Re: Index question
Date: Tue, 25 Aug 2009 22:32:46 +0100
Message-ID: <HLCdnXsNUOzhxQnXnZ2dnUVZ8hudnZ2d_at_bt.com>
<stevedhoward_at_gmail.com> wrote in message news:3923e951-8b83-4ad0-b8c4-1d89f0584828_at_h21g2000yqa.googlegroups.com... On Aug 25, 1:22 pm, joel garry <joel-ga..._at_home.com> wrote:
> As Michel and Jonathan noted, it is probably corruption.
>
> We need to execute a rebuild for this index anyway (we just deleted
> 250 million "increasing sequence type" rows which is making MIN
> queries such as this horrific in terms of performance).
I'd favour an optimizer error over corruption - the plan that got the right answer had to find the bottom left corner of each index and scan from there - so there wasn't any corruption in that area of the index partitions.
If you have time, I'd suggest coalesce rather than rebuild.
On a related topic - if you've got an index that's inconsistent with the table, then a coalesce (or rebuild that scans the index) could still leave you with an inconsistent index.
-- Regards Jonathan Lewis http://jonathanlewis.wordpress.com Author: Cost Based Oracle: Fundamentals http://www.jlcomp.demon.co.uk/cbo_book/ind_book.html The Co-operative Oracle Users' FAQ http://www.jlcomp.demon.co.uk/faq/ind_faq.htmlReceived on Tue Aug 25 2009 - 16:32:46 CDT