Re: Hot Backup - Need for Rollback Segs

Let me state this simply first for all of those who may be a little confused by the flurry of postings on this matter:

WHEN PERFORMING A HOT BACKUP, IT IS ABSOLUTELY ESSENTIAL TO HAVE A BACKUP OF THE ROLLBACK SEGMENT TABLESPACES. In order to be able to recover from a media failure, your hot backup must have backed of the following files:

• All datafiles including system, data, indexes, and rollback tablespaces
• a backup of the control files created with the "alter database backup controlfile to <file>" statement
• All archive log files generated during and after the hot backup

It is not necessary to backup the online redo log files. The only tablespace that I would feel comfortable not backing up is the temporary tablespace, assuming of course that only temporary segments are written to that tablespace.

Mind you, this is the simple answer and that you can get clever with how and when you backup the different files.

Let's look at the rollback tablespace question. If you loose a rollback tablespace or any one of its datafiles, Oracle must recover it just like any other tablespace. The datafile(s) must be restored and the archive log/online redo log files must be applied in order to get the database back in sync. Without a backup of the rollback segments, you could get the database running again with a couple of undocumented init.ora parameters, but the consistency of the data in your database would be highly questionable.

I have experienced this very problem and did not have a backup of the rollback segments(nor anything else). Fortunately, it wasn't a database that was critical and we forced it open.

Therefore, I would treat rollback tablespaces with the same attention I give the system tablespace.

Matt

In article <DD5CKw.LMM_at_txnews.amd.com>, darryl snedeker <darryl.snedeker_at_amd.com> says:
>
>Larry,
>
>The worst kind of recovery to experience is a time-base recovery. Although not necessary, we
>have been in a position, several times, where we had to create a completely new database
>instance for the purpose of recovery. Under this situation you will need the following:
>-Backup of Control files (Can use current version from production)
>-Datafile backups (most importantly System tablespace datafiles, more frequently if possible)
>-Archive Log file backups.
>
>If you have a device failure, you fix the device, restore the datafiles and apply archive log
>files. If someone drops a table, consider shooting them (joke) and then create a recovery
>instance that consists of the system datafile(s) and the datafiles comprising the tablespace
>where the dropped table once existed. Apply archive log files until you're happy and if
>necessary, OPEN RESETLOGS, export, import, and you're back on the map again. Not much detail, I
>know, but characterizes which backups are necessary.
>
>I cannot find the justification for backing up rollback segment datafiles or Redo log files.
>If anyone else can find reason for backing up more than the three file types above I would like
>to have those reasons explained without having the text transposed.
>
>Thanks.
>
>-D
>
  Received on Fri Aug 11 1995 - 00:00:00 CEST