A slightly different running sum in a hierarchy query

Question

Hi,
I've got stuck with this problem since quite some time, so pls help:
I am basically trying to rollup/aggregate values from the lower level nodes to the upper level nodes in a hierarchy. But this aggregate is a bit different than a plain rollup of values.
create table ent_rel (parent number, child number, child_amount number);
insert into ent_rel
select null, 100, null from dual
union all select 100, 101, null from dual
union all select 101, 102, 20 from dual
union all select 102, 103, null from dual
union all select 103, 104, null from dual
union all select 104, 105, null from dual
union all select 105, 106, 40 from dual
union all select 106, 107, null from dual
union all select 107, 108, 20 from dual
union all select 107, 109, 10 from dual
union all select 101, 203, null from dual
union all select 203, 205, 50 from dual
union all select 205, 207, null from dual
union all select 207, 209, null from dual
union all select 209, 210, 10 from dual
commit;
select RPAD(' ', (LEVEL - 1) * 2, '-') || child child
,level
,child_amount
from ent_rel
start with parent is null
connect by prior child = parent
CHILD LEVEL CHILD_AMOUNT
100 1
--101 2
----102 3 20
------103 4
--------104 5
----------105 6
------------106 7 40
--------------107 8
----------------108 9 20
----------------109 9 10
----203 3
------205 4 50
--------207 5
----------209 6
------------210 7 10
Now, I need a SQL query or even a simple PL/SQL code to get the DESIRED_AMOUNT field as shown below:
Pls note that there might be many more parallel branches to a node, but I've kept it simple just to get the concept.
CHILD LEVEL CHILD_AMOUNT DESIRED_AMOUNT
100 1 70
--101 2 70
----102 3 20 20
------103 4 40
--------104 5 40
----------105 6 40
------------106 7 40 40
--------------107 8 30
----------------108 9 20 20
----------------109 9 10 10
----203 3 50
------205 4 50 50
--------207 5 10
----------209 6 10
------------210 7 10 10
Thanks a lot in advance.
thanks and regards,
Knl

Answer 1

Hi,
Welcome to the forum!
Thanks for posting the CREATE TABLE and INSERT statements! That's very helpful.
It's also helpful if you format your code and output. When posting formatted text on this site, type these 6 characters:
\(all small letters, inside curly brackets) to keep the site from compressing the spaces.
You can do two CONNECT BY queries: one to get the desired amount, and the other like you're already doing, to get the indented child column.  The first will differ in that it has no START WITH clause, which means every node in your table will be the root of a tree, and that there's an additional CONNECT BY condition to quit looking for descendants as soon as an amount is found.  This corresponds to your apparant requirement that the desired_amount is the amount, if it is not NULL, and only otherwise is the sum of the desired_amounts of one's descendants.WITH     desired_amount_tree     AS
     SELECT     CONNECT_BY_ROOT child     AS root
     ,     child
     ,     child_amount
     FROM     ent_rel
     CONNECT BY     PRIOR child          = parent
          AND     PRIOR child_amount     IS NULL
,     got_desired_amount     AS
     SELECT     root
     ,     SUM (child_amount)     AS desired_amount
     FROM     desired_amount_tree
     GROUP BY root
select      RPAD ( ' '
     , (LEVEL - 1) * 2
     ) || e.child      AS child
,     level
,     e.child_amount
,     desired_amount
from      ent_rel               e
JOIN     got_desired_amount     d     ON     e.child     = d.root
start with      parent           is null
connect by      prior child      = parent
It's usually more efficient to do join after doing the CONNECT BY.  For better performance, try doing your original query by itself, and then joining its result set to got_desired_amount.
Of course, it would be a lot more efficient if we could eliminate the extra CONNECT BY query.  That might be possible using MODEL or (starting in Oracle 11.2) a recursive sub-query.