Re: On-Line Tablespace backup

Jill,

  Until last week, I was one of those who had the same misconception. The thread you are referring to is probably the one that I was involved in last week.

  Here's what I learned during the whole thing. When you issue a "begin backup", this notifies the database to NOT write the SCN to the datafile header (as Sybrand wrote). But the rest of the datafile can be written to. You then use OS commands (or the like) to back up the datafile. This results in an INCONSISTENT datafile backup. Then you issue the "end backup" command. While the tablespace is in backup mode, the changes applied to the datafiles are written to the redo logs in full blocks (not just rows). This is important for recovery. When you recover, you bring back the inconsistent datafile and then can apply changes on a block by block basis. Even though the datafile is inconsistent, you can fully recover since you have the datafile block changes in the redo logs. You are backing up a file that is changing while the redo logs keep notes on the changes. This lets you resolve your inconsistency.

  For further information, refer to the Oracle documentation. I currently have the 8.0.5 version of the documentation installed on my workstation. For that version, refer to Oracle8 Backup and Recovery Guide, Chapter 2 (What are you backing up?), section Types Of Backups.

HTH,
Brian

Sybrand Bakker wrote:
>
> On Sat, 30 Oct 1999 08:18:35 -0700, "Jill" <jc_va_at_hotmail.com> wrote:
>
> >I have been troubled by something I read on this newsgroup earlier. I
> >thought that when you "alter tablespace begin backup", during that period
> >nothing is written to the tablespace, only to the redo logs until "..end
> >backup". Now I am reading that the tablespace is written to while doing an
> >operating system copy, and that full blocks, not just rows, are written to
> >the redo logs. Even our senior DBA in our office had the same misconception
> >as I.
> >
> >How does the operating system take a consistent copy of the tablespace,
> >especially if it is very large? How does it sort if out upon recovery? I
> >know these are broad questions, but if someone could correct my
> >understanding, I'd appreciate it.
> >
> >
> >
> >
> >
> To my understanding what happens during online backup, the System
> Change Number (SCN) is NOT being written to the datafile anymore, it
> is written to the redolog only. This is inline with Oracle writing
> complete blocks. Whenever anything goes wrong, there will be always
> recovery needed, even if you did have a consistent copy of your
> tablespaces. So the redolog files are applied to the database
> regardless. A consistent copy is not really necessary.
>
> Hth,
>
> Sybrand Bakker, Oracle DBA
Received on Mon Nov 01 1999 - 09:36:47 CST