Re: Shrewd way to avoid 'snapshot too old'?

Shut the hell up Hans. Just shut the hell up. Doing what Jap suggested is true only
in the case he is the ONLY user. In any other case doing that will be like giving a monkey
a gun, theres gonna be blood somewhere. Sybrand is correct! In most of the cases the order by
will decrease performance. And as for the language I prefer what he says rather than the
tight contrived air of mock civility. God , show me something before you lashey out in public Hans.

                                                                Tapan

Hans-Peter Sloot wrote:

> If this is such utter nonsense as you call it, explain why!
> Watch you language a bit otherwise!
>
> There might be something missing in your understanding of
> snapshot too old.
>
> Regards anyway1
>
> Sybrand Bakker wrote:
>
> > Start reading the Oracle Concepts Manual asap please.
> > This is utter nonsense.
> > The only shrewd way to avoid this problem is
> > a) to have rollback segments of sufficient size
> > b) don't commit inside a fetch.
> >
> > Sorry,
> >
> > Sybrand Bakker, Oracle DBA
> >
> > Jaap W. van Dijk <j.w.vandijk_at_kpn.com> schreef in berichtnieuws
> > 01bfbf4e$f99af6a0$2d2b15ac_at_HKTGN0003049994...
> > > If I have a lengthy query on a table (withOUT an FOR UPDATE clause) and at
> > > the same time I update the records I am querying, I sometimes run into a
> > > 'snapshot too old' error.
> > > This is an wellknown situation I believe: records from the query have to
 be
> > > retrieved from the rollbacksegments due to the updates, and at a certain
> > > point the required records are overwritten.
> > >
> > > Can this not be avoided by putting a dummy ORDER BY in the query? Oracle
> > > will read ALL the records, sort them, and, while reading the records from
> > > the temporary tablespace (and NOT from the rollbacksegment) update the
> > > records.
> > >
> > > The records will still be written to the rollbacksegments while updating
 of
> > > course, but the data for the SELECT will come from elsewhere.
> > >
> > > For the sort you could use ROWNUM, so in effect no sorting is
> > > required/performed at all.
> > >
> > > How about that?
> > >
> > > Jaap.
> > >
Received on Thu May 18 2000 - 00:00:00 CDT