Help calculating complex sum in hierarchical dataset

Question

I have an interesting SQL problem. I have a hierarchic table of parts that make a bill of material. similar to this:

ASSEMBLY
---------
parent_part_id
part_id
quantity

I get the hierarchy of this structure with a query like this:

SELECT level, part_id, quantity
from assembly
start with parent_part_id = 1
connect by parent_part_id = prior part_id;

the output might look like this:

level  part_id  quantity
-----  -------  ---------
1      2        2
2      3        10
1      4        2
2      5        1    
3      3        5

so far so good.

the question is this: how do I calculate the total number of each part required in order to make the top level assembly (part 1)?

Grouping this result set by part and summing the quantity is not correct, since the quantity should be multiplied by the quantity of the part immediately above the current part in the hierarchy, recursively up the tree.

I am thinking this is a LAG function, but having trouble visualizing it.

edit: expected results:

part_id  quantity
-------  --------
2        2
3        30
4        2
5        2

more edit: i get interesting results with this query

SELECT rownum, level lvl, part_id, quantity, unit_of_measure
                , connect_by_isleaf || sys_connect_by_path(quantity,'*') math
            from assembly
            start with parent_part_id = 1
            connect by parent_part_id = prior part_id

the math column returns a string representation of the calculation i want to perform :) for instance it may say:

1*1*2*10

or something similar and appropriate... perhaps making a function to parse this and return the result will solve the problem.. anyone think this is outrageous?

Answer 1

In Oracle 11 R2 its possible with a common table expression:

The test data:

--  drop table assembly;

create table assembly (
  part_id              number, 
  parent_part_id       number,
  quantity             number
);

insert into assembly values (2, 1,  2);
insert into assembly values (3, 2, 10);
insert into assembly values (4, 1,  2);
insert into assembly values (5, 4,  1);
insert into assembly values (3, 5,  5);

The select statement:

select 
  part_id, 
  sum(quantity_used) as quantity
from (
  with assembly_hier (lvl, part_id, quantity, quantity_used) as (
    select 
      1        lvl,
      part_id,
      quantity ,
      quantity        quantity_used
    from
      assembly
    where
      parent_part_id = 1 
  union all
    select
      assembly_hier.lvl      + 1 lvl,
      assembly     .part_id,
      assembly     .quantity,
      assembly_hier.quantity_used * assembly.quantity quantity_used
    from
      assembly_hier, assembly
    where
      assembly_hier.part_id = assembly.parent_part_id
  )
  select * from assembly_hier
)
group by part_id
order by part_id;

Edit Prior to Ora11R2, it might work with a model clause:

select 
  part_id,
  sum(quantity) quantity 
from (
  select
    lvl
    parent_part_id,
    part_id,
    quantity
  from (
    select 
      lvl,
      parent_part_id,
      part_id,
      quantity
    from (
      select  
        rownum r, 
        level lvl, 
        parent_part_id,
        part_id, 
        quantity
      from 
        assembly
      start with parent_part_id = 1
      connect by parent_part_id = prior part_id
    )
  )
  model
    dimension by (lvl, part_id)
    measures (quantity, parent_part_id)
    rules upsert (
       quantity[     any, any          ] order by lvl, part_id =   quantity[cv(lvl)  , cv(part_id)] * 
                                          nvl( quantity[cv(lvl)-1,    parent_part_id[cv(lvl), cv(part_id)] ], 1)
    )
)
group by part_id
order by part_id;

Edit II Another possibility would be to sum the logarithms of quantity and then take the sum's exponent:

select 
  part_id,
  sum(quantity) quantity
from (
  select 
    part_id,
    exp(sum(quantity_ln) over (partition by new_start order by r)) quantity
  from (
    select 
      r,
      lvl,
      part_id,
      quantity_ln,
      sum(new_start) over(order by r) new_start
    from (
      select 
        rownum r, 
        level lvl, 
        part_id, 
        ln(quantity) quantity_ln,
        nvl(lag(connect_by_isleaf,1) over (order by rownum),0) new_start
      from assembly
      start with parent_part_id = 1
      connect by parent_part_id = prior part_id
    )
  )
)
group by part_id
order by part_id
;

Answer 2

i ended up here: this works on oracle 10 and 11, the connect_by_isleaf can be used to adjust the logic whether you want to sum only the leafs, or all nodes.

select part_id,  new_rec.quantity*sum(math_calc( math,2)) m, unit_of_measure
from ( SELECT rownum, level lvl, part_id, quantity, unit_of_measure
            , connect_by_isleaf || sys_connect_by_path(quantity,'*') math
from assembly
start with parent_part_id = new_rec.part_id
connect by parent_part_id = prior part_id ) p
group by part_id, unit_of_measure