Re: Archivelog Mode Recoverability: How Close to Point of Failure?

"Skinny Guy" <skinnychineseguy_at_hotmail.com> wrote in message news:f9b5a61d.0402121603.1730933b_at_posting.google.com...
> Hi,
>
> If I have my DB in archivelog mode and a good backup, how close to the
> point of failure can I recover to?

Define your failure before you start talking about recovery. There's media failure, which can be completely recovered from without the loss of any committed transactions. And there's user error/failure, which cannot. Media failure is what happens when a hard disk blows up. User error is what happens when a user issues a 'delete from emp' command, forgetting to supply a 'where' clause -and then commits instead of rolling back.

Media failures require media recoveries. Media recoveries are complete. User errors require incomplete recoveries (or "point-in-time" recoveries) which, by definition, are incomplete.

> Don't the archive logs get copied only after a log switch and a log
> switch only occurs when a log fills up.

Correct.

> So what if there are commits
> that don't cause the log file to fill up so that those commits haven't
> yet made it to the archive log file. Won't you lose those tranactions
> when you tried to recover from the full backup + the archive logs?

Ah. A slight misunderstanding, I see. No.. a media recovery will apply ALL redo since the time of the backup, whether that redo comes from the archived logs, or the online ones. Therefore, provided your current online log is still safe and sound, you'll get everything back.

If you lose your current online redo log, however, then yes, you have a problem. You can recover only up to the point immediately prior to LGWR switching into that log (loss of redo, be that an archive or an online log, is another reason for having to perform incomplete recoveries). But you won't ever lose your current online redo log, will you? Because like any sensible DBA you will have used the multiplexing features for the online logs which Oracle made available sometime back in version 6 or 7, so that the loss of one log file won't completely knock out an entire log group.

I suppose the fundamental point you're making is that the current online log is Oracle's weak spot because it hasn't been archived yet. There is thus no backup for that file (and there cannot be). Absolutely true. That's why multiplexing the redo logs ought to be compulsory.

Regards
HJR

-- 
--------------------------------------------
Oracle Insights: www.dizwell.com
--------------------------------------------