Question about redo generation

Question

select * from v$version;
Oracle Database 11g Enterprise Edition Release 11.2.0.1.0 - Production
PL/SQL Release 11.2.0.1.0 - Production
"CORE     11.2.0.1.0     Production"
TNS for Linux: Version 11.2.0.1.0 - Production
NLSRTL Version 11.2.0.1.0 - ProductionSetup for test
create table parent_1 (id number(12) NOT NULL);
alter table parent_1 add constraint parent_1_pk primary key (id);
create table parent_2 (id number(12) NOT NULL);
alter table parent_2 add constraint parent_2_pk primary key (id);
create table child_table (ref_id number(12) NOT NULL,ref_id2 number(12) NOT NULL, created_at timestamp(6));
alter table child_table add constraint child_table_pk primary key (ref_id, ref_id2);
alter table child_table add constraint child_table_fk1 foreign key (ref_id) references parent_1(id);
alter table child_table add constraint child_table_fk2 foreign key (ref_id2) references parent_2(id);
insert into parent_1 select rownum from all_objects;
insert into parent_2 values (1);
insert into parent_2 values (2);
insert into child_table (select id, 1, systimestamp from parent_1);
insert into child_table (select id, 2, systimestamp from parent_1);
commit;Code version 1:
declare
   type t_ids is table of NUMBER(12);
   v_ids t_ids;
   start_redo NUMBER;
   end_redo NUMBER;
  cursor c_data is SELECT id FROM parent_1;
begin
   select value into start_redo from v$mystat where statistic# = (select statistic# from v$statname where name like 'redo size');
   open c_data;
   LOOP
    FETCH c_data
    BULK COLLECT INTO v_ids LIMIT 1000;
    exit;
   end loop;
  CLOSE c_data;
    for pos in v_ids.first..v_ids.last LOOP
  BEGIN
    insert into child_table values (v_ids(pos), 2, systimestamp);
    EXCEPTION
      WHEN DUP_VAL_ON_INDEX THEN
        update child_table set created_at = systimestamp where ref_id = v_ids(pos) and ref_id2 = 2;
  END;
  END LOOP;
end;
/Version 2:
declare
   type t_ids is table of NUMBER(12);
   v_ids t_ids;
   start_redo NUMBER;
   end_redo NUMBER;
  cursor c_data is SELECT id FROM parent_1;
  ex_dml_errors EXCEPTION;
  PRAGMA EXCEPTION_INIT(ex_dml_errors, -24381);
  pos NUMBER;
  l_error_count NUMBER;
begin
   select value into start_redo from v$mystat where statistic# = (select statistic# from v$statname where name like 'redo size');
   open c_data;
   LOOP
    FETCH c_data
    BULK COLLECT INTO v_ids LIMIT 1000;
    exit;
   end loop;
  CLOSE c_data;
  BEGIN
    FORALL i IN v_ids.first .. v_ids.last SAVE EXCEPTIONS
    insert into child_table values (v_ids(i), 2, systimestamp);
  EXCEPTION
    WHEN ex_dml_errors THEN
      l_error_count := SQL%BULK_EXCEPTIONS.count;
      FOR i IN 1 .. l_error_count LOOP
        pos := SQL%BULK_EXCEPTIONS(i).error_index;
        update child_table set created_at = systimestamp where ref_id = v_ids(pos) and ref_id2 = 2;
      END LOOP;
  END;
   select value into end_redo from v$mystat where statistic# = (select statistic# from v$statname where name like 'redo size');
  dbms_output.put_line('Created redo : ' || (end_redo-start_redo));
end;
/Version 1 output:
Created redo : 682644
Version 2 output:
Created redo : 7499364
Why is version 2 generating significant more redo ?

Answer 1

As both pieces of code erroneously replace non-procedural code by procedural code, ignoring the power of a RDBMS to process sets, and are examples of slow by slow programming,
both pieces of code are undesirable, so the difference in redo generation doesn't matter.
Sybrand Bakker
Senior Oracle DBA