RE: tuning a massive delete

Perhaps this small example can make it clear? I have two tables, orders and order_to_delete. I want to find orders to KEEP - i.e. the order_id is not in table order_to_delete, or it's in table order_to_delete with a status 'N'.

If I'm using outer joins, I think I need a a union, nicht wahr?

drop table order_to_delete ;
drop table orders ;
create table orders (order_id number not null, order_date date) ; create table order_to_delete (order_id number not null, delete_flag varchar2 (1)) ; insert into orders (order_id, order_date)  values (1, sysdate - 1) ;
insert into orders (order_id, order_date)  values (2, sysdate - 2) ;
insert into orders (order_id, order_date)  values (3, sysdate - 3) ;
insert into order_to_delete (order_id, delete_flag)  values (1, 'Y') ;
insert into order_to_delete (order_id, delete_flag)  values (2, 'N') ;
commit ;
 select a.order_id, a.order_date
  from orders a, order_to_delete b
  where a.order_id = b.order_id

        and b.delete_flag = 'N'
union
 select a.order_id, a.order_date
  from orders a, order_to_delete b
  where a.order_id = b.order_id (+)

        and b.order_id is null ;

This query will return order ids 2 and 3, the ones I want to KEEP. The query returns the same value in Oracle 8.1 and Oracle 9.2

> -----Original Message-----
> From: ml-errors_at_fatcity.com [mailto:ml-errors_at_fatcity.com]On Behalf Of
> Binley Lim
>
> On a general note, this older outer join syntax to simulate a not-in
> requires ALL the "b" columns to include the (+) sign, including the
> "nvl...", except the "b.cusip is null", which is the not-in
> itself. If you
> miss one, the logic is completely changed.
>
> 9i's new syntax does outer-joins by doing the outer on the
> tables, rather
> than the columns, so you can no longer do this outer "simulation"!
>
> In this case, you can achieve the same thing with a not-in,
> rather than (+),
> in the SQL syntax (assuming you see a HASH AJ in the plan somewhere).
>
> ----- Original Message -----
> To: "Multiple recipients of list ORACLE-L" <ORACLE-L_at_fatcity.com>
> Sent: Thursday, September 18, 2003 9:04 AM
>
>
> > rgaffuri_at_cox.net wrote:
> > >
> > > create table ani_prx_faster parallel (degree 5) nologging
> > > as
> > > select b.*
> > > from bo_owner_master.ani_prx b,
> > > bo_owner_stage.ani_prx a
> > > where a.cusip = b.cusip (+)
> > > and a.fund_no = b.fund_no (+)
> > > and a.add_cymd = b.add_cymd (+)
> > > and nvl(b.ba_reccode, 'X') != 'V'
> > > and b.cusip is null
> > >
> > > This query I got from here I got only 638k, and I have
> 27m records in my
> file and there are 17m deletes. so 12m records left. Also all
> the records
> are null????
> > >
> > > what do i need to change?
> > > >
> >
> > Ooops, I have permuted a's and b's in the where clause ...
> >
> > Otherwise not to sure about the nvl(). Wait a minute ...
> we are getting
> > rows for which there is no matching (cusip, fund_no,
> add_cymd) in the
> > staging table. Perhaps that the 'cusip is null' condition
> should go ...
> > the nvl() should be enough.

-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.net
-- 
Author: Jacques Kilchoer
  INET: Jacques.Kilchoer_at_quest.com

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).