Comments on: Upgrade issues

By: Jonathan Lewis

Jonathan Lewis — Mon, 01 Aug 2011 11:08:13 +0000

Randolf,

Thanks for the note. It’s possible that Doug had various patches applied – I think I recall him moaning about the number of patches and fixes going on all the time.

By: Randolf Geist

Randolf Geist — Tue, 26 Jul 2011 18:53:01 +0000

I’ve just run this test on 10.2.0.5 and 11.2.0.2 – strangely 11.2.0.2 with optimizer features set to 10.2.0.5 gives me the new result, whereas the “real” 10.2.0.5 shows me the old-style, 10.2.0.4 result. I’m not sure if the 10.2.0.5 behaviour changes with some PSUs applied to it – nevertheless having no further PSUs or the like applied to both 11.2.0.2 and 10.2.0.5 there seems to be an inconsistency in results.

By: Val Kaviyarasan

Val Kaviyarasan — Mon, 28 Feb 2011 20:14:55 +0000

Thanks for the warning …

Other than doing optimizer_features_enable, is there an expected patch from oracle to fix this issue?

By: Philosophy – 13 « Oracle Scratchpad

Philosophy – 13 « Oracle Scratchpad — Thu, 10 Feb 2011 14:58:42 +0000

[...] Of course, being Oracle, you may find that someone, somewhere, has exactly such a histogram that appears to work brilliantly for them – but that will be because the optimizer has messed up the arithmetic so much that they are getting a great execution plan for completely the wrong reason … so they need to watch out for the next upgrade or patch release in case the optimizer gets enhanced. [...]

By: oraculix

oraculix — Fri, 04 Feb 2011 14:47:43 +0000

Thank you for the early warning, Jonathan! I’m on the verge of migrating a DB with sub-partitioning to 11.2.0.2, and this note might just save my day.
Kind regards,
Uwe

By: Mark W. Farnham

Mark W. Farnham — Fri, 04 Feb 2011 11:19:35 +0000

Nicely done. One additional solution (with all the usual caveats about lying to the CBO to fix one problem may cause other, potentially worse and more numerous problems) is to set the blevel to match (probably setting the one column index UP rather than making the PK look excessively cheap, since if you’re that close on the cost estimate on a plan using the single column index versus a table scan it probably won’t hurt too much to do a table scan, and the PK winning over the single column index when it shouldn’t probably won’t hurt too much.)

Oracle publishing changes in factor weightings and formulas used to generate the estimated cost from release to release and in patches sure would be a nice thing to do.

mwf

By: Doug Burns

Doug Burns — Thu, 03 Feb 2011 23:57:51 +0000

it helps to emulate Doug’s problem

I think it will take a much better test case than this to cover the full range of my problems ;-)

But thanks a lot for looking at this and the detailed explanation.