Re: index node split autonomy

<ctcgag_at_hotmail.com> wrote in message
news:20030527120537.213$RI_at_newsreader.com...
> "Richard Foote" <richard.foote_at_bigpond.com> wrote:
> > >
> > > If an DML statement causes a index block to split, and then
> > > that statement rolls back, is the split also rolled back, or does
> > > it persist?
> > >
> > >
> > Hi Xho,
> >
> > It persists.
> >
> > Much in the same way that a bunch of inserts increments the HWM only to
> > be rolled back. The new HWM persists..
> >
> > The issue to remember here is that Oracle is a multi-user database and
if
> > an index node splits, it's quite plausible that another transaction has
> > inserted an entry into a split node as well. If Oracle could be bothered
> > to redo all the work involved in the split, what of the effects of other
> > transactions ?
>
> Hi Richard,
>
> Thanks you (and the others who responded) for the answer. I thought that
> this must be the case, based on pretty much the same reasoning, but I
> wanted to make sure before wading in farther.
>
> So this must make the undo for indices especially complex, as a failed
> transaction may need to roll data back into blocks other than the ones
> they originally came from, and a transaction making a read-consistent view
> may need to look for undo which came from blocks other than the one it is
> rolling back.
>
> Since the API for extensible indexing doesn't seem to allow this kind of
> intimate interaction with the rollback mechanism, I'm thinking that any
> domain index requiring special node-split logic is going to be very, very
> hard to make "long trousers", as Paul would say.
>
> Regards,
>
> Xho
>

... and Paul hereby offers humble apologies for an oversimplistic and condescending answer, posted in a moment of irritation at something completely different.

Sorry.

Regards,
Paul Received on Tue May 27 2003 - 13:35:23 CDT