10

Given a SourceTable and a TargetTable, I would like to programmatically create a string with all joins required.

In short, I am trying to find a way to create a string like this:

FROM SourceTable t
JOIN IntermediateTable t1 on t1.keycolumn = t.keycolumn
JOIN TargetTable t2 on t2.keycolumn = t1.keycolumn

I have a query that returns all foreign keys for a given table, but am running into limitations in trying to run through all this recursively to find the optimal join path and make the string.

SELECT 
    p.name AS ParentTable
    ,pc.name AS ParentColumn
    ,r.name AS ChildTable
    ,rc.name AS ChildColumn
FROM sys.foreign_key_columns fk
JOIN sys.columns pc ON pc.object_id = fk.parent_object_id AND pc.column_id = fk.parent_column_id 
JOIN sys.columns rc ON rc.object_id = fk.referenced_object_id AND rc.column_id = fk.referenced_column_id
JOIN sys.tables p ON p.object_id = fk.parent_object_id
JOIN sys.tables r ON r.object_id = fk.referenced_object_id
WHERE fk.parent_object_id = OBJECT_ID('aTable')
ORDER BY ChildTable, fk.referenced_column_id

I am sure this has been done before, but I can't seem to find an example.

6
  • 2
    What if there are 2 or more paths from source to target? May 13, 2015 at 18:49
  • 2
    Yeah, I'd be concerned about multiple potential paths, and also about a single path that is more than 2 steps. Also, keys comprised of more than one column. Those scenarios will all throw a wrench in any automated solution. May 13, 2015 at 18:53
  • Notice that even one single foreign key between two tables will allow 2 or more paths (actually an unlimited number of paths of arbitrary length). Consider the query "find all items that have been placed at least once in the same order with item X". You'll need to join OrderItems with Orders and back with OrderItems. May 13, 2015 at 19:06
  • 2
    @ypercube Right, also, what exactly does "the optimal path" mean? May 13, 2015 at 19:11
  • "Optimal JOIN path" means "the shortest series of joins that will join Target table to the Source table." If T1 is referenced in T2 and T3, T2 is referenced in T4 and T3 is referenced in T4. The optimal path from T1 to T3 is T1,T2,T3. The T1,T2,T4,T3 path would not be optimal because it is longer.
    – Metaphor
    May 13, 2015 at 19:33

2 Answers 2

5

I had a script that does a rudimentary version of foreign key traversal. I adapted it quickly (see below), and you might be able to use it as a starting point.

Given a target table, the script attempts to print the join string for the shortest path (or one of them in the case of ties) for all possible source tables such that single-column foreign keys can be traversed to reach the target table. The script seems to be working well on the database with a couple thousand tables and many FK connections that I tried it on.

As others mention in the comments, you'd need to make this more complex if you need to handle multi-column foreign keys. Also, please be aware that this is not by any means production-ready, fully-tested code. Hope it's a helpful starting point if you do decide to build out this functionality!

-- Drop temp tables that will be used below
IF OBJECT_ID('tempdb..#paths') IS NOT NULL
    DROP TABLE #paths
GO
IF OBJECT_ID('tempdb..#shortestPaths') IS NOT NULL
    DROP TABLE #shortestPaths
GO

-- The table (e.g. "TargetTable") to start from (or end at, depending on your point of view)
DECLARE @targetObjectName SYSNAME = 'TargetTable'

-- Identify all paths from TargetTable to any other table on the database,
-- counting all single-column foreign keys as a valid connection from one table to the next
;WITH singleColumnFkColumns AS (
    -- We limit the scope of this exercise to single column foreign keys
    -- We explicitly filter out any multi-column foreign keys to ensure that they aren't misinterpreted below
    SELECT fk1.*
    FROM sys.foreign_key_columns fk1
    LEFT JOIN sys.foreign_key_columns fk2 ON fk2.constraint_object_id = fk1.constraint_object_id AND fk2.constraint_column_id = 2
    WHERE fk1.constraint_column_id = 1
        AND fk2.constraint_object_id IS NULL
)
, parentCTE AS (
    -- Base case: Find all outgoing (pointing into another table) foreign keys for the specified table
    SELECT 
        p.object_id AS ParentId
        ,OBJECT_SCHEMA_NAME(p.object_id) + '.' + p.name AS ParentTable
        ,pc.column_id AS ParentColumnId
        ,pc.name AS ParentColumn
        ,r.object_id AS ChildId
        ,OBJECT_SCHEMA_NAME(r.object_id) + '.' + r.name AS ChildTable
        ,rc.column_id AS ChildColumnId
        ,rc.name AS ChildColumn
        ,1 AS depth
        -- Maintain the full traversal path that has been taken thus far
        -- We use "," to delimit each table, and each entry then has a
        -- "<object_id>_<parent_column_id>_<child_column_id>" format
        ,   ',' + CONVERT(VARCHAR(MAX), p.object_id) + '_NULL_' + CONVERT(VARCHAR(MAX), pc.column_id) +
            ',' + CONVERT(VARCHAR(MAX), r.object_id) + '_' + CONVERT(VARCHAR(MAX), pc.column_id) + '_' + CONVERT(VARCHAR(MAX), rc.column_id) AS TraversalPath
    FROM sys.foreign_key_columns fk
    JOIN sys.columns pc ON pc.object_id = fk.parent_object_id AND pc.column_id = fk.parent_column_id 
    JOIN sys.columns rc ON rc.object_id = fk.referenced_object_id AND rc.column_id = fk.referenced_column_id
    JOIN sys.tables p ON p.object_id = fk.parent_object_id
    JOIN sys.tables r ON r.object_id = fk.referenced_object_id
    WHERE fk.parent_object_id = OBJECT_ID(@targetObjectName)
        AND p.object_id <> r.object_id -- Ignore FKs from one column in the table to another

    UNION ALL

    -- Recursive case: Find all outgoing foreign keys for all tables
    -- on the current fringe of the recursion
    SELECT 
        p.object_id AS ParentId
        ,OBJECT_SCHEMA_NAME(p.object_id) + '.' + p.name AS ParentTable
        ,pc.column_id AS ParentColumnId
        ,pc.name AS ParentColumn
        ,r.object_id AS ChildId
        ,OBJECT_SCHEMA_NAME(r.object_id) + '.' + r.name AS ChildTable
        ,rc.column_id AS ChildColumnId
        ,rc.name AS ChildColumn
        ,cte.depth + 1 AS depth
        ,cte.TraversalPath + ',' + CONVERT(VARCHAR(MAX), r.object_id) + '_' + CONVERT(VARCHAR(MAX), pc.column_id) + '_' + CONVERT(VARCHAR(MAX), rc.column_id) AS TraversalPath
    FROM parentCTE cte
    JOIN singleColumnFkColumns fk
        ON fk.parent_object_id = cte.ChildId
        -- Optionally consider only a traversal of the same foreign key
        -- With this commented out, we can reach table A via column A1
        -- and leave table A via column A2.  If uncommented, we can only
        -- enter and leave a table via the same column
        --AND fk.parent_column_id = cte.ChildColumnId
    JOIN sys.columns pc ON pc.object_id = fk.parent_object_id AND pc.column_id = fk.parent_column_id 
    JOIN sys.columns rc ON rc.object_id = fk.referenced_object_id AND rc.column_id = fk.referenced_column_id
    JOIN sys.tables p ON p.object_id = fk.parent_object_id
    JOIN sys.tables r ON r.object_id = fk.referenced_object_id
    WHERE p.object_id <> r.object_id -- Ignore FKs from one column in the table to another
        -- If our path has already taken us to this table, avoid the cycle that would be created by returning to the same table
        AND cte.TraversalPath NOT LIKE ('%_' + CONVERT(VARCHAR(MAX), r.object_id) + '%')
)
SELECT *
INTO #paths
FROM parentCTE
ORDER BY depth, ParentTable, ChildTable
GO

-- For each distinct table that can be reached by traversing foreign keys,
-- record the shortest path to that table (or one of the shortest paths in
-- case there are multiple paths of the same length)
SELECT *
INTO #shortestPaths
FROM (
    SELECT *, ROW_NUMBER() OVER (PARTITION BY ChildTable ORDER BY depth ASC) AS rankToThisChild
    FROM #paths
) x
WHERE rankToThisChild = 1
ORDER BY ChildTable
GO

-- Traverse the shortest path, starting from the source the full path and working backwards,
-- building up the desired join string as we go
WITH joinCTE AS (
    -- Base case: Start with the from clause to the child table at the end of the traversal
    -- Note that the first step of the recursion will re-process this same row, but adding
    -- the ParentTable => ChildTable join
    SELECT p.ChildTable
        , p.TraversalPath AS ParentTraversalPath
        , NULL AS depth
        , CONVERT(VARCHAR(MAX), 'FROM ' + p.ChildTable + ' t' + CONVERT(VARCHAR(MAX), p.depth+1)) AS JoinString
    FROM #shortestPaths p

    UNION ALL

    -- Recursive case: Process the ParentTable => ChildTable join, then recurse to the
    -- previous table in the full traversal.  We'll end once we reach the root and the
    -- "ParentTraversalPath" is the empty string
    SELECT cte.ChildTable
        , REPLACE(p.TraversalPath, ',' + CONVERT(VARCHAR, p.ChildId) + '_' + CONVERT(VARCHAR, p.ParentColumnId)+ '_' + CONVERT(VARCHAR, p.ChildColumnId), '') AS TraversalPath
        , p.depth
        , cte.JoinString + '
' + CONVERT(VARCHAR(MAX), 'JOIN ' + p.ParentTable + ' t' + CONVERT(VARCHAR(MAX), p.depth) + ' ON t' + CONVERT(VARCHAR(MAX), p.depth) + '.' + p.ParentColumn + ' = t' + CONVERT(VARCHAR(MAX), p.depth+1) + '.' + p.ChildColumn) AS JoinString
    FROM joinCTE cte
    JOIN #paths p
        ON p.TraversalPath = cte.ParentTraversalPath
)
-- Select only the fully built strings that end at the root of the traversal
-- (which should always be the specific table name, e.g. "TargetTable")
SELECT ChildTable, 'SELECT TOP 100 * 
' +JoinString
FROM joinCTE
WHERE depth = 1
ORDER BY ChildTable
GO
0

You can put the list of keys of a table with two fields TAB_NAME,KEY_NAME for all the tables you want to connect.

Example, for table City

  • City|City_name
  • City|Country_name
  • City|Province_name
  • City|City_Code

likewise Province and Country.

Collect the data for the tables and put into a single table ( Eg. Metadata table)

Now draft the query like below

select * from
(Select Table_name,Key_name from Meta_Data 
where Table_name in ('City','Province','Country')) A,
(Select Table_name,Key_name from Meta_Data 
where Table_name in ('City','Province','Country')) B,
(Select Table_name,Key_name from Meta_Data 
where Table_name in ('City','Province','Country')) C

where

A.Table_Name <> B.Table_name and
B.Table_name <> C.Table_name and
C.Table_name <> A.Table_name and
A.Column_name = B.Column_name and
B.Column_name = C.Column_name

This will get you how can you link the tables based on the matching keys(same key names)

If you think the keys name may not match you can include an alternate key field and try to use it in the where condition.

1
  • Note that the asker wanted to use the existing sys tables in SQL Server that describe the columns in a table, how tables are linked together, etc. All that already exists. Building your own tables that define your table structure to meet a specific need could be a fallback position, but the preferred answer would use what already exists, as the accepted answer does.
    – RDFozz
    Aug 21, 2018 at 15:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.