Re: shutdown immediate or abort for cold-backup?

"Leonard F Clark" <lfc_at_zoom.co.uk> wrote in message news:3b7e520c.5005337_at_192.168.0.1...
> I'm reluctant to get involved in what looks like becoming a minor
> flame war but I hope my comments may add value.

Well it does liven up a rather wet and miserable Saturday here in the UK.

> I thought it was standard practice (and is certainly advice from
> Oracle) to do the following for batch shutdowns:
>
> 1. Shutdown abort
> 2. Startup normal (probably best in restricted mode)
> 3. Shutdown normal

This is standard advice, but in the hope of starting another argument its rather like the advice about large redo logs limiting recovery time. It is not based on any reasonable argument, but rather a "feeling" that shutdown abort is bad. Now, it may be that shutdown abort is bad, but in that case it doesn't matter whether you do a startup restrict or not. I may have come over a bit strong on this, but it partly reflects my irritation at the unreliability of shutdown immediate.

>
> Discussion:
>
> a) Although I have always been warned clear of shutdown aborts, I have
> never had any subsequent problems. In fact, if you think about it, a
> shutdown abort is no different from a shutdown immediate except that
> the rollback of active processes occurs on recovery (at startup)
> rather than as part of the shutdown process.

As I understand it (from many years ago), this is not true. Shutdown abort does no housekeeping at all, but simply kills all the background processes - that's why it worries people.

> b) by doing the three-step approach, the backup is taken after a
> normal shutdown and, therefore, should give clean results. If a
> problem *were* to result from the abort, by building enough checks
> into the script (return code and check for ORA- errors), you can abort
> the backup and raise warnings immediately.

But you're now in the situation where your database is corrupt and you may have to go back to the previous backup ie your backup routine has caused loss of data and that really isn't acceptable. If this situation is really possible and given that shutdown immediate is very unreliable, it means that a reliable backup must be a hot backup.

> c) the one clear broblem in backing up an aborted shutdown is that, if
> you ever need to recover the database, you will always have to recover
> from the abort (which you have captured on tape). By clearing the
> recovery on step 2, you have a nice clean shutdown to recover from.

Absolutely! Received on Sat Aug 18 2001 - 08:34:41 CDT