Header datafile and SCN

Question

Hi every body,
The controlfile contains the current SCN and the checkpoint SCN. that is right ?
In my search, I found that every datafile online and in read write mode, contain in his header the scn checkpoint.
But, I think that it contains only the last scn applied to this datafile. that is right?
Without this last scn in the datafile, how oracle know when he is stopped writing to the datafile in case of failure, to recover it ?
Thank in advance for your answers.

Answer 1

user622061 wrote:
Thank you a lot for the links.
I did a test :
BEFORE TEST
v$database :
CHECKPOINT_CHANGE# CURRENT_SCN
13114719 13117108
v$datafile :
NAME CHECKPOINT_CHANGE# LAST_CHANGE#
C:\ORACLE\ORADATA\BDD10G\TBS1_01.DBF 13117132
v$datafile_header :
NAME CHECKPOINT_CHANGE# LAST_CHANGE#
C:\ORACLE\ORADATA\BDD10G\TBS1_01.DBF 13117132
v$logfile and v$log :
MEMBER STATUS
C:\ORACLE\ORADATA\BDD10G\REDO01.LOG INACTIVE
C:\ORACLE\ORADATA\BDD10G\REDO02.LOG CURRENT
C:\ORACLE\ORADATA\BDD10G\REDO05.LOG UNUSED
TEST
create table toto ( a varchar2(30)) tablespace tbs1 ;
insert into table toto values ('SEBASTIAN');
commit;
AFTER TEST
v$datafile :
NAME CHECKPOINT_CHANGE# LAST_CHANGE#
C:\ORACLE\ORADATA\BDD10G\TBS1_01.DBF 13117132
v$datafile_header :
NAME CHECKPOINT_CHANGE# LAST_CHANGE#
C:\ORACLE\ORADATA\BDD10G\TBS1_01.DBF 13117132
the checkpoint_change# for this datafile is the same, before and after insert and commit a data in this tablespace.That's correct! The reason for this is that Oracle won't go and ask DBWR to write into the datafile with each and every commit of yours. You have ended a transaction, that's all! Now, the results of that transaction is updated already by LGWR into the redo log file. The checkpoint number(not SCN) is not updated by DBWR since this would happen in the case it would do a full Checkpoint.
To inspect the contents of this datafile, I use some OS commands : I search the word "SEBASTIEN" !
The first time, I didn't find it.
few minutes later, I found the word "SEBASTIEN" --> So my datafile is updated by the insert I did.Correct again! DBWR would write when it would think it's good to write. So not at the time of commit but after some time, DBWR updated the contents into the datafile.
But, the problem is that his checkpoint_change# still the same = 13117132As I have explained , it won't change, not so quickly.
>
>
You can reproduce the test !
My problem is to understand how Oracle ( DBWR) can update a datafile with data related to the SCN newer than the SCN registered in the header of datafile ?DBWR would have the list of the dirty buffer's list with it which it would periodically. If there is a commit and the data is updated in the logfile, the contents logged into the redo log file would be checkpointed in thedatafile after some time. This commit event, stored in the log file, is also recorded inthe control file. That's how Oracle knows, in the event of a crash, that to what extent they have to do the recovery since the stop checkpoint is already noted in the controlfile.
If Oracle did it, how can he know the last SCN on every datafile.It doesn't need to. Control file has it.
HTH
Aman....