"Howard J. Rogers" <hjr_at_dizwell.com> wrote in message news:<40ade064$0$3036$afc38c87_at_news.optusnet.com.au>...
> "yls177" <yls177_at_hotmail.com> wrote in message
> news:c06e4d68.0405202128.1da062e1_at_posting.google.com...
> > from my understanding after reading the posts, i realized that with
> > until cancel, oracle will recover till the databae crash or have some
> > problems.
>
> No, "recover until cancel" means "keep applying redo until I tell you to
> stop". You pick the point where you want to type the word "cancel", and
> hence it has nothing to do with the point where the database crashes,
> necessarily.
>
> >after this, u can alter database open resetlogs for oracle
> > to start fine but then u will lose some data, right?
>
> If you ever stop applying redo before the point where Oracle would naturally
> stop applying it, then by definition you have not applied some redo that was
> generated by transactions. Therefore, you have not repeated those
> transactions. Therefore, you have lost data.
>
> It' called an "incomplete recovery" for a reason.
>
> > and if u dont use until cancel, oracel will automatically recover by
> > itself.
>
> Well, sort of. You have to type in the words "recover database", so it's
> hardly automatic. But yes, if you don't use the words 'until cancel' or
> 'until time' then Oracle will apply every last drop of redo.
>
> > but then i still dont think that it makes a difference, only that with
> > until cancel, oracle will stop recovering till the time which it
> > crashed.
>
> It's got nothing to do with when the database crashed. Instance crashes
> don't even need any manual recovery commands at all. Instance recoveries
> really are totally and utterly automatic.
>
> > also, i have searched google and from below
>
> I'd suggest http://tahiti.oracle.com. There's a lot about recovery you don't
> seem to be getting. Which is fair enough, but you need to nail this subject.
>
>
> >
> >
> > quote:
> > --------------------------------------------------------------------------
> ------
> >
> > the "UNTIL CANCEL" clause specifies that you're performing an
> > incomplete
> > recovery. If you omit it then Oracle assumes that you want to perform
> > a
> > complete recovery.
> >
> > If you specify a BACKUP CONTROLFILE then Oracle uses the file headers
> > rather
> > than control file information to control the restore. For the record,
> > if you
> > cancel it as you did in scenario 2 (where it failed) you can usually
> > rerun
> > "recover database using backup controlfile until cancel;", enter
> > "cancel"
> > immediately and then "alter database open resetlogs". This should open
> > the
> > database properly.
> >
> > I have to admit that I don't know what difference these clauses make
> > at the
> > instance/control file level which is why I didn't reply at first.
> > I could speculate that using a backup controlfile forces Oracle to
> > search
> > the file headers for the highest SCN and uses that as it's target SCN
> > for
> > the restore.
>
> Precisely so. Spot on. Ordinarily, the control file is our 'master clock'
> and tells us the SCN to which all the datafiles should be synchronised. The
> 'using backup controlfile' clause tells Oracle not to pay any attention to
> what the control file says, because you've flagged it to be well out of date
> before you start.
>
> >I could further theorise that "UNTIL CANCEL" overrides
> > this
> > behaviour and lets you terminate the restore at any point without
> > regard to
> > the SCNs stored in the datafile headers.
>
> They're not mutually exclusive. Until cancel means, as I've said, "apply
> redo (ie, recover) until I tell you to stop". The "using backup controlfile"
> means "and don't trust what the control file tells you". You can do the one
> without the other. It depends on whether you can trust what the control
> file's SCN is (if you haven't restored it from a backup, then yes you can).
>
> > Mind you without going and playing with a recovery this is all
> > conjecture.
> > If anyone can authoritatively confirm or deny this I'd appreciate it
> > as it
> > doesn't seem to be documented anywhere.
>
> Well, that's a bit silly. Of course it's documented. In lots of places. That
> the documentation provided by Oracle might not be terribly understandable is
> a rather different matter.
>
> You first need to get the distinction between complete and incomplete
> recoveries clear in your head.
>
> Then you need to understand that all recoveries, complete and incomplete
> work by knowing what the control file's SCN is.
>
> Then you will understand what a mess you'll be in when the control file's
> SCN is known to be wrong, and why you need a mechanism to tell Oracle 'don't
> trust it'.
>
> And then you'll see that you can do a complete recovery 'using backup
> controlfile' just as well as you can do an incomplete recovery 'using backup
> controlfile'.
>
> Regards
> HJR
thanks HJR...
Received on Fri May 21 2004 - 22:14:43 CDT
