Re: Why control files are needed for restore of hot backup?

From: Sybrand Bakker <postbus_at_sybrandb.demon.nl>
Date: Thu, 23 May 2002 00:03:31 +0200
Message-ID: <ueo6619tpeki37@corp.supernews.com>

"Sean M" <smckeownNO_at_BACKSIESearthlink.net> wrote in message news:3CEC0588.B1F94A00_at_BACKSIESearthlink.net...
> Sybrand Bakker wrote:
> >
> > On 21 May 2002 17:48:29 -0700, danielroy10_at_hotmail.com (Daniel Roy)
> > wrote:
> >
> > >Hi everyone,
> > > I have a question which will probably look obvious to many, but I'm
> > >just starting with Oracle! If I take a hot backup of a database, using
> > >for each tablespace:
> > >- alter tablespace bla-bla begin backup
> > >- I copy the tablespace datafiles to the backup directory
> > >- alter tablespace bla-bla end backup
> > >
> > >I also include in my backup the archive logs generated during the
> > >backup.
> > >Now my question is if the datafile structure of the database doesn't
> > >change, do I really need the control file to also be backed up? It
> > >seems to me that at recovery time, re-creating the control files
> > >should be the same result than using a control file from during the
> > >backup. My feeling is that in both cases, Oracle relies exclusively on
> > >the SCN's of the datafile headers in order to decide which datafile
> > >needs recovery, and which one doesn't.
> > >
> > >Daniel
> >
> > Study the difference between *complete* recovery (all files present)
> > and *incomplete* recovery (backup controlfile) and don't try to be
> > 'smart'. During *complete* recovery Oracle compares the SCNs in the
> > controlfile and the datafiles.

>
>

> I disagree. The act of recreating a controlfile does *not* always
> require an incomplete recovery. You can lose all copies of your
> controlfile, recreate it with the NORESETLOGS clause, and do a fully
> complete recovery (i.e. open the database without the 'resetlogs'
> clause, no loss of committed transactions). You'll need to have all
> archive logs since the last backup and all online redo logs up to the
> point of the media failure, of course.

>

> [Think about it. You have a good backup of your db (hot/cold/whatever,
> as long as it's good). Say you loose a disk that contains all your
> controlfiles (dumb, but let's stick to the hypothetical) and, say, the
> USER_DATA tablespace. Everything else is OK. You restore the USER_DATA
> datafiles from the backup. You startup nomount, issue "create
> controlfile noresetlogs blah blah", then recover the database. Oracle
> applies all archives to USER_DATA since the backup, and then applies all
> online redo logs to USER_DATA and any other datafiles who need to be
> caught up. Next step is 'alter database open' (notice no resetlogs
> clause) and bingo, recovered database, no commited transactions lost, no
> resetlogs issued, and all with a new controlfile.]

>

> Daniel's question is valid. The answer is that, no, you don't
> necessarily *have* to have a current copy of the controlfile to restore
> a backup. But since you multiplex your controlfiles (you do multiplex
> them, don't you?), this would rarely be necessary. Best advice -
> multiplex the controlfile, and make a backup controlfile to trace after
> every structural change to the db. It'll save your bacon.

>

> Regards,
> Sean

Your scenario would still constitute *manual/point in time* recovery as opposed to *complete* automatic recovery. You say there isn't a difference between the two methods of recovery? With all archived redolog files and the controlfile being present, what would happen?
In your scenario you would have at least to press <ENTER> for every individual archive.

If you want to nit pick next time, make sure you're better prepared.

Regards

--
Sybrand Bakker
Senior Oracle DBA

to reply remove '-verwijderdit' from my e-mail address

Received on Wed May 22 2002 - 17:03:31 CDT