Deleting rows from parent and child tables

Assume two tables in Oracle 10G

TableA (Parent) --> TableB (Child)

Every row in TableA has several child rows related to it in TableB.

I want to delete specific rows in TableA which means i have to delete the related rows in tableB first.

This deletes the child entries

delete from tableB where last_update_Dtm = sysdate-30;

To delete the parent rows for the rows just deleted in the child table I could do something like this

Delete from TableA where not exists (select 1 from tableB where tableA.key=tableB.key);

The above will will also delete rows in the child table where (last_update_Dtm = sysdate-30) is false. TableA does not have a last_update_dtm column so there is no way of knowing which rows to delete without the entries in the child table.

I could save the keys in the child table prior to deleting but this seems like an expensive approach. What is the correct way of deleting the rows in both tables?

Edit

To explai开发者_如何学Cn better what i am trying to achieve, the following query would have done what i am trying to do if there was no constraint between the two table.

Delete from tableA
Where exists (
Select 1 from tableB
where tableA.key=tableB.key
and tableB.last_update_dtm=sysdate-30)

Delete from tableB where last_update_dtm=systdate-30

---

Two possible approaches.

1. If you have a foreign key, declare it as on-delete-cascade and delete the parent rows older than 30 days. All the child rows will be deleted automatically.

2. Based on your description, it looks like you know the parent rows that you want to delete and need to delete the corresponding child rows. Have you tried SQL like this?

     delete from child_table
         where parent_id in (
              select parent_id from parent_table
                   where updd_tms != (sysdate-30)
   

   -- now delete the parent table records

   delete from parent_table
   where updd_tms != (sysdate-30);
   

---- Based on your requirement, it looks like you might have to use PL/SQL. I'll see if someone can post a pure SQL solution to this (in which case that would definitely be the way to go).

declare
    v_sqlcode number;
    PRAGMA EXCEPTION_INIT(foreign_key_violated, -02291);
begin
    for v_rec in (select parent_id, child id from child_table
                         where updd_tms != (sysdate-30) ) loop

    -- delete the children
    delete from child_table where child_id = v_rec.child_id;

    -- delete the parent. If we get foreign key violation, 
    -- stop this step and continue the loop
    begin
       delete from parent_table
          where parent_id = v_rec.parent_id;
    exception
       when foreign_key_violated
         then null;
    end;
 end loop;
end;
/

---

If the children have FKs linking them to the parent, then you can use DELETE CASCADE on the parent.

e.g.

CREATE TABLE supplier 
( supplier_id numeric(10) not null, 
 supplier_name varchar2(50) not null, 
 contact_name varchar2(50),  
 CONSTRAINT supplier_pk PRIMARY KEY (supplier_id) 
); 



CREATE TABLE products 
( product_id numeric(10) not null, 
 supplier_id numeric(10) not null, 
 CONSTRAINT fk_supplier 
   FOREIGN KEY (supplier_id) 
  REFERENCES supplier(supplier_id) 
  ON DELETE CASCADE 
); 

Delete the supplier, and it will delate all products for that supplier

---

Here's a complete example of how it can be done. However you need flashback query privileges on the child table.

Here's the setup.

create table parent_tab
  (parent_id number primary key,
  val varchar2(20));

create table child_tab
    (child_id number primary key,
    parent_id number,
    child_val number,
     constraint child_par_fk foreign key (parent_id) references parent_tab);

insert into parent_tab values (1,'Red');
insert into parent_tab values (2,'Green');
insert into parent_tab values (3,'Blue');
insert into parent_tab values (4,'Black');
insert into parent_tab values (5,'White');

insert into child_tab values (10,1,100);
insert into child_tab values (20,3,100);
insert into child_tab values (30,3,100);
insert into child_tab values (40,4,100);
insert into child_tab values (50,5,200);

commit;

select * from parent_tab
where parent_id not in (select parent_id from child_tab);

Now delete a subset of the children (ones with parents 1,3 and 4 - but not 5).

delete from child_tab where child_val = 100;

Then get the parent_ids from the current COMMITTED state of the child_tab (ie as they were prior to your deletes) and remove those that your session has NOT deleted. That gives you the subset that have been deleted. You can then delete those out of the parent_tab

delete from parent_tab
where parent_id in
  (select parent_id from child_tab as of scn dbms_flashback.get_system_change_number
  minus
  select parent_id from child_tab);

'Green' is still there (as it didn't have an entry in the child table anyway) and 'Red' is still there (as it still has an entry in the child table)

select * from parent_tab
where parent_id not in (select parent_id from child_tab);

select * from parent_tab;

It is an exotic/unusual operation, so if i was doing it I'd probably be a bit cautious and lock both child and parent tables in exclusive mode at the start of the transaction. Also, if the child table was big it wouldn't be particularly performant so I'd opt for a PL/SQL solution like Rajesh's.