Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Important: wrong Oracle results in parallel query

Re: Important: wrong Oracle results in parallel query

From: Telemachus <telemachus_at_ulysseswillreturn.net>
Date: Thu, 18 Jul 2002 10:41:43 +0100
Message-ID: <s3wZ8.3074$zX3.2567@news.indigo.ie>


I can give you a shorter sample of a more invidious bug that won't be fixed till forever
(I've just checked and it's in 9.2.0 ) and had the same feelings last year

but do you have a workaround ?

If the worst comes to the worst perhaps you might code your own sort routine....
"Martin Haltmayer" <Martin.Haltmayer_at_d2mail.de> wrote in message news:3D367BBC.2438D96_at_d2mail.de...
> This does not have to do anything with the real sort order of "german"
because
> if I have the same table twice it has to order it twice the same way no
matter
> how wrong it is (otherwise ordering would not be deterministic).
>
> Thomas Kyte (thank you very much) and our Oracle support informed me that
it was
> bug 1231199. I looked it up in metalink and found that this bug has been
known
> since March 2000! I find it very gross that it is not yet fixed in 9.2! It
was
> announced for fixing in version 10.0 (whenever that will be). I am really
angry
> about this behaviour of Oracle corporation. If any craftsman would deliver
goods
> that s/he knows are faulty you could sue her/him for fraud. Not so in IT.
They
> stuff all lots of new features in their releases instead of fixing old
ones.
> That maybe the real meaning of n.e.w. (not even working).
>
> Martin
>
>
>
> Jonathan Lewis wrote:
> >
> > I have to protect myself with a claim of ignorance
> > of German, but doesn't it have some option that a
> > double S can be expressed as a single Beta ?
> >
> > If so, you could get SS sorting identically to B
> > and disappearing on a German NLS sort, but
> > not disappearing on a binary NLS sort.
> >
> > The details may be wrong - but I think the
> > argument could be appropriate.
> >
> > --
> > Jonathan Lewis
> > http://www.jlcomp.demon.co.uk
> >
> > Next Seminars
> > UK July / Sept
> > Australia August
> > Malaysia September
> > USA (MI) November
> >
> > http://www.jlcomp.demon.co.uk/seminar.html
> >
> > Martin Haltmayer wrote in message <3D357CDA.64EA27E_at_d2mail.de>...
> > >Hi all,
> > >
> > >we did the following on Sun Solaris 64bit 9.2.0.1.0, Sun Solaris 32bit
> > >8.1.7.4.0, Sun Solaris 32bit 8.1.6.3.0, Windows2000 8.1.7.2.1:
> > >
> > >drop table t_xxx;
> > >drop table t_yyy;
> > >
> > >create table t_xxx as select rpad (object_name, 30, ' ') as object_name
> > from
> > >all_objects;
> > >create table t_yyy as select rpad (object_name, 30, ' ') as object_name
> > from
> > >all_objects;
> > >
> > >alter table t_yyy modify object_name char (30);
> > >
> > >describe t_xxx
> > >describe t_yyy
> > >
> > >alter table t_xxx parallel (degree 4);
> > >alter table t_yyy parallel (degree 4);
> > >
> > >alter session set nls_sort = german;
> > >
> > >select object_name from t_xxx
> > >minus
> > >select object_name from t_yyy
> > >/
> > >
> > >alter session set nls_sort = binary;
> > >
> > >select object_name from t_xxx
> > >minus
> > >select object_name from t_yyy
> > >/
> > >
> > >The last two selects yield different results. Under any circumstances I
> > would
> > >call this an Oracle bug. Please note that the rpad usage ensures that
the
> > data
> > >are *really* the same. Please also note that even a different sort
> > mechanism
> > >does not explain this because the set minus operation would be
consistent
> > in its
> > >semantics!
> > >
> > >Bottom line: when doing parallel operations make sure you have nls_sort
> > >explicitly set to binary. Attention: it will be set to e. g. german in
case
> > you
> > >set nls_language!
> > >
> > >Any comments?
> > >
> > >Martin
Received on Thu Jul 18 2002 - 04:41:43 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US