Re: Production problem: Fuzzy standby database

  • From: Howard Latham <howard.latham@xxxxxxxxx>
  • To: Andrey.Kriushin@xxxxxxxx
  • Date: Sun, 19 Jul 2009 10:58:06 +0100

get support involved.

2009/7/18 KRIUSHIN, Andrey <Andrey.Kriushin@xxxxxxxx>

> Hi,
>
> I remember the discussion on fuzzy files recovery back in 2006.
> Here is the URL
> //www.freelists.org/post/oracle-l/Restore-and-Recover-Fuzzy-datafiles
>
> George, I think you are a little bit optimistic. Can not imagine any
> structure in the
> server code (X$-) or in database/flashback logs/block change tracing
> files, were one
> can find any label/mark showing that "all blocks in the files ..." in
> fuzzy mode are pulled
> out of it. The only mark i'm aware of - the end of thread redo.
>
> Which means you've performed complete recovery, and there are no other
> changes
> left to apply. For example, you've copied current online redo "on the
> fly" (with a possibility
> of corruption), or made a shutdown before such a copy... And even
> probaly then started up
> the instance again... The important point is, that such a redo will have
> a last redo block in
> it. The next one would be either zero or will contain older (sequence
> number?) redo
> records.
>
> Charles, you are in danger right now,
>             as you'll fail to activate your standby in usual way,
> unless you have all the redo (including the current one) from the
> production. Even in this
> case you'll need to emulate "manual switchover", i.e. put current redo
> in place,
> recover database (including that current redo) and then recreate the
> controlfile.
>            This is working scenario (at least in 9iR2 and 10g).
>
> There is still some hope to avoid "flood of blood": switchover.
>  Indeed, it applyes ALL of the redo from production to the standby,
> and, hopefully,
> resets the fuzziness sign as well. Though I've not tested this scenario
> personally.
> You should experiment yourself.
>
>
> --Andrey
>
> Johnson, George wrote:
> > If I can ask, how were the file backups taken? If you do simple copy or
> > split mirror type backups and forget to "begin backup", it does this,
> > block SCNs in the files cause the files to go "fuzzy", the answer is to
> > keep on applying redo to until you pull all blocks in the files forward
> > out of fuzzy mode.
> --
> //www.freelists.org/webpage/oracle-l
>
>
>


-- 
Howard A. Latham

Other related posts: