Help with generating a report from data in a paren

问题:

I need help with a problem regarding data saved in a parent-children model table and a report I need to build upon it. I've already tried searching for topics about parent-children issues, but I couldn't find anything useful in my scenario.

What I have

A Microsoft SQL Server 2000 database server.

A categories table, which has four columns: category_id, category_name, father_id and visible; the categories have x root categories (where x is variable), and could be y level deep (where y is variable), if a category is a root level one it has father_id null otherwise it's filled with the id of the father category.

A sales table, which has z columns, one of which is category_id, a foreign key to categories.category_id; a sale must always have a category, and it could be linked anywhere in the aforementioned y level.

What I need

I've been asked a report displaying only the root (first level) categories, and the quantity of sales belongings to each of these, or their children, no matter how deep. I.e. if one of the root categories is food, which has a children category named fruit, which has a children category named apple, I need to count every item belonging to food or fruit or apple.

Couldn't you use the nested set data model?

I know of the nested set model, but I already have the table this way, and migrating it to the nested set model would be a pain (let alone I didn't even fully grasp how nested set works), not counting the changes needed in the application using the database. (If someone thinks this is still the least pain way, please explain why and how the current data could be migrated.)

Couldn't you use CTE (Common Table Expressions)?

No, it's a Microsoft SQL Server 2000, and Common Table Expressions are introduced in the 2005 edition.

Thanks in advance, Andrea.

回答1:

SQL 2000 Based solution

DECLARE @Stack TABLE (
  StackID INTEGER IDENTITY
  , Category VARCHAR(20)
  , RootID INTEGER
  , ChildID INTEGER
  , Visited BIT)

INSERT INTO @Stack
SELECT  [Category] = c.category_name
        , [RootID] = c.category_id
        , [ChildID] = c.category_id
        , 0
FROM    Categories c

WHILE EXISTS (SELECT * FROM @Stack WHERE Visited = 0)
BEGIN
  DECLARE @StackID INTEGER
  SELECT  @StackID = MAX(StackID) FROM    @Stack

  INSERT INTO @Stack
  SELECT  st.Category
          , st.RootID
          , c.category_id
          , 0
  FROM    @Stack st
          INNER JOIN Categories c ON c.father_id = st.ChildID  
  WHERE   Visited = 0

  UPDATE  @Stack
  SET     Visited = 1
  WHERE   StackID <= @StackID
END

SELECT  st.RootID
        , st.Category
        , COUNT(s.sales_id)
FROM    @Stack st
        INNER JOIN Sales s ON s.category_id = st.ChildID
GROUP BY st.RootID, st.Category
ORDER BY st.RootID

SQL 2005 Based solution

A CTE should get you what you want

• Select each category from Categories to be the root item
• recursively add each child of every root item
• INNER JOIN the results with your sales table. As every root is in the result of the CTE, a simple GROUP BY is sufficient to get a count for each item.

SQL Statement

;WITH QtyCTE AS (
  SELECT  [Category] = c.category_name
          , [RootID] = c.category_id
          , [ChildID] = c.category_id
  FROM    Categories c
  UNION ALL 
  SELECT  cte.Category
          , cte.RootID
          , c.category_id
  FROM    QtyCTE cte
          INNER JOIN Categories c ON c.father_id = cte.ChildID
)
SELECT  cte.RootID
        , cte.Category
        , COUNT(s.sales_id)
FROM    QtyCTE cte
        INNER JOIN Sales s ON s.category_id = cte.ChildID
GROUP BY cte.RootID, cte.Category
ORDER BY cte.RootID

回答2:

Something like this?

CREATE TABLE #SingleLevelCategoryCounts
{ 
    category_id,
    count,
    root_id
}

CREATE TABLE #ProcessedCategories
{
    category_id,
    root_id
}

CREATE TABLE #TotalTopLevelCategoryCounts
{ 
    category_id,
    count
}

INSERT INTO #SingleLevelCategoryCounts
SELECT 
    category_id, SUM(*), category_id
FROM
    Categories 
    INNER JOIN Sales ON Categories.category_id = sales.category_id
WHERE
     Categories.father_id IS NULL
GROUP BY
     Categories.category_id


WHILE EXISTS (SELECT * FROM #SingleLevelCategoryCounts)
BEGIN
   IF NOT EXISTS(SELECT * FROM #TopLevelCategoryCounts)
   BEGIN
      INSERT INTO #TopLevelCategoryCounts
      SELECT
          root_id, count
      FROM
          #SingleLevelCategoryCounts
   END 
   ELSE
   BEGIN
       UPDATE top
       SET
          top.count = top.count + level.count
       FROM
          #TopLevelCategoryCounts top
          INNER JOIN #SingleLevelCategoryCounts level ON top.category_id = level.count
   END

   INSERT INTO #ProcessedCategories
   SELECT category_id, root_id FROM #SingleLevelCategoryCounts

   DELETE #SingleLevelCategoryCounts

   INSERT INTO #SingleLevelCategoryCounts
   SELECT 
        category_id, SUM(*), pc.root_id
   FROM
        Categories 
        INNER JOIN Sales ON Categories.category_id = sales.category_id
        INNER JOIN #ProcessedCategories pc ON Categories.father_id = pc.category_id
   WHERE
         Categories.category_id NOT IN
         (
             SELECT category_id in #ProcessedCategories
         )
   GROUP BY
         Categories.category_id


END