Re: Bind Peeking with java application

On Mar 14, 1:30 am, "Cristian Cudizio" <cristian.cudi..._at_yahoo.it> wrote:
> On Mar 13, 7:54 pm, "joel garry" <joel-ga..._at_home.com> wrote:
>
>
>
>
>
> > On Mar 13, 3:57 am, "Cristian Cudizio" <cristian.cudi..._at_yahoo.it>
> > wrote:
>
> > > On Mar 12, 7:55 pm, sybra..._at_gmail.com wrote:
>
> > > > On 12 Mar 2007 10:02:48 -0700, "Cristian Cudizio"
>
> > > > <cristian.cudi..._at_yahoo.it> wrote:
> > > > >So i suspect that on eavy load Oracle during hard parse does not make
> > > > >bind peeking
>
> > > > Any reference in the docs?
> > > > Because obviously this is nonsense.
>
> > > > --
> > > > Sybrand Bakker
> > > > Senior Oracle DBA
>
> > > sorry, i've some problems with posting. I thing you are rigth "this is
> > > nonsense"
> > > there are no reference in the docs but it appens
>
> > Seehttp://jonathanlewis.wordpress.com/2006/12/11/bind-peeking/-you
> > need to run a trace on the problem customers system to find out if it
> > is indeed doing the peek. Are you sure the problem isn't because it
> > _is_ doing the peek? That is, after all, the essence of the bind
> > peeking problem - the first access to the db "correctly" does an FTS,
> > when most of the accesses should not. So you flush, and the
> > subsequent access is ok. It is difficult to reproduce that without
> > knowing what the first access to the customer's database was. You say
> > it is during heavy load - that would perhaps set up the conditions for
> > the sql to be aged out of memory and need to be hard parsed again.
>
> > Metalink not-a-bug 5763227 notes that the 9.2 JDBC driver does not
> > support bind peeking, so going from 9.2 to 10g makes symptoms similar
> > to yours, FWIW.
>
> > There probably is other stuff going on too.
>
> >http://www.freelists.org/archives/oracle-l/05-2004/msg01444.htmlif
> > you are brave.
>
> > jg
> > --
> > @home.com is bogus.http://www.sfdt.com/
>
> The reference to note 5763227 is VERY INTERESTING, i've never
> suspected that bind peeking may
> depend by drivers.
> I've already read Jonathan Lewis post,
> unfortunately not being able to reproduce it is difficult for me
> activate tracing on event 10053 and
> verify what appens (and our application has a connection pool that
> makes impossible to individuate
> what is the connection tha run the query.
> As i've already written we use JDBC OCI 10.2. But now i try to write
> a small java
> program to test with event 10053 and verify what appens. What you say
> is that if there is a bug in
> our software for wich the incorret value is passed as bind value for
> the query that makes correct
> to choose a FTS. To prove that i will use Fine Graned Audit to find if
> is that

Does it really have to be a bug in your app? Could it be your app sometimes just has to pass a value that the optimizer decides is best served by FTS? Also, how is your customer generating histograms? Can you use the 9 driver?

Also google on trcsess.

jg

--
@home.com is bogus.
"You still have the freedom to do things that are stupid in America.
Unfortunately, many people are exercising that freedom with abandon."
- Christopher Cruise