Update 1 million records in batches of 1000

Question

Hello,
I have to update 1 million records in groups of 1000. This is the best code I've got but it takes about 1hr and 15 minutes. (I'm also doing a commit every 5k for rollback purposes.) Does anybody have any better ideas?
thanks,
-Kevin
set time on
spool c:\testrowbatch.log
declare
vrow_id varchar2(15);
pcount number:=0;
icount number:=0;
vbnum number:=1;
CURSOR cs_01 is
select row_id from eim_activity where if_row_batch_num = '10';
BEGIN
OPEN cs_01;
LOOP
FETCH cs_01 INTO vrow_id;
IF cs_01%NOTFOUND THEN
dbms_output.put_line('End of Data');
CLOSE cs_01;
END IF;
Update eim_activity set if_row_batch_num = vbnum where row_id = vrow_id;
icount:=icount + 1;
pcount:=pcount + 1;
IF icount = 5000 THEN
commit;
icount:=0;
CLOSE cs_01;
OPEN cs_01;
END IF;
IF pcount = 1000 THEN
vbnum := vbnum +1;
pcount:=0;
end if;
END LOOP;
commit;
end;

Answer 1

There are three problems with commiting inside the loop, particularly when you are updating the table that you have created the cursor on.
First, as everyone pointed out, it makes it slower.
Second, you run a serious risk of getting an ora 1555 Snapshot too old error. In which case you may not be able to reliably restart the procedure.
Third, it actually takes more rollback doing it that way than doing it in a single transaction.
The problem with Todd's approach is that you will update some records multiple times, whether you commit or not. You will set if_row_batch_num = 10 for 1000 records during the 10th iteration of the loop. These records will then be found in a subsequent iteration.
If you are doing this in PL/SQL just to increment the variable for every 1000 records, then this single update will solve that problem. It will be faster than your procedure, it will only update each row once per run, and it will always be an all or nothing update. You will still set if_row_batch_num = 10 for 1000 existing records.
UPDATE eim_activity
SET if_row_batch_num = FLOOR(rownum/1001) +1
WHERE if_row_batch_num = 10;TTFN
John