Re: What *really* happens at ALTER TBSP END BKP ?

Daniel Nichols <daniel.nichols_at_NOSPAMvirgin.net> wrote in message news:<cl7hgvo3issk7qv3ovn3jeg4kb8r7aljq0_at_4ax.com>...
> Thank you, that's a very helpful illustration.
>
> The site where I work has only recently switched to hot backups and
> the questions I and colleagues have needed answering really don't
> appear in the manuals. For years I'd done ad-hoc hot backups and
> actually had the misconception that the datafiles were untouched
> whilst in hot backup. (Another debate had been whether you needed to
> use OCOPY (on NT) to copy the files out during hot backup. The
> conclusion reached is that as long as your copy method can actually
> copy the file without error then it's okay to use it.)

Good question. I've *never* used ocopy, and I've never had a problem. On the other hand, I don't recall ever actually having to work with NT4 (it was always W2K and upwards).

Not sure if it's relevant: XP has a facility called 'shadow copy' which allows files to be copied when they're in use. Nothing to do with Oracle. But I imagine that was the issue that OCOPY was designed to address. I would have thought that with W2K and above, the matter was entirely moot.

>
> The definitions I used were simply taken from Oracle's glossary -
> "Oracle8i Generic Documentation Master Glossary".
> The 8i one was:
> Checkpoint = A pointer indicating that all changes prior to the SCN
> specified by a redo record have been written to the datafiles by DBWn.
> ...
>
> The 9i Release 2 is using:
> Checkpoint = A data structure that defines an SCN in the redo thread
> of a database. Checkpoints are recorded in the control file and each
> datafile header, and are a crucial element of recovery.
>
> So a checkpoint has changed from being a pointer to being the actual
> data structure. If anything the new definition is woollier since it's
> not telling you what happens at a checkpoint.

Yessir. Still as clear as mud. A checkpoint is a process. Not a pointer, not a structure... but the mere process of DBWR doing its work and then posting CKPT to do its thing as well. (Their secon description allows no scope whatsoever, just as a for example) to distinguish between major and minor checkpoints. Thnk of it as a DBWR write, however, and the distinction is clear: sometimes DBWR writes lots (entire buffer cache at a shutdown, for example) and sometimes not very much (buffers used by a table about to be dropped, for example).

A checkpoint *marker* is simply a bit of data in the redo thread to indicate that, to that point, DBWR has done X amount of work. And that's what read during instance recoveries... though I notice the new definition has nothing much to say about that either. Ho hum.

I agree. It can't be *that* hard to come up with a description of what's involved without it sounding like a physics dissertation. Or without it sounding so vague and half-baked no-one is quite sure what on Earth it means.

~QM Received on Sun Jul 06 2003 - 23:15:22 CDT