Question

Using SQL Server 2008 and any available features of TSQL, I am trying to work out how to if a set-based solution that does not involve a temp table exists for the following problem:

Given a set of nodes that have a parent-child relationship, and a set of key-value pairs that apply to each, and given that
the value (for a given key-value pair) at a deeper level of the node hierarchy will override a value with the same key that
is inherited from an ancestor node, select:

1. the full set of key-value pairs that apply to a given node
2. the set of inherited values for that node

The schema is as follows:

create table Node
(
    ID bigint identity primary key,
    ParentID bigint null foreign key references Node(ID),
    Name nvarchar(100)
);

create table KeyValuePair
(
    ID bigint identity primary key,
    KeyName nvarchar(100) not null,
    Value nvarchar(1000) not null,
    NodeID bigint not null foreign key references Node(ID),

    unique (KeyName, NodeID)
);

The result set would essentially include the columns KeyName, Value, InheritedValue.

I’ve been trying to do this using a common table expression but the logic of it is a bit tricky.

Answer 1

I setup the Node and KeyValuePair tables as per the question, and populated with some sample values, such that my hierarchy was as follows:

Root
|—A
|   |—A1
|   |—A2
|
|—B
    |—B1
    |—B2

I assigned two properties, named “Property 1” and “Property 2”, each defined in root with values “Root Prop 1” and “Root Prop 2” respectively. In A, I overrode “Property 1” with value “A Prop 1” and in B, I overrode “Property 2” with value “B Prop 2”.

set identity_insert Node on
insert into Node(ID,ParentID,Name)
values (1,null,'Root'),(2,1,'A'),(3,1,'B'),(4,2,'A1'),(5,2,'A2'),
       (6,3,'B1'),(7,3,'B2')
set identity_insert Node off

insert into KeyValuePair(KeyName, [Value], NodeID)
values ('Property 1','Root Prop 1',1),
('Property 2','Root Prop 2',1),
('Property 1','A Prop 1',2),
('Property 2','B Prop 2',3)

Calling Nathan’s solution for node A1 yields no rows!

The where clause in Nathan’s solution should be a condition of the join between keys and v, resulting in the revised procedure shown below (also I’ve renamed DataValue to KeyValuePair to be consistent with the original question):

create procedure dbo.ListDataValues
    @nodeid bigint
as
begin
    with nodes as (
        select ID, ParentID, 0 as Level
        from Node n where ID=@nodeid
        union all
        select n.ID, n.ParentID, c.Level+1 as Level
        from Node n inner join nodes c on c.ParentID = n.ID
    ),
    keys as (
        select distinct(KeyName)
        from KeyValuePair
        where NodeID in (select ID from nodes)
    )
    select
        keys.KeyName,
        v.Value,
        i.Value as [InheritedValue],
        i.NodeID as [InheritedFromNodeID]
    from
        keys
        left join KeyValuePair v on v.KeyName = keys.KeyName
                                     and v.NodeID = @nodeid
        left join KeyValuePair i on i.KeyName = keys.KeyName
            and i.NodeID = (select top 1 NodeID from KeyValuePair d
                            inner join nodes k on k.ID = d.NodeID
                            where Level > 0 and d.KeyName = i.KeyName
                            order by [Level])
end
go

This yielded the correct results as expected:

KeyName      Value   InheritedValue    InheritedFromNodeID
------------ ------- ----------------- --------------------
Property 1   NULL    A Prop 1          2
Property 2   NULL    Root Prop 2       1