Re: Re: oracle can ignore hints

From: <ryan.gaffuri_at_cox.net>
Date: Wed, 10 Mar 2004 8:37:09 -0500
Message-Id: <20040310133709.NEID12895.lakemtao08.cox.net@smtp.central.cox.net>

unless i make typos that i cannot see, I have seen oracle repeatedly ignore my hints lately. I cannot reproduce, since I cannot share our data.
>
> From: Jared.Still_at_radisys.com
> Date: 2004/03/10 Wed AM 08:22:22 EST
> To: oracle-l_at_freelists.org
> Subject: Re: oracle can ignore hints
>
> Correct, so did Oracle 'ignore' the hint?
>
> I guess the point is this: does Oracle ignore your hints,
> or are they merely unusable?
>
> You could also try this with a bitmapped index.
>
> Jared
>
>
>
>
>
>
> Vasu Balla <vballa_at_triniti.com>
> Sent by: oracle-l-bounce_at_freelists.org
> 03/09/2004 10:53 PM
> Please respond to oracle-l
>
>
> To: oracle-l_at_freelists.org
> cc:
> Subject: Re: oracle can ignore hints
>
>
>
> Hi Jared,
>
> i think, a index will not have a row pointer having a null for indexed
> column. Thats the reason the index is ignored in first case, where you are
> checking for null. i may be wrong also
>
> Vasu
>
> Jared.Still_at_radisys.com wrote:
>
> It would be nice to see a reproducible test case of a hint being ignored.
>
> Here's one you can try:
>
> drop table t;
>
> create table t( c1 varchar2(30));
>
> create index tidx on t(c1);
>
> insert into t values(null);
>
> commit;
>
> exec dbms_stats.gather_table_stats(user,'T')
>
> set autotrace on
>
> select /*+ index(t tidx) */
> c1
> from t
> where c1 is null
> /
>
> set autotrace off
>
> update t set c1 = 'DATA';
> commit;
>
> exec dbms_stats.gather_table_stats(user,'T')
> set autotrace on
>
> select /*+ index(t tidx) */
> c1
> from t
> where c1='DATA'
> /
>
> select
> c1
> from t
> where c1='DATA'
> /
>
> set autotrace off
>
>
> You will notice that the hint in the first query is 'ignored'.
>
> If you do a 10053 trace on the first query you will see that the index
> TIDX is not
> even considered for use as an access path. Does this mean that the index
> hint was ignored?
>
> The second query uses the index because the hint tells the CBO to do so.
> The
> third query is a FTS because that's what the CBO would rather do.
>
>
> HTH
>
> Jared
>
> --
> -----------------------------------------------------
> Vasu Balla
> * email : vballa_at_triniti.com
> ( phone :+91 40 27893939 X 1291
> -----------------------------------------------------
> ---------------------------------------------------------------- Please
> see the official ORACLE-L FAQ: http://www.orafaq.com ---------------------------------------------------------------- To
> unsubscribe send email to: oracle-l-request_at_freelists.org put
> 'unsubscribe' in the subject line. -- Archives are at http://www.freelists.org/archives/oracle-l/ FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html -----------------------------------------------------------------
>
>
>

Correct, so did Oracle 'ignore' the hint?

I guess the point is this:  does Oracle ignore your hints,
or are they merely unusable?

You could also try this with a bitmapped index.

Jared

Vasu Balla <vballa@triniti.com> Sent by: oracle-l-bounce@freelists.org  03/09/2004 10:53 PM  Please respond to oracle-l

To:        oracle-l@freelists.org         cc:                 Subject:        Re: oracle can ignore hints

Hi Jared,

i think, a index will not have a row pointer having a null for indexed column. Thats the reason the index is ignored in first case, where you are checking for null. i may be wrong also

Vasu

Jared.Still@radisys.com wrote:

It would be nice to see a reproducible test case of a hint being ignored.

Here's one you can try:

drop table t;

create table t( c1 varchar2(30));

create index tidx on t(c1);

insert into t values(null);

commit;

exec dbms_stats.gather_table_stats(user,'T')

set autotrace on

select /*+ index(t tidx) */
c1
from t
where c1 is null
/

set autotrace off

update t set c1 = 'DATA';
commit;

exec dbms_stats.gather_table_stats(user,'T')
set autotrace on

select /*+ index(t tidx) */
c1
from t
where c1='DATA'
/

select
c1
from t
where c1='DATA'
/

set autotrace off


You will notice that the hint in the first query is 'ignored'.

If you do a 10053 trace on the first query you will see that the index TIDX is not
even considered for use as an access path.  Does this mean that the index
hint was ignored?

The second query uses the index because the hint tells the CBO to do so.  The
third query is a FTS because that's what the CBO would rather do.


HTH

Jared

--
-----------------------------------------------------
Vasu Balla
* email : vballa@triniti.com
( phone :+91 40 27893939 X 1291
-----------------------------------------------------
---------------------------------------------------------------- Please see the official ORACLE-L FAQ: http://www.orafaq.com ---------------------------------------------------------------- To unsubscribe send email to: oracle-l-request@freelists.org put 'unsubscribe' in the subject line. -- Archives are at http://www.freelists.org/archives/oracle-l/ FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html -----------------------------------------------------------------

---

Please see the official ORACLE-L FAQ: http://www.orafaq.com

---

To unsubscribe send email to: oracle-l-request_at_freelists.org put 'unsubscribe' in the subject line.

--
Archives are at http://www.freelists.org/archives/oracle-l/
FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------

Received on Wed Mar 10 2004 - 07:37:20 CST