SQL Server Recursive Hierarchy Query - sql

Question

This is for SQL Server 2012. I need to generate a dataset containing links, and all links of links from a given starting ParentId given the following table
CREATE TABLE Relations (
Id INT NOT NULL PRIMARY KEY,
ParentId INT NOT NULL,
ChildId INT
);
So for the following dataset:
1 A B
2 B C
3 C D
4 F D
5 F G
6 X Y
7 Y Z
Starting with C, I'd expected to get back rows 1 to 5 as they're all linked to C through either parent or child hierarchies. E.g. G has parent F, which is parent of D, which is child of C.
It's not a standard hierarchy query as there's no real root, and I need to get links in both directions. So this means I can't use the CTE recursion trick.. here was my attempt:
--Hierarchical Query using Common Table Expressions
WITH ReportingTree (Id, Parent, Child, Lvl)
AS
(
--Anchor Member
SELECT Id,
ParentId,
ChildId,
0 as Lvl
FROM Relations WHERE ParentId = 9488
UNION ALL
--Recusive Member
SELECT
cl.Id,
cl.ParentId,
cl.ChildId,
r1.Lvl+1
FROM [dbo].[CompanyLinks] cl
INNER JOIN ReportingTree r1 ON ReportingTree.Parent = cl.Child
INNER JOIN ReportingTree r2 ON cl.FromCompanyId = r2.Parent <-- errors
)
SELECT * FROM ReportingTree
My second attempt involved a temp table and while loop. This works but turns out to be very slow:
BEGIN
CREATE TABLE #R (
Id INT NOT NULL PRIMARY KEY NONCLUSTERED,
ParentId INT NOT NULL,
ChildId INT
);
CREATE CLUSTERED INDEX IX_Parent ON #R (ParentId);
CREATE INDEX IX_Child ON #R (ChildId);
INSERT INTO #R
SELECT Id,ParentId ChildId
FROM Relations
WHERE ParentId = 9488 OR ChildId = 9488;
WHILE ##RowCount > 0
BEGIN
INSERT INTO #R
SELECT cl.Id,cl.ParentId, cl.ChildId
FROM #R INNER JOIN
Relations AS cl ON cl.ChildId = #R.ChildId OR cl.ParentId = #R.ParentId OR cl.ChildId = #R.Parent OR cl.ParentId = #R.Child
EXCEPT
SELECT Id,ParentId, ChildId
FROM #R;
END
SELECT * FROM Relations cl inner Join #Relations r ON cl.Id = #R.Id
DROP TABLE #R
END
Can anyone suggest a workable solution for this?

Answer 1

We match every row with every other row based on every combination of parent and child ids, and save the path along the way. Recursively we do this matching and make the path, in order to avoid infinite loops we check the path is not traversed previously, finally we have all nodes that has a path to the desired node(#Id):
WITH cte AS (
SELECT CompanyLinks.*, cast('(' + cast(ParentId as nvarchar(max)) + ','
+ cast(ChildId as nvarchar(max))+')' as nvarchar(max)) Path
FROM CompanyLinks
WHERE ParentId = #Id OR ChildId = #Id
UNION ALL
SELECT a.*,
cast(
c.Path + '(' +
cast(a.ParentId as nvarchar(max)) + ',' +
cast(a.ChildId as nvarchar(max)) + ')'
as nvarchar(max)
) Path
FROM CompanyLinks a JOIN cte c ON
a.ParentId = c.ChildId
OR c.ParentId = a.ChildId
OR c.ParentId = a.ParentId
OR c.ChildId = a.ChildId
where c.Path not like cast(
'%(' +
cast(a.ParentId as nvarchar(max)) + ',' +
cast(a.ChildId as nvarchar(max)) +
')%'
as nvarchar(max)
)
)
SELECT DISTINCT a.id, Company.Name, path from (
SELECT distinct ParentId as id, path FROM cte
union all
SELECT distinct ChildId as id, path FROM cte
) a inner join Company on Company.Id = a.Id
Here is a fiddle for it.
If you want distinct nodes just use:
SELECT DISTINCT id from (
SELECT distinct ParentId as id FROM cte
union all
SELECT distinct ChildId as id FROM cte
) a
at the end of query.
This query is actually a Breadth First Search on an un-directed graph.
Note: Based on Hogan comment, there is no need for checking the path, as there is a primary key in the relation table(which I did not noticed) we can look for the primary key in prior recursions to avoid infinite loops.