How to check circular reference in data of millions of rows

Question

Hi all,
I am new as an oracle developer.
I have a table(say table1) of hierarchical data with columns child_id and parent_id. The table has around 0.5 million records. There are around 0.2 million root level records are present.
I need to update some (the number variies on different conditions, might differ from 50K to 0.4 million) of the parent_ids from other table only if that do not cause any circular reference in data of table1.
I am using Oracle9i.
The approach I have tried is:
I copied all data from table1 to a work table table3. Update the parent_id on columns that matches the condition and updated a column source_table to table2.
I copied all the child_id from table2 that are to be updated into another table table4 and built tree for each of those childs. If it builds the tree successfully updated a column is_circular in the table3 to 0 and then deleted the child_id from table4.
This whole process needs to run recursively until no tree is built successfully in an iteration.
But this process is running slow and at the rate it is executing it would run for days for the 0.3 million records.
Please suggest a better way to do this.
The code I am using is :
INSERT /*+ append parallel (target,5) */
INTO table3 target
select /*+ parallel (src,5) */
child_id, parent_id, 'table1' source_table,null
from table1 src;
INSERT INTO table4
select distinct child_id, parent_id
from table2
where status_cd is null;
commit;
Update dtable3 a
set (parent_id,source_table) = (select parent_id,'table2' source_table
from table4 b
where b.child_id = a.child_id);
WHILE v_loop_limit<>v_new_count
LOOP
select count(1)
into v_loop_limit
     from table4;
-+-----------------------------------------------------------
--| You need to actually traverse the tree in order to
--| detect circular dependencies.
+-----------------------------------------------------------           
For i in 1 .. v_new_count
BEGIN
select child_id
into v_child_id from(
select child_id, row_number() over (order by child_id) pri from table4)
where pri = i;
SELECT COUNT (*) cnt
INTO v_cnt
FROM table3 dl
START WITH dl.child_id = v_child_id
CONNECT BY PRIOR dl.child_id = dl.parent_id ;
UPDATE table3SET is_circular = '0'
where child_id= v_child_id
and source_table = 'table2';
delete from table3
where child_id = v_child_id;
select count(1)
into v_new_count
     from table4;
COMMIT;
EXCEPTION WHEN e_infinite_loop
THEN dbms_output.put_line ('Loop detected!'||v_child_id);
WHEN OTHERS THEN
     dbms_output.put_line ('Other Exceptions!');
END;
END LOOP;
END LOOP;
Thanks & Regards,
Ruchira

Answer 1

Hello,
First off, if you're so new to a technology why do you brush off the first thing an experienced, recognized "guru" tells you? Take out that WHEN OTHERS ! It is a bug in your code, period.
Secondly, here is some code to try. Notice that the result depends on what order you do the updates, that's why I run the same code twice but in ASCENDING or DESCENDING order. You can get a circular reference from a PAIR of updates, and whichever comes second is the "bad" one.drop table tab1;
create table TAB1(PARENT_ID number, CHILD_ID number);
create index I1 on TAB1(CHILD_ID);
create index I2 on TAB1(PARENT_ID);
drop table tab2;
create table TAB2(PARENT_NEW number, CHILD_ID number);
insert into TAB1 values (1,2);
insert into TAB1 values (1,3);
insert into tab1 values (2,4);
insert into TAB1 values (3,5);
commit;
insert into TAB2 values(4,3);
insert into TAB2 values(3,2);
commit;
begin
for REC in (select * from TAB2 order by child_id ASC) LOOP
  merge into TAB1 O
  using (select rec.child_id child_id, rec.parent_new parent_new from dual) n
  on (O.CHILD_ID = N.CHILD_ID)
  when matched then update set PARENT_ID = PARENT_NEW
  where (select COUNT(*) from TAB1
    where PARENT_ID = n.CHILD_ID
    start with CHILD_ID = n.PARENT_NEW
    connect by CHILD_ID = prior PARENT_ID
    and prior PARENT_ID <> n.CHILD_ID) = 0;
end LOOP;
end;
select distinct * from TAB1
connect by PARENT_ID = prior CHILD_ID
order by 1, 2;
rollback;
begin
for REC in (select * from TAB2 order by child_id DESC) LOOP
  merge into TAB1 O
  using (select rec.child_id child_id, rec.parent_new parent_new from dual) n
  on (O.CHILD_ID = N.CHILD_ID)
  when matched then update set PARENT_ID = PARENT_NEW
  where (select COUNT(*) from TAB1
    where PARENT_ID = n.CHILD_ID
    start with CHILD_ID = n.PARENT_NEW
    connect by CHILD_ID = prior PARENT_ID
    and prior PARENT_ID <> n.CHILD_ID) = 0;
end LOOP;
end;
select distinct * from TAB1
connect by PARENT_ID = prior CHILD_ID
order by 1, 2;One last thing, be sure and have indexes on child_id and parent_id in your main table!