Computing aggregated values in a hierarchical structure bottom-up

Question

Hi,

I need to recompute quantities in a BOM structure which only guarantees correct quantity values for leaf nodes.

If the above is a bit blurry don't worry, below is a simplified example.

Consider a hierarchical table defining three columns: ID (PK), PID (parent ID in the hierarchy) and Q (quantity - of something, for the current record).

set nocount on
declare @results table(
        ID  nvarchar(3), 
        PID nvarchar(3), 
        Q   int,
        lvl int,
        ord int
    )

The ord column serves only to sort results.

The lvl column defines the level for the current record in the hierarchy.

For those wondering how these columns are maintained, the @results table is populated in the real world by a function which does all the tricks; for this example they'll be given hardcoded values.

Now, the problem is that correct Q values are guaranteed only for leaf-level nodes in the hierarchy. For the other nodes Q may or may be not correctly defined. My task is to recompute Q values for these nodes.

Sample data:

insert into @results(ord, lvl, ID, PID, Q)
select 1, 0, 'A' as ID, null as PID, null as Q union
select 2, 1, 'B'  , 'A' , 15 union
select 3, 1, 'C'  , 'A' , 10 union
select 4, 2, 'B1' , 'B' , 6  union
select 5, 2, 'B2' , 'B' , 4  union
select 6, 2, 'C1' , 'C' , 5  union
select 7, 2, 'C2' , 'C' , 3  union
select 8, 3, 'C11', 'C1', 4  union
select 9, 3, 'C12', 'C1', 3

As you may see, the quantities for B and C1 are wrong: they're 15 and 5 but should be 10 and 7:

select * from @results order by ord

Here's the initial data:

ID   PID            Q         lvl         ord
---- ---- ----------- ----------- -----------
A    NULL        NULL           0           1
B    A             15           1           2
C    A             10           1           3
B1   B              6           2           4
B2   B              4           2           5
C1   C              5           2           6
C2   C              3           2           7
C11  C1             4           3           8
C12  C1             3           3           9

Finally, the question: is there any way to update this table in a set-based manner so all quantities are updated with bottom-up summed quantities of children nodes?

The best I came with can be seen below, but it's not set based:

declare @level int

select @level = max(lvl) from @results

while (@level > 0)
begin
    update r set Q = s.SumQ
    from @results r inner join (
            select PID, sum(Q) as SumQ 
            from   @results 
            where lvl = @level group by PID
        ) s on ( r.ID = s.PID )

    set @level = @level - 1
end

select * from @results

which gives the correct quantities:

ID   PID            Q         lvl         ord
---- ---- ----------- ----------- -----------
A    NULL          20           0           1
B    A             10           1           2
C    A             10           1           3
B1   B              6           2           4
B2   B              4           2           5
C1   C              7           2           6
C2   C              3           2           7
C11  C1             4           3           8
C12  C1             3           3           9

Thanks!

Answer 1

;WITH   q AS
        (
        SELECT  *, id AS initial
        FROM    @results
        UNION ALL
        SELECT  r.*, initial
        FROM    q
        JOIN    @results r
        ON      r.pid = q.id
        )
UPDATE  ru
SET     q = qn.nq
FROM    @results ru
JOIN    (
        SELECT  initial,
                SUM(rq) AS nq
        FROM    q
        LEFT JOIN
                (
                SELECT  id,
                        CASE
                        WHEN EXISTS
                        (
                        SELECT  NULL
                        FROM    @results ri
                        WHERE   ri.pid = r.id
                        )
                        THEN NULL
                        ELSE q
                        END AS rq
                FROM    @results r
                ) r
        ON      r.id = q.id
                AND r.id <> q.initial
        GROUP BY
                q.initial
        ) qn
ON      ru.id = qn.initial
        AND qn.nq IS NOT NULL