Re: Archive logs

"Joel" <Joel_member_at_newsguy.com> wrote in message news:9gt89l01ajp_at_drn.newsguy.com...
> Our checkpoints/logfiles are changing about every hour during peak times.
 Any
> way we can recover that hour if something blows up?
>

Do you mean you are switching redo logs every hour? Good for you if so, because that's quite a decent rate at which to be switching, and it means you've sized your redo logs appropriately (much faster switches than that, and your overall performance would suffer as all the I/O associated with checkpoints would be clogging up the works).

Now, if your Instance blows up (Junior DBA pulls power plug out of wall socket, for example), then of course you can recover that last hour's work. Just issue the startup command, and let SMON do its stuff.

If a datafile (or entire hard disk containing datafiles) blows up, then yes, you can still get everything back, by simply restoring the relevant datafiles from the last backup, and applying all redo from the time of that backup (from the subject line of your post, I am assuming you are taking archives, and thus can actually supply that entire redo stream from both the archives and the online logs).

If a log file blows up, then you can still get the hour's work back, because that log file, having a status of inactive, has long since been archived, and therefore a perfectly good copy of it already exists.

If the CURRENT log file blows up, you *still* have nothing to worry about, because you have (I hope) already mirrored your redo log groups, and the remaining members of the current group will do perfectly fine for the purposes of recovery.

If ALL the members of the CURRENT redo log blows up, then you are stuffed: you have at that point to perform incomplete recovery, and recover to the end of the previous redo log. Potentially, an hour's worth of transactions have been lost.

If you lose the only copy of an archive log, that's also potential bad news, since if you need to perform recovery, you won't be able to roll forward past the resulting gap in your archive sequence (you'd have to lose both a datafile and an archive log for that to be a problem, though).

So, boiling it all down, the only real dramas start when all members of the current redo log group are lost. 3-way multiplexing of redo logs onto separate physical devices should minimise the risk of that ever happening, but yes, if it ever does happen, you've just discovered Oracle's Achilles' heel.

Regards
HJR Received on Mon Jun 25 2001 - 07:17:47 CDT