After a hot backup, where all tablespaces are placed in backup mode seperately; i.e.

alter tablespace user_data start backup;

each datafile will have a different Checkpoint SCN in its header; with the equivalent redo log files which will bring it up to date.

Why doesn't this non-sync of Checkpoint SCN cause a problem when restoring and recovering these datafiles? The "checkpoint" SCN of each datafile is different from other datafiles and probably different from the control file's recorded SCN as well; doesn't this cause a problem?


Doesn't the inconsistency of the SCN in the datafiles cause a problem when SMON checks them once we decide to restore and recover the database using the hot backup?

  • 2
    It doesn't cause a problem because it will still always be possible to restore everything to the same SCN using a combination of logs and datafiles that have been backed up "hot". Obviously it'd be a problem without the redo :)
    – Philᵀᴹ
    Feb 3, 2015 at 15:45
  • 1
    Because a hot backup itself is done in a configuration that is constantly doing backups (via redolog archival). The system that allows to bridge differences between (full) backups, is always up and running. Hence, there's no issue later, when you want to return to any time. Be it a time BEFORE a backup, DURING a backup, or AFTER a backup. The first one though, requires that another backup is done earlier. Just putting your DB in archivelog, not making hot backup, is not sufficient.
    – tvCa
    Feb 3, 2015 at 17:27
  • Thanks for your answers. I am assuming that SMON checks the combination of the datafiles and redo logs for finding the latest SCN then...
    – Stelios
    Feb 3, 2015 at 20:33
  • scn's are written to redo logs continuously - when you commit they are emitted into the redo stream. and semantically speaking, after the checkpoint COMPLETES, not after the checkpoint is INITIATED . asktom.oracle.com/pls/asktom/…
    – Stelios
    Feb 4, 2015 at 7:59

1 Answer 1


When in doubt ask Tom...

backup up and recovery generating extra redo Which is good reason to use RMAN for backups rather than scripted hot backups using begin backup/end backup.

  • There's a reason why OP is on level of tablespaces. If you want a database backup, yes, but if you want to backup specific tablespaces (and, that implies NOT backing up others), it may still be a good idea, no ? Not everybody needs a full backup.
    – tvCa
    Feb 3, 2015 at 18:59
  • RMAN can backup individual tablespaces without using the old start backup/end backup syntax. docs.oracle.com/cd/B19306_01/backup.102/b14192/…
    – Gandolf989
    Feb 3, 2015 at 20:09
  • Thanks for the link. I understand the reason behind the extensive redo log generation and its purpose during a hot backup. what does not click in my mind is how you can restore and recover the database afterwards. I remember reading that SMON checks the control file and the datafiles for the same SCN before opening the database. In the case of a hot backup, then restore and recover; since we will be putting the tablespaces in backup mode in different discrete time periods the datafiles within the tablespaces will not have consistent SCN s nor will the control file.
    – Stelios
    Feb 3, 2015 at 20:27
  • When you restore a database you need to restore the control file first, then mount the database. When you do that the control file knows where all of the data files need to be as well as it has meta data for all of the backups. Ideally you want to up control_file_record_keep_time to keep track of at least 10 days worth of backups if not more. Then when you recover the database knows what RMAN backups files to use and where they should be. Ideally if you can't do a daily full backup you want to do incremental backups. Hopefully you have enterprise edition and can use a block change file.
    – Gandolf989
    Feb 3, 2015 at 20:38
  • As far as I remember when doing this type of backup I had to create also a standby controlfile first. When restoring I used this standby controlfile for redolog aply and media recovery.
    – ibre5041
    Feb 4, 2015 at 10:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.