Find parent rows that have identical sets of child rows

Question

Suppose I have a structure like this:

Recipes table

RecipeID
Name
Description

RecipeIngredients table

RecipeID
IngredientID
Quantity
UOM

The key on RecipeIngredients is (RecipeID, IngredientID).

What are some good ways for finding duplicate recipes? A duplicate recipe is defined as having the exact same set of ingredients and quantities for each ingredient.

I've thought of using FOR XML PATH to combine the ingredients into a single column. I haven't fully explored this but it should work if I make sure the ingredients/UOMs/quantities are sorted in the same sequence and have a proper separator. Are there better approaches?

There are 48K recipes and 200K ingredient rows.

Answer 1

For the following assumed schema and example data

CREATE TABLE dbo.RecipeIngredients
    (
      RecipeId INT NOT NULL ,
      IngredientID INT NOT NULL ,
      Quantity INT NOT NULL ,
      UOM INT NOT NULL ,
      CONSTRAINT RecipeIngredients_PK 
          PRIMARY KEY ( RecipeId, IngredientID ) WITH (IGNORE_DUP_KEY = ON)
    ) ;

INSERT INTO dbo.RecipeIngredients
SELECT TOP (210000) ABS(CRYPT_GEN_RANDOM(8)/50000),
                     ABS(CRYPT_GEN_RANDOM(8) % 100),
                     ABS(CRYPT_GEN_RANDOM(8) % 10),
                     ABS(CRYPT_GEN_RANDOM(8) % 5)
FROM master..spt_values v1,                     
     master..spt_values v2


SELECT DISTINCT RecipeId, 'X' AS Name
INTO Recipes 
FROM  dbo.RecipeIngredients 

This populated 205,009 ingredient rows and 42,613 recipes. This will be slightly different each time due to the random element.

It assumes relatively few dupes (output after an example run was 217 duplicate recipe groups with two or three recipes per group). The most pathological case based on the figures in the OP would be 48,000 exact duplicates.

A script to set that up is

DROP TABLE dbo.RecipeIngredients,Recipes
GO

CREATE TABLE Recipes(
RecipeId INT IDENTITY,
Name VARCHAR(1))

INSERT INTO Recipes 
SELECT TOP 48000 'X'
FROM master..spt_values v1,                     
     master..spt_values v2

CREATE TABLE dbo.RecipeIngredients
    (
      RecipeId INT NOT NULL ,
      IngredientID INT NOT NULL ,
      Quantity INT NOT NULL ,
      UOM INT NOT NULL ,
      CONSTRAINT RecipeIngredients_PK 
          PRIMARY KEY ( RecipeId, IngredientID )) ;

INSERT INTO dbo.RecipeIngredients
SELECT RecipeId,IngredientID,Quantity,UOM
FROM Recipes
CROSS JOIN (SELECT 1,1,1 UNION ALL SELECT 2,2,2 UNION ALL  SELECT 3,3,3 UNION ALL SELECT 4,4,4) I(IngredientID,Quantity,UOM)

The following completed in less than a second on my machine for both cases.

CREATE TABLE #Concat
  (
     RecipeId     INT,
     concatenated VARCHAR(8000),
     PRIMARY KEY (concatenated, RecipeId)
  )

INSERT INTO #Concat
SELECT R.RecipeId,
       ISNULL(concatenated, '')
FROM   Recipes R
       CROSS APPLY (SELECT CAST(IngredientID AS VARCHAR(10)) + ',' + CAST(Quantity AS VARCHAR(10)) + ',' + CAST(UOM AS VARCHAR(10)) + ','
                    FROM   dbo.RecipeIngredients RI
                    WHERE  R.RecipeId = RecipeId
                    ORDER  BY IngredientID
                    FOR XML PATH('')) X (concatenated);

WITH C1
     AS (SELECT DISTINCT concatenated
         FROM   #Concat)
SELECT STUFF(Recipes, 1, 1, '')
FROM   C1
       CROSS APPLY (SELECT ',' + CAST(RecipeId AS VARCHAR(10))
                    FROM   #Concat C2
                    WHERE  C1.concatenated = C2.concatenated
                    ORDER  BY RecipeId
                    FOR XML PATH('')) R(Recipes)
WHERE  Recipes LIKE '%,%,%'

DROP TABLE #Concat 

One caveat

I assumed that the length of the concatenated string will not exceed 896 bytes. If it does this will raise an error at run time rather than silently failing. You will need to remove the primary key (and implicitly created index) from the #temp table. The maximum length of the concatenated string in my test setup was 125 characters.

If the concatenated string is too long to index then performance of the final XML PATH query consolidating the identical recipes could well be poor. Installing and using a custom CLR string aggregation would be one solution as that could do the concatenation with one pass of the data rather than a non indexed self join.

SELECT YourClrAggregate(RecipeId)
FROM #Concat
GROUP BY concatenated

I also tried

WITH Agg
     AS (SELECT RecipeId,
                MAX(IngredientID)          AS MaxIngredientID,
                MIN(IngredientID)          AS MinIngredientID,
                SUM(IngredientID)          AS SumIngredientID,
                COUNT(IngredientID)        AS CountIngredientID,
                CHECKSUM_AGG(IngredientID) AS ChkIngredientID,
                MAX(Quantity)              AS MaxQuantity,
                MIN(Quantity)              AS MinQuantity,
                SUM(Quantity)              AS SumQuantity,
                COUNT(Quantity)            AS CountQuantity,
                CHECKSUM_AGG(Quantity)     AS ChkQuantity,
                MAX(UOM)                   AS MaxUOM,
                MIN(UOM)                   AS MinUOM,
                SUM(UOM)                   AS SumUOM,
                COUNT(UOM)                 AS CountUOM,
                CHECKSUM_AGG(UOM)          AS ChkUOM
         FROM   dbo.RecipeIngredients
         GROUP  BY RecipeId)
SELECT  A1.RecipeId AS RecipeId1,
        A2.RecipeId AS RecipeId2
FROM   Agg A1
       JOIN Agg A2
         ON A1.MaxIngredientID = A2.MaxIngredientID
            AND A1.MinIngredientID = A2.MinIngredientID
            AND A1.SumIngredientID = A2.SumIngredientID
            AND A1.CountIngredientID = A2.CountIngredientID
            AND A1.ChkIngredientID = A2.ChkIngredientID
            AND A1.MaxQuantity = A2.MaxQuantity
            AND A1.MinQuantity = A2.MinQuantity
            AND A1.SumQuantity = A2.SumQuantity
            AND A1.CountQuantity = A2.CountQuantity
            AND A1.ChkQuantity = A2.ChkQuantity
            AND A1.MaxUOM = A2.MaxUOM
            AND A1.MinUOM = A2.MinUOM
            AND A1.SumUOM = A2.SumUOM
            AND A1.CountUOM = A2.CountUOM
            AND A1.ChkUOM = A2.ChkUOM
            AND A1.RecipeId <> A2.RecipeId
WHERE  NOT EXISTS (SELECT *
                   FROM   (SELECT *
                           FROM   RecipeIngredients
                           WHERE  RecipeId = A1.RecipeId) R1
                          FULL OUTER JOIN (SELECT *
                                           FROM   RecipeIngredients
                                           WHERE  RecipeId = A2.RecipeId) R2
                            ON R1.IngredientID = R2.IngredientID
                               AND R1.Quantity = R2.Quantity
                               AND R1.UOM = R2.UOM
                   WHERE  R1.RecipeId IS NULL
                           OR R2.RecipeId IS NULL) 

This works acceptably when there are relatively few duplicates (less than a second for the first example data) but performs badly in the pathological case as the initial aggregation returns exactly the same results for every RecipeID and so doesn't manage to cut down the number of comparisons at all.

Answer 2

This is a generalization of the relational division problem. No idea how efficient this will be:

; WITH cte AS
( SELECT RecipeID_1 = r1.RecipeID, Name_1 = r1.Name,
         RecipeID_2 = r2.RecipeID, Name_2 = r2.Name  
  FROM Recipes AS r1
    JOIN Recipes AS r2
      ON r1.RecipeID <> r2.RecipeID
  WHERE NOT EXISTS
        ( SELECT 1
          FROM RecipeIngredients AS ri1
          WHERE ri1.RecipeID = r1.RecipeID 
            AND NOT EXISTS
                ( SELECT 1
                  FROM RecipeIngredients AS ri2
                  WHERE ri2.RecipeID = r2.RecipeID 
                    AND ri1.IngredientID = ri2.IngredientID
                    AND ri1.Quantity = ri2.Quantity
                    AND ri1.UOM = ri2.UOM
                )
         )
)
SELECT c1.*
FROM cte AS c1
  JOIN cte AS c2
    ON  c1.RecipeID_1 = c2.RecipeID_2
    AND c1.RecipeID_2 = c2.RecipeID_1
    AND c1.RecipeID_1 < c1.RecipeID_2;

Another (similar) approach:

SELECT RecipeID_1 = r1.RecipeID, Name_1 = r1.Name,
       RecipeID_2 = r2.RecipeID, Name_2 = r2.Name 
FROM Recipes AS r1
  JOIN Recipes AS r2
    ON  r1.RecipeID < r2.RecipeID 
    AND NOT EXISTS
        ( SELECT IngredientID, Quantity, UOM
          FROM RecipeIngredients AS ri1
          WHERE ri1.RecipeID = r1.RecipeID
        EXCEPT 
          SELECT IngredientID, Quantity, UOM
          FROM RecipeIngredients AS ri2
          WHERE ri2.RecipeID = r2.RecipeID
        )
    AND NOT EXISTS
        ( SELECT IngredientID, Quantity, UOM
          FROM RecipeIngredients AS ri2
          WHERE ri2.RecipeID = r2.RecipeID
        EXCEPT 
          SELECT IngredientID, Quantity, UOM
          FROM RecipeIngredients AS ri1
          WHERE ri1.RecipeID = r1.RecipeID
        ) ;

And another, different one:

; WITH cte AS
( SELECT RecipeID_1 = r.RecipeID, RecipeID_2 = ri.RecipeID, 
          ri.IngredientID, ri.Quantity, ri.UOM
  FROM Recipes AS r
    CROSS JOIN RecipeIngredients AS ri
)
, cte2 AS
( SELECT RecipeID_1, RecipeID_2,
         IngredientID, Quantity, UOM
  FROM cte
EXCEPT
  SELECT RecipeID_2, RecipeID_1,
         IngredientID, Quantity, UOM
  FROM cte
)

  SELECT RecipeID_1 = r1.RecipeID, RecipeID_2 = r2.RecipeID
  FROM Recipes AS r1
    JOIN Recipes AS r2
      ON r1.RecipeID < r2.RecipeID
EXCEPT 
  SELECT RecipeID_1, RecipeID_2
  FROM cte2
EXCEPT 
  SELECT RecipeID_2, RecipeID_1
  FROM cte2 ;

Tested at SQL-Fiddle

---

Using the CHECKSUM() and CHECKSUM_AGG() functions, test at SQL-Fiddle-2:
(ignore this as it may give false positives)

ALTER TABLE RecipeIngredients
  ADD ck AS CHECKSUM( IngredientID, Quantity, UOM )
    PERSISTED ;

CREATE INDEX ckecksum_IX
  ON RecipeIngredients
    ( RecipeID, ck ) ;

; WITH cte AS
( SELECT RecipeID,
         cka = CHECKSUM_AGG(ck)
  FROM RecipeIngredients AS ri
  GROUP BY RecipeID
)
SELECT RecipeID_1 = c1.RecipeID, RecipeID_2 = c2.RecipeID
FROM cte AS c1
  JOIN cte AS c2
    ON  c1.cka = c2.cka
    AND c1.RecipeID < c2.RecipeID  ;