Re: transaction tables consistent reads - undo records applied

From: joel garry <joel-garry_at_home.com>
Date: Fri, 25 Jun 2010 09:01:38 -0700 (PDT)
Message-ID: <6814438c-9eb4-4d21-9489-136448838082_at_a30g2000yqn.googlegroups.com>

On Jun 24, 7:25 pm, Steve Howard <stevedhow..._at_gmail.com> wrote:
> On Jun 24, 10:20 pm, Steve Howard <stevedhow..._at_gmail.com> wrote:
>
> > Hi All,
>
> > 10.2.0.4 three node cluster EE on SLES 10
>
> ...I should also mention the explain plan for the query is OK.  It is
> a range scan for the time period being retrieved, which may be two or
> three hours, representing 300,000 rows (out of 300 million).
>
> Also, when I look at "table fetch by rowid" for the querying session,
> it will periodically just stall.  It is as if the session is reading
> thousands upon thousands of undo blocks for that next row.  The
> arraysize is 15 (standard SQL*Plus)
>
> Sometimes this happens, sometimes it doesn't.
>
> We did apply patch 7527908

I have no idea, but I speculate you can use Tanel Poders' latchprofx and poke around on his site about in memory undo to figure this out. Does your plan show lots of recursion?

jg

--
_at_home.com is bogus.
http://www.businessweek.com/news/2010-06-25/oracle-rises-after-sun-acquisition-fuels-profit-gain.html

Received on Fri Jun 25 2010 - 11:01:38 CDT