Re: how is v$backup populated in mount stage?

> It's obvious that the information about whether something is in hot backup
> mode or not is stored within the data files themselves. If it were
> otherwise, then crashing your instance in the middle of a hot backup would
> not result in 'file 6 needs media recovery' errors, because having noticed
> file 6's checkpoint change number was out of synch, the control file would
> have been able to say 'but don't worry about it, because I know that file is
> in hot backup mode'.
>
> The fact is, you *do* get such error messages, therefore the control file
> cannot be the source of information about whether something is in hot backup
> mode or not.

Proving something (or not) with specific test cases rather than deducing and speculating based on your example above is my preference.

>
> Since no-one these days should be doing O/S-style backups, however, the
> issue is akin to worrying about angelic hosts, terpsichore, and small sharp
> metal objects.

You lost me with those remarks.

My observation is based on some funny things noticed in an RMAN backup and recovery scenario I am looking at.

I have specific test cases that show oracle in a mount state "appears" to read and get this information from data files (since swapping in and out control files does not change what is returned).

Oracle documentation as such tends to infer that only the pfile/spfile and control files are processed in mount stage.

I am beginning to think that based on queries made at mount stage, some parts of oracle software may look beyond the pfile/spfile and control files.

If that's the case, what queries cause these actions under what circumstances seems to be an undocumented feature.

It's somewhat of a puzzle that I thought some people might have solved or have specific knowledge of.

Probably will open a tar eventually if no one can shed any more light on this. Received on Thu Nov 04 2004 - 19:08:45 CST