Re: online backup

jhy <jhy_at_earthling.net> seems to have written:
>I have the same book. The reason the checkpoint SCN is not updated in
>the file header(s) for a tablespace being backed up is because no
>updates have been logged to the file since the backup started.

[deletia]

This has to be wrong.

What would happen when a query attemped to drag data out of the file that had been updated since the hot-backup began? The block buffer can't be guaranteed to hold all changes made during a hot backup, and the db can't go around dragging data out of the online/archived redo logs to satisfy a query.

It's my belief that during a hot backup, updates to the datafile continue as normal - ie, the product of your backup operation is inconsistent to a point in time or SCN, because your backup operation is reading data that's being changed by dbwr on the fly. The datafile itself remains current - to satisfy queries - but your OS copy of it is inconsistent.

The redo logs contain hugely more information than they do for normal operation because instead of containing the _changes_ to be made to each data block, they contain the full image of each block after each change has been applied.

This is because the DB has no guarantee that your OS backup of the datafile has the "before image" of the data to which it might apply a change made during the enable-backup phase. Thus, it needs to replace the whole of each altered block during recovery to attain a clean datafile image, consistent to the very end of the hot backup operation.

Ie, your hot backup is useless without the archived redo logs from the the period of the hot backup. But we knew that already.

Or something like that.

Some or all of the above is probably wrong, but it's late and my brain hurts.

-michael
--
St michael (mainly) Erroneous http://goliath.mersinet.co.uk/~ishamael/  "The Lord's my backup, I shall not DAT.   He maketh me pay cash-in-hand for DLT jukeboxen.   He leadeth me in the Legato Networker restore." Received on Sat Aug 15 1998 - 19:52:53 CDT