Need clarification on doing bulk updates

Question

Hi All,
I got a requirement where I need to insert around 100 Million records and do updates on the entire content of the table.
Requirement is I have a customer order table as given below
CREATE TABLE customer_order(group_no number, order_no number) nologging no compress parallel 10;
This is going to have millions of rows (100 Million ).
When we are inserting the data we are inserting only order_no.
After insertion we need to update the group_no for all these orders. Here every 1000 order_no should have one group_no assigned to it. But when are picking up the 1000 order_no's those 1000 should be randomly picked up from the table and fors those 1000 orders we need to assign group no.
We are able to insert 100 Million rows into this table in under 5 minutes. (FYI there are no indexes on this table as we found indexes are of no use here and creation of indexes itself is taking more than 3 hours and the index size is greater than table size).
To get the random order numbers when are using DBMS_RANDOM..VALUE package it is taking for ever. Can anyone let us know how to tackle this challenging situation.
Appreciate your inputs here.
Thanks,
Maddy.

Answer 1

Do you need exactly 1000 order_no per group_no? If close enough is good enough I'd use:
INSERT
  INTO customer_order
  SELECT  ora_hash(level,100000 - 1),
          level
    FROM  dual
    CONNECT BY level <= 100000000
/For example:
SQL> CREATE TABLE customer_order(group_no number, order_no number)
  2  /
Table created.
SQL> INSERT
  2    INTO customer_order
  3    SELECT  ora_hash(level,10 - 1),
  4            level
  5      FROM  dual
  6      CONNECT BY level <= 10000
  7  /
10000 rows created.
SQL> select  group_no,
  2          count(*),
  3          min(order_no),
  4          max(order_no)
  5    from  customer_order
  6    group by group_no
  7  /
  GROUP_NO   COUNT(*) MIN(ORDER_NO) MAX(ORDER_NO)
         1       1001             4          9992
         6        991             2          9999
         2        989             6          9995
         4        989            18         10000
         5       1028            37          9996
         8       1006             5          9990
         7       1021             1          9988
         3        973            10          9998
         9        985             3          9977
         0       1017             8          9986
10 rows selected.
SQL> As you can see number of order_no per group no is close enough to 1000 and order_no within group_no are mixed (not sequential). And in any case, it would be more efficient to populate both order_no and group_no in single insert and then, if you need exact count per group no, issue smaller update.
SY.