
SYSTEM
UNDOTBS1
SYSAUX
USERS
TEMP
EXAMPLE
SONY
7 rows selected.
SQL> alter tablespace sony begin backup;
Tablespace altered.
SQL> select checkpoint_change#, last_change# from V$datafile;
CHECKPOINT_CHANGE# LAST_CHANGE#
           1762853
           1762853
           1762853
           1762853
           1762853
           1762999
6 rows selected.
SQL> select * from V$backup;
     FILE# STATUS                CHANGE# TIME
         1 NOT ACTIVE                  0
         2 NOT ACTIVE                  0
         3 NOT ACTIVE                  0
         4 NOT ACTIVE                  0
         5 NOT ACTIVE                  0
         6 ACTIVE                1762999 08-JUN-09
6 rows selected.
SQL> create table sony_table ( a number) tablespace sony;
Table created.
SQL> begin
  2  for i in 1..100 loop
  3  insert into sony_table values(100);
  4  end loop;
  5  end;
  6  /
PL/SQL procedure successfully completed.
SQL> commit;
Commit complete.
SQL> alter system flush buffer_cache;
System altered.
SQL> alter tablespace sony end backup;
Tablespace altered.
SQL> select * from sony_table;
         A
       100
       100
       100
       100
       100
       100
       100
       100
       100
....You can see here that the tablespace got a table created into it. In that table, we insserted some data which surely was smaller than the cache itself. So we flushed the buffer cache, making sure that the data of the buffer cache is going to be into the data file. After that we got the tablespace out from the backup and we got our data.
The only thing that does get frozen down in the begin backup mode is that the file headers get freezed at that point when the backup got started. Makes sense in doing so as this point would be remembered by the ocontrol file to identify that when the tablespace would be undergoing recovery, from what point the redo vectors need to get applied. The data does NOT go in to the file header but into the object blocks which has the System Change Number in them. This number gets frozen. So even when your tablespace is in the backup mode, the buffers stilll move as they were moving when the tablespace was in the normal mode. With that. there is no point what so ever to keep the committed in the data buffer cache . Even if you say that your logic is correct, still there can be a situation where your logic would be proved wrong. Buffer cache wont ever be equl to that amount of data in size which is changed by the users. So if we keep the committed data in the cache which is of , say 10meg and you changed 1000 meg, what about this data? Where would we keep it if we wont let it be flished to the data file?
HTH
Aman....