Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: Anyone dealt with address not mapped to object errors?
Kai,
To the best of my knowledge this parameter is new as of 8.1.6. The default
setting for it is cursor_sharing=exact
When this parameter is changed to cursor_sharing=FORCE, then Oracle
substitues all literal values with system generated bind variables. When
set to exact, no action is taken. This is extremely useful in situations
where the code running in the DB does not use bind variables and as a result
the library cache hit ratio (in particular the sql area) would be very low.
Since Oracle has to hard parse more statements when bind variables are not
used, this creates unnecessary pressure on the CPU(s).
The problem with cursor_sharing=FORCE, in my experience, is that Oracle doesn't always find the literal value of the bind variable at execution time. The result is a core dump and trace file with a 600 and 7445 errors. The session that was passing the sql is disconnected and from what I can see, is not allowed to reconnect. Naturally, most applications can't deal with this.
On my system, we get in excess of 10 million sql statements per day in the DB and get about 50-100 core dumps per day with cursor_sharing=FORCE, and none with it set to EXACT.
I have requested that Oracle list this as a bug, but as of yet they haven't done so.
In your case, this doesn't apply as I am fairly certain 8.1.5 doesn't have this parameter. If you are getting addressed not mapped to object errors, this *probably* indicates memory corruption. The source could be either Oracle screwing up or the OS or physical problems with the memory. I would suggest you examine the hardware to see if it is reporting any faults(/var/adm/messages?), if not then look at Oracle. Try setting all init.ora parameters to their default (insofar as you can) and see if the problem persists. If it does, then ensure your OS has all the latest patches or at least meets Oracles requirements as defined in the installation guide. I am pretty sure you need at least the jumbo pack if you are on Solaris 2.6.
-- Dave A "Kai Regi" <kai_at_rs.ee> wrote in message news:39E8B98E.51BF202_at_rs.ee...Received on Sat Oct 14 2000 - 18:02:52 CDT
> Hi, Dave
>
> I can't find this initialization parameter cursor_sharing anywhere in the
docs.
> V$parameter view shows no rows either.
> Do I miss something, it should be listed in Oracle8i Reference? What value
> should I use instead of FORCE?
> Anyway, I have same problem with Oracle 8.1.5 on Sun and I am looking for
> solutions.
>
> Kai
>
> Dave A wrote:
>
> > No, as it turned out the problem was being caused by the init.ora
parameter
> > curshor_sharing=FORCE. Once turned off the problem went away.
> >
>