RE: index hint returns wrong results (retraction of absolute claim!)
Date: Tue, 8 Mar 2016 13:17:29 -0500
Message-ID: <034c01d17966$c85f9f30$591edd90$_at_rsiz.com>
I agree there are circumstances where you have instructed Oracle to ignore errors such that during the period when those errors are allowed Oracle may reasonably deliver an incorrect result. I would not call that a bug since Oracle is behaving according to your special instructions.
It is also not an Oracle bug to destroy the integrity of your data when directed to gloss over anything during update or recovery. Ignoring corrupted rollback segments comes immediately to mind, but there are plenty of other things (some of which require fairly subtle analysis to reveal how it is you've actually directed Oracle to allow an inconsistency.)
However even if your data is corrupt, in lieu of such settings being set in the current system or session, I believe it is incumbent on Oracle to deliver either correct results reflecting the contents of your tables or else toss an error.
EXCEPT, as you point out, if you have explicitly allowed data into the wrong partitions.
My opinion in that case is that Oracle should set a "no pruning" directive on any partition so abused and only ignore that that setting by a session or system exception when delivering results. However that [bug fix | enhancement request] has not been accepted into the product set. I regard that as a bug, but I cannot really argue with Oracle tossing a tuxedo on it.
It is not apparent to me the OP is currently allowing any such errors, ergo my request for a reproducible case from creation that might reveal something. (Which is a superset that includes your data in wrong partition [currently not bug] evolution of the data.)
I do thank you for reminding me and alerting us all to the fact there are several combinations of instructions to Oracle that literally request delivery of questionable result sets. And especially I thank you for reminding me of the data in wrong partition exception currently in force regarding non-errored results on the table viewed as a whole when pruning takes place.
In that context, I stand corrected. It is not apparent to me these apply to the OP, but your precise question may drive out the truth of the situation short of a full re-creation scenario.
Thanks,
mwf
-----Original Message-----
From: Jonathan Lewis [mailto:jonathan_at_jlcomp.demon.co.uk]
Sent: Tuesday, March 08, 2016 12:06 PM
To: Mark W. Farnham; Tom.Terrian.ctr_at_dla.mil; oracle-l_at_freelists.org
Subject: RE: index hint returns wrong results
I'm not sure that I can agree with you, Mark.
What if you get "wrong results" because you set the query_rewrite_integrity
?
It's not quite the same, but you've said to Oracle that you will be
responsible for wrong results.
Similarly what if you set a constraint to RELY disable novalidate and then break it. You volunteered to get wrong results.
If you tell Oracle that you don't want it to validate the data that you've put into the wrong partition why should Oracle be at fault when you get the wrong results ?
(That's assuming my suggested scenario is correct, of course - I'll try to build a model this evening).
Regards
Jonathan Lewis
http://jonathanlewis.wordpress.com
_at_jloracle
From: Mark W. Farnham [mwf_at_rsiz.com]
Sent: 08 March 2016 16:23
To: Tom.Terrian.ctr_at_dla.mil; Jonathan Lewis; oracle-l_at_freelists.org Subject: RE: index hint returns wrong results
First, let's be clear: If Oracle returns incorrect results, then there most definitely is a bug in the stack somewhere. Even if the data has been FUBAR'd due to PEBKAC, it is incumbent on the RDBMS to prevent delivery of a response without tossing an error.
Second, mmm, it seems to me it would be very useful to try to develop a degenerate (meaning no more complicated than required) case from creation and population scripts to wrong results so others can help isolate which versions may be affected and any required "avoid doing this" work-around advice to the community until the bug is fixed.
mwf
-----Original Message-----
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org]
On Behalf Of Terrian, Thomas J CTR (US)
Sent: Tuesday, March 08, 2016 8:51 AM
To: Jonathan Lewis; oracle-l_at_freelists.org
Subject: RE: index hint returns wrong results
That is interesting.
Even when I use the partition name, I still get bad results:
select /*+ index(t1 ind1) */ count(*)
from scott.t1 partition (sys_p14)
where f1 = 'one' and f2 = ' ';
COUNT(*)
2
There is only 1 row in that table where f1 = 'one' and f2 = ' '.
And I see partition pruning in the explain plan when it tries to access the table (not for the global index of course).
-----Original Message-----
From: Jonathan Lewis [mailto:jonathan_at_jlcomp.demon.co.uk]
Sent: Tuesday, March 08, 2016 8:05 AM
To: Terrian, Thomas J CTR (US); oracle-l_at_freelists.org
Subject: [Non-DoD Source] RE: index hint returns wrong results
If the table is partitioned then I think it's possible for the error to appear as a result of procedural error on the part of the application.
I'd have to check but I think you might be able to do an exchange partition including indexes without validation that got data into the wrong partition which wouldn't be seen by a (partition eliminating) tablescan but was seen by a range scan of a global index.
This may be version dependent, of course thanks to enhancements in global index usage.
Regards
Jonathan Lewis
http://jonathanlewis.wordpress.com
_at_jloracle
From: oracle-l-bounce_at_freelists.org [oracle-l-bounce_at_freelists.org] on behalf of Terrian, Thomas J CTR (US) [Tom.Terrian.ctr_at_dla.mil] Sent: 08 March 2016 12:32
To: oracle-l_at_freelists.org
Subject: index hint returns wrong results
Oracle 12c.
Am I missing something or is this a bug? Using an index hint returns the wrong results. Does it matter if the table is partitioned and the index is global or even what fields are in the index? I can't see how any of that should matter. The results should be the same:
select /*+ index(t1 ind1) */ count(*)
from scott.t1
where f1 = 'one' and f2 = ' ';
COUNT(*)
2
select count(*)
from scott.t1
where f1 = 'one' and f2 = ' ';
COUNT(*)
1
--
http://www.freelists.org/webpage/oracle-l
--
http://www.freelists.org/webpage/oracle-l
=
--
http://www.freelists.org/webpage/oracle-l
Received on Tue Mar 08 2016 - 19:17:29 CET