Re: why aren't ORA-1555 errors MORE frequent?

"Richard Kuhler" <noone_at_nowhere.com> wrote in message news:gewC9.15096$2z1.6176523_at_twister.socal.rr.com...
> "Howard J. Rogers" wrote:
>
> <snip>
>
> > > Interesting. You're saying that the cleanouts themselves have to be
> > > written to a RBS? I can't envision any reason for this. Why is this
> > > done? This also seems to imply that a pure SELECT could run out of
> > > rollback, right?
> >
> > You can certainly get 1555s when no-one is doing anything other than
pure
> > selects, true enough. They are incredibly hard to demonstrate, however,
> > being as rare as hen's teeth.
> >
> > Regards
> > HJR
>
> Actually, I was thinking that if the cleanouts have to be written to a
> RBS then the SELECT could actually get "ORA-1650 Unable to extend
> rollback segment" or similar, right?. More significantly, I'm trying to
> figure out why cleanouts would have to be written to a RBS to begin
> with.

Hi again Richard,

Part of the "clean-out" involves updating the header with the SCN of the original transaction (or the SCN as close as Oracle can reproduce it so they we know for the sure that the original transaction was *no later* than the recorded SCN). Note that this overwrites the previously recorded SCN.

From a read consistency point of view this previous SCN is a vital link. If the current SCN is not early enough for a consistent read, we need to determine what was the value of the previous SCN. Hence the need to record this info in the RBS.

Cheers

Richard

>
>
> Thanks,
> Richard Kuhler
>
Received on Wed Nov 20 2002 - 07:33:27 CST