Two table recursive lookup without hierarchy, is this possible? - sql

Question

This is the current design of the SQL Server database I am working with.
I have two tables:
recipes
recipe ingredients
Recipes consist of recipe ingredients but an ingredient can be another recipe. In theory there are infinite levels as each recipe can have another ingredient that is also a recipe.
In the above data example, the Fresh Salsa recipe (ID 3047) has 7 ingredients. Six are raw Materials but one is another recipe (Recipe ID 3008). This recipe ID references another recipe in the 'recipes' table.
There is no hierarchy and I don't think I can create a hierarchy.
The goal is to extract all the recipe items for a particular recipe that have a 'sub' recipes and 'sub-sub' recipes etc.
It would seem like a recursive lookup would be the answer but because there is no hierarchy, this doesn't seem to work.
Here's my attempted query (the recipeItem list variable is a list of all the recipeitems that are also recipes created in a previous query):
<cfquery name="whatever">
WITH MenuPrepOfPreps (recipe_id, depth, otherRecipe_id, recipe_name)
AS
(
SELECT r.recipe_id,
0 as depth,
ri.otherRecipe_id,
r.recipe_name
FROM menu_recipes r
JOIN menu_recipeItems ri
ON ri.otherRecipe_id = r.recipe_id
WHERE ri.otherRecipe_id in (#recipeItemList#)
UNION ALL
-- recursive members
SELECT
mop.recipe_id,
mop.depth + 1 as depth,
ri.otherRecipe_id,
r.recipe_name
FROM menu_recipes r
JOIN menu_recipeItems ri
ON ri.otherRecipe_id = r.recipe_id
INNER JOIN MenuPrepOfPreps AS MOP
ON ri.otherRecipe_id = MOP.recipe_id
)
SELECT top(6)recipe_id, recipe_name
FROM MenuPrepOfPreps
GROUP BY recipe_id, recipe_name
</cfquery>
It keeps creating an infinite loop. When I limit the results to the first few rows (top 6), it does give the desired result.
It is possible that the design of the database is not correct so this might never work.
Any help is appreciated.
[UPDATED QUERY BASED ON #NewBie20200101 PROPOSED SOLUTION WITH CHANGES TO VARIABLE/COLUMN NAMES]
<cfquery name="whatever">
WITH MenuPrepOfPreps AS
(
SELECT otherrecipe_id,
CASE
when
otherRecipe_id = 0 then null
else
otherRecipe_id
end
as sub_recipe
FROM menu_recipeItems as a -- anchor
UNION ALL
SELECT
a.otherrecipe_id,
CASE
when
b.otherRecipe_id = 0 then null
else
b.otherRecipe_id
end
as sub_recipe
FROM menu_recipeItems as b
where b.recipe_id = a.otherRecipe_id --recursion
and a.otherRecipe_id is null --stopper
), allrecipeitems as (
SELECT recipe_id, sub_recipe
FROM MenuPrepOfPreps
)
Select
c.recipe_id,
d.otherRecipe_id
From MENU_recipes c
INNER JOIN MENU_recipeItems d on c.recipe_id = d.otherRecipe_id
Where c.recipe in (#recipelist#)
</cfquery>
Does not work and gives the following error:
The multi-part identifier "a.otherRecipe_id" could not be bound.

Answer 1

Not sure if this is gonna work:
With preppreprep as (
Select
Recipeid,
Case when otherrecipeId = 0 then null else otherrecipeID end as otherrecipeID, ——remove 0 it might be a problem
From
Recipeitems as a ——————anchor
Left outer join recipeitems as b on a.otherrecipeID = b.recipeID
Union all
Select
C.recipeid,
Case when c.otherrecipeID = 0 the null else c.other recipeID end as otherrecipeID, also remove 0
From preprepprep as c
Left outer join recipeitems as d
Where c.recipeid = d.otherrecipeID———--recursion
), allrecipeitems as (
Select
RecipeID,
OtherrecipeID
From preprepprep
)
Select
C.RecipeID
D.OtherRecipeID
From recipe c
Inner recipeitems d on c.recipeid = d.recipeid
Where c.recipe in (##)
——extract unpacked sub recipes based on recipe
If you think there are more than 99 levels, add option max recursion 0

Answer 2

I figured this out myself with help from all contributors (thank you). First off as was mentioned by #Charlieface I was approaching this from the wrong direction. With this solution I started from the bottom of the hierarchy and worked up until there is no more data to get in the hierarchy.
The last query has multiple filters and grouping in order to get exactly the information I need.
If anyone is interested, here's the solution:
WITH MenuPrepOfPreps (recipe_id, otherRecipe_id, depth, recipe_name)
AS (
SELECT
mria.recipe_id, mria.otherRecipe_id, 0, mr.recipe_name
FROM
menu_recipeItems mria, menu_recipes mr
where
mria.otherrecipe_id <> 0
and mria.recipe_id = mr.recipe_id
UNION ALL
SELECT
MenuPrepOfPreps.recipe_id, mri.otherrecipe_id, MenuPrepOfPreps.depth+1, mr.recipe_name
FROM
menu_recipeItems mri, MenuPrepOfPreps, menu_recipes mr
WHERE MenuPrepOfPreps.otherrecipe_id = mri.recipe_id
AND MRI.otherrecipe_id <> 0
AND mr.recipe_id = mri.recipe_id
)
SELECT
mopsA.otherrecipe_id, mopsA.recipe_id, mopsA.depth, mra.recipe_name AS thePrepRecipeName
FROM
MenuPrepOfPreps mopsA, MENU_recipes MRA
WHERE
mopsA.recipe_id in (#recipelist#)
AND mopsA.otherRecipe_id = MRA.recipe_id
AND MRA.recipeType_id = 2
GROUP BY mra.recipe_name, mopsA.otherrecipe_id, mopsA.recipe_id, mopsA.depth
I will have to see how this behaves with a larger dataset than the sample set I have now, but so far it's pretty fast.

Answer 3

I would create a table valued function that returns all leaf recipeItem for a single recipe_id (that flattens the hierarchy), without a recursion but a loop that resolves one recursion level at a time (instead of following each single branch). If you have a million sub recipes down to 10 levels deep it takes 12 iterations to get the data (10 (the depth) + 1 (determines nothing has changed anymore) + 1 (to get all the leaves from the recipes).
This approach has also the advantage that no limit must be set, if there is an endless loop defined in the database it does not bother us the slightest.
Here the code for the table valued function:
CREATE FUNCTION [dbo].[GetRecipeItemsFlat](#recipe_id INT)
RETURNS #recipeItems TABLE (recipeItem_id INT NOT NULL)
AS
BEGIN
DECLARE #recipeList TABLE (recipe_id INT);
DECLARE #previousCount INT = 0;
DECLARE #currentCount INT = 1;
-- Get all recipe_ids recursively until infinity (no endless loop possible)
INSERT INTO #recipeList SELECT #recipe_id;
WHILE (#previousCount < #currentCount) BEGIN
-- Adding not yet added child recipe_ids
INSERT INTO #recipeList SELECT DISTINCT otherRecipe_id FROM recipeItem WHERE recipe_id IN (SELECT recipe_id FROM #recipeList) AND (otherRecipe_id IS NOT NULL) AND (otherRecipe_id != 0) AND (otherRecipe_id NOT IN (SELECT recipe_id FROM #recipeList));
SET #previousCount = #currentCount ;
SET #currentCount = (SELECT COUNT(*) FROM #recipeList);
END
INSERT INTO #recipeItems SELECT recipeItem_ID FROM recipeItem WHERE recipe_id IN (SELECT * FROM #recipeList) AND (rawMaterial_id IS NOT NULL) AND (rawMaterial_id != 0);
RETURN;
END
And then I would create a view that provides all leaf recipeItem for a recipe like this:
CREATE VIEW recipeItemFlat
AS
SELECT c.recipe_id, c.recipeItem_id, c.ItemQuantity, c.rawMaterial_id
FROM recipe a
CROSS APPLY dbo.GetRecipeItemsFlat(recipe_id) b
INNER JOIN recipeItem c ON b.recipeItem_id = c.recipeItem_id;
And then the query to get all items for recipe 3047 is trivial:
SELECT * FROM recipeItemFlat WHERE recipe_id = 3047
There is one possible issue with this solution which I usually use for rights and roles resolutions and such things where it does not matter but here it could:
If there is a recipe A that has two recipes B and C and both B and C have a recipe D, then D is going to be only once in the result and adding up the quantities would give a wrong result. If that is relevant, don't use DISTINCT to determine the end of the loop in the table valued function but a maximum level like you did in your example.