Re: Help with hot backups

"Howard J. Rogers" <hjr_at_dizwell.com> wrote in message news:<40a96c41$0$1586$afc38c87_at_news.optusnet.com.au>...
> Joel Garry wrote:
>
> >>Really appreciate any assistance anyone could offer on this.
> >
> >
> > Metalink also has a number of discussions and notes, see
> > http://metalink.oracle.com/metalink/plsql/ml2_documents.showDocument?p_database_id=NOT&p_id=155656.1
> > among others.
> >
> > There are bugs/misfeatures in the way this works in 8i, you need to
> > keep some range of logs before and after the backup anyways.
>
> Well, we rather hope that you always keep the logs produced after a
> backup. It's not much of a backup, otherwise! And that's regardless of
> whether you're using RMAN or doing it with good old shell scripts.
>
> Likewise, no-one sensible deletes all logs from before a
> freshly-performed backup. It's been discussed here many times, as you know.

Of course, _I_ know, but the OP may not, and there is that delete input command to confuse the issue... and besides that, it just doesn't work the same every time as far as whether it keeps enough logs for a self-contained backup, at least when creating standby db's IME.
>
> > You
> > definitely want to do the one after the db backup to be sure
> > checkpoint_change# in v$database is updated properly.
>
> Absolutely.
>
> > An additional
> > one before couldn't hurt,
>
> Yes it could. It involves a log switch and hence a checkpoint. Don't do
> things on the grounds they 'can't hurt' unless you know with absolute
> certainty that they can't. And this isn't one of those things.
>
> > and some people recommend it, since it waits
> > for the archiving to occur (as opposed to the alter system switch
> > logfile command, which doesn't wait around before returning control).
>
>
> But as I said before, if he puts an 'archive log current' *after* the
> data file backup and *before* the archivelog backup, RMAN will wait anyway.
>
> There is nothing that putting in an early archive log current does for
> you that is not exactly achieved by doing it the way our original poster
> already had it (ie, after the data file backup). Which happens,
> incidentally, to be exactly what the Metalink paper you quoted says one
> should do, prefacing their advice with the words "The backup order is
> very important".

As usual, you are correct, precise and informative.

>
> However, throwing in an extra, early bit of archiving definitely causes
> ARCn to do more work, causes a log switch, causes a checkpoint, and
> therefore has a direct hit on the database. It isn't a free gift, and it
> shouldn't therefore be recommended.

The general situations I've dealt with have been little or no usage of the db during the RMAN backup. So I'm not familiar with the situation of some load on the db, but can't help but wonder if there are situations where you would want the hit to occur at a known time, rather than saving everything towards the end.

jg

--
@home.com is bogus.
You _will_ cooperate:
http://www.signonsandiego.com/uniontrib/20040518/news_1b18lucent.html