Re: Question about Archivemode

"Howard J. Rogers" wrote:
>
> There is one scenario where, despite not being in archivelog mode, you might
> get away with a complete and perfect recovery: if the continuous stream of
> redo you require still happens to be sitting in the online logs. If you have
> 5 online redo log groups, which switch every half hour, then you have 2.5
> hours of redo available to you. So if you backed up last night, opened the
> database at 9.00am, and the intruder swiped the system tablespace at
> 11.00am, you're doomed. But if he swiped it at 9.45am, the redo required to
> recover the backup of the system datafile is still available in the online
> logs, so you could just restore that one file and do a recovery on it.

Interesting... I'm not sure I agree, though I've never actually tested it, so please take what I'm about to say with a grain of salt. (I'm not talking about your timing thing - I saw your other post re: noon.) I'm talking about the ability to take a datafile from a cold backup of a noarchivelog database, insert it into the mix, and recover using only online redo (assuming, of course, that you haven't cycled through the onlines since the cold backup so you have an unbroken stream of redo available). The problem (at least, I suspect, but I don't have noarchivelog mode databases handy to test) is that although you have all the redo in online logs still available since the backup, Oracle only lets you do instance recovery (as opposed to media recovery) for a noarchivelog mode database. And during instance recovery, Oracle only applies redo since the last checkpoint to the datafile. Now, in your scenario, that datafile has likely been checkpointed more than once since you started up after the cold backup. So even though Oracle hasn't yet overwritten the redo generated since the startup, it's only smart enough to apply the redo since the last checkpoint. So if you replace a file with one who's checkpoint in its header doesn't match what the controlfile thinks it should be, I suspect Oracle would freak out during instance recovery. Oracle would look at the controlfile and say "cool, I need to apply changes since the last checkpoint at 11:30 to this guy". But the datafile's header says "but that's not right - my last checkpoint is actually from 9:00 a.m." Now, maybe if you recreated your controlfile so that the only info Oracle has to go on is the datafile header you could fake it out... hmm... I smell a test coming... Howard, do you have a noarchivelog test database handy?

> Clearly, there's no guarantees, though, that the required redo would be
> available. Only archivelog mode can make that guarantee.

Definitely.

Regards,
Sean Received on Wed Jul 17 2002 - 18:45:24 CDT