Re: Recovery - online redo log issue (ORA-00338)

Edzard wrote:
>
> Hello,
>
> Still wish to give my opinion on your statement:
> >
> > No, backing up online redo logs is never required for any backup.
> >
> True, if the redo logs, both archived and active, are at any time 100%
> save.

Ah, but I didn't say "archived" redo logs. I specifically said "online" redo logs. Backing up *online* redo logs is never required for any backup - perhaps I should have been more clear and said "any valid, supported backup." Cold backups after a shutdown abort don't count.

> First the case of a hot backup. While taking the backup some
> transaction may occur. This transaction may only be partially
> reflected in the backup. Some changed blocks may have been written to
> disk, others may not and still be in buffers. So when it ever comes to
> restoring this backup, the database needs recovery. You then need the
> redo logs that were active during the cause of the backup, which were
> not backed up to tape. These redo logs may have been archived by the
> time the crash occurs but still they are not on tape. I can not see
> how the database can survive a total-disaster scenario, that is were
> the only thing left is a backup tape. Unless you have some special
> provision, like generating log switches after backing up the
> tablespace and before backing up the archives.

Again, I was only speaking of online redos. Once they've been archived, you should follow normal procedures to avoid a single point of failure - multiplex your archives on disk, then send multiple copies of them to tape, etc. If you need logs for recovery that haven't yet been archived (i.e. still online), then you still have multiple copies of those since you multiplex your online redos (right?).

> In the case of a cold backup the same is true if you happen to shut
> down the database by aborting it. Is that still a cold backup?

No. At least, it's not a valid, supported method of backing up your database cold. That's not to say you can't manage to recover it in many cases, but you're on your own at that point.

> Anyway
> the redo logs are required when starting up after shutdown abort. On
> the Windows NT platform, this may occur unexpectedly. When shutting
> down the database by stopping the service on the NT services control
> panel, it caused a shutdown abort in Oracle8i version 8.1.7.0, no
> matter if the shutdown mode was set to "i" (immediate) or "n" (normal)
> when creating the instance. This also occured when using the oradim
> command. This bug was fixed in the first patch set for 8.1.7 but I
> don't know about earlier versions. The alert*.log will show it.

If NT does that, it's either a bug that's likely already fixed, or just another good reason not to use NT for anything important (or both).

> I'd conclude that a decision to not backup red logs requires
> specialist attention.

I'd conclude that designing a backup and recovery strategy for Oracle requires the attention of a specialist.

Regards,
Sean Received on Fri Apr 19 2002 - 11:10:10 CDT