RE: Update of Clobs *Performance*

Mark,

>there is a good chance it will blow up later as the tables grow.
Kidding right?

You bring up a good point, but I disagree. My reasons are:

1. one commit will outperform the counter.
2. if the table is so huge that a single commit is a problem then he shouldn't be looping one by one in the first place
3. undo is a little different here, these are lobs - pctversion

The idea of a counter is useful if you want "partial loads". If half get in and the rest fail, and that's what you want, and you have a reliable point where you can restart, then cool. Otherwise, I really disagree.=20

• Anthony

-----Original Message-----
From: Mark W. Farnham [mailto:mwf_at_rsiz.com]=20 Sent: Wednesday, November 03, 2004 3:17 PM To: Anthony Molinaro; oracle-l_at_freelists.org Subject: RE: Update of Clobs *Performance*

Putting in a counter, picking a reasonable size, commit'n and reset'n the counter when you hit the limit is usually useful.

Even if a single commit for the whole table will work for him now, there is a good chance it will blow up later as the tables grow. It's pretty likely committing each row is unreasonable, but committing monoliths is a recipe for future problems and driving UNDO out of cache without need. I recommend avoiding monolithic commits unless there is a hard requirement for reversibility (rollback) and avoiding a program architecture that drives a need for monolithic commits is up there with the golden mean as far as I'm concerned.

mwf

-----Original Message-----
From: Anthony Molinaro [mailto:amolinaro_at_wgen.net] Sent: Wednesday, November 03, 2004 3:05 PM To: mwf_at_rsiz.com; oracle-l_at_freelists.org Subject: RE: Update of Clobs *Performance*

In regard to: >>>>>>>> Even better, just commit once at the end...

-----Original Message-----
From: Mark W. Farnham [mailto:mwf_at_rsiz.com] Sent: Wednesday, November 03, 2004 3:00 PM To: oracle-l_at_freelists.org
Subject: RE: Update of Clobs *Performance*

create index why_full_scan_all_my_clobs_for_each_one_row_update on tableB(tabB_num)

change your where clause to where tabB_num =3D to_number(v_id)

Think about a commit counter within the loop less than the entire table. Maybe 1000 or 10000?

Regards,

mwf

<snip>

Procedure
declare
v_clob varchar2(32500);
v_id varchar(10);

cursor cont_rep_clob is
select tabA_char, tabA_clob
from Table_A;

begin
open cont_rep_clob;
loop
fetch cont_rep_clob into v_id, v_clob;

exit when cont_rep_clob%NOTFOUND;

update Table_B
set tabB_clob =3D v_clob
where to_char(tabB_num) =3D v_id;

commit;

end loop;
close cont_rep_clob;

--
http://www.freelists.org/webpage/oracle-l