Question

以下SQL根据它们的关系分隔表。问题在于在3000系列下排序的表格。作为外键的一部分并使用外键的表。任何人都有一些聪明的递归CTE或一个存储过程来进行必要的排序？程序连接到数据库不被视为解决方案。

编辑：我根据第一个解决方案在“答案”中发布了答案任何重新发布我自己的“正确”答案的人都可以获得免费的“正确答案”！

WITH 
 AllTables(TableName) AS
 (
 SELECT OBJECT_SCHEMA_NAME(so.id) +'.'+ OBJECT_NAME(so.id) 
 FROM dbo.sysobjects so 
 INNER JOIN sys.all_columns ac ON 
  so.ID = ac.object_id
 WHERE
  so.type = 'U'
 AND
  ac.is_rowguidcol = 1
 ),

  Relationships(ReferenceTableName, ReferenceColumnName, TableName, ColumnName)  AS
  (
  SELECT  
    OBJECT_SCHEMA_NAME (fkey.referenced_object_id) + '.' +  
    OBJECT_NAME (fkey.referenced_object_id) AS ReferenceTableName 
    ,COL_NAME(fcol.referenced_object_id, 
              fcol.referenced_column_id) AS ReferenceColumnName 
    ,OBJECT_SCHEMA_NAME (fkey.parent_object_id) + '.' +  
    OBJECT_NAME(fkey.parent_object_id) AS TableName 
    ,COL_NAME(fcol.parent_object_id, fcol.parent_column_id) AS ColumnName 
  FROM sys.foreign_keys AS fkey 
    INNER JOIN sys.foreign_key_columns AS fcol ON 
               fkey.OBJECT_ID = fcol.constraint_object_id 
  ),

 NotReferencedOrReferencing(TableName) AS
 (
 SELECT TableName FROM AllTables
 EXCEPT
 SELECT TableName FROM Relationships
 EXCEPT
 SELECT ReferenceTableName FROM Relationships
 ),

 OnlyReferenced(Tablename) AS
 (
 SELECT ReferenceTableName FROM Relationships
 EXCEPT
 SELECT TableName FROM Relationships
 ),
-- These need to be sorted based on theire internal relationships
 ReferencedAndReferencing(TableName, ReferenceTableName) AS
 (
 SELECT r1.Tablename, r2.ReferenceTableName FROM Relationships r1 
 INNER JOIN Relationships r2
 ON r1.TableName = r2.ReferenceTableName   
 ),

 OnlyReferencing(TableName) AS
 (
 SELECT Tablename FROM Relationships
 EXCEPT
 SELECT ReferenceTablename FROM Relationships
 )
SELECT TableName, 1000 AS Sorting FROM NotReferencedOrReferencing
UNION
SELECT TableName, 2000 AS Sorting FROM OnlyReferenced 
UNION
SELECT TableName, 3000 AS Sorting FROM ReferencedAndReferencing
UNION
SELECT TableName, 4000 AS Sorting FROM OnlyReferencing
ORDER BY Sorting

Answer 1

我的调整是适度的调整：这个是SQL-2005 +，适用于没有“rowguidcol”的数据库：

WITH TablesCTE(SchemaName, TableName, TableID, Ordinal) AS
(
    SELECT
        OBJECT_SCHEMA_NAME(so.object_id) AS SchemaName,
        OBJECT_NAME(so.object_id) AS TableName,
        so.object_id AS TableID,
        0 AS Ordinal
    FROM
        sys.objects AS so
    WHERE
        so.type = 'U'
        AND so.is_ms_Shipped = 0
    UNION ALL
    SELECT
        OBJECT_SCHEMA_NAME(so.object_id) AS SchemaName,
        OBJECT_NAME(so.object_id) AS TableName,
        so.object_id AS TableID,
        tt.Ordinal + 1 AS Ordinal
    FROM
        sys.objects AS so
    INNER JOIN sys.foreign_keys AS f
        ON f.parent_object_id = so.object_id
        AND f.parent_object_id != f.referenced_object_id
    INNER JOIN TablesCTE AS tt
        ON f.referenced_object_id = tt.TableID
    WHERE
        so.type = 'U'
        AND so.is_ms_Shipped = 0
)

SELECT DISTINCT
        t.Ordinal,
        t.SchemaName,
        t.TableName,
        t.TableID
    FROM
        TablesCTE AS t
    INNER JOIN
        (
            SELECT
                itt.SchemaName as SchemaName,
                itt.TableName as TableName,
                itt.TableID as TableID,
                Max(itt.Ordinal) as Ordinal
            FROM
                TablesCTE AS itt
            GROUP BY
                itt.SchemaName,
                itt.TableName,
                itt.TableID
        ) AS tt
        ON t.TableID = tt.TableID
        AND t.Ordinal = tt.Ordinal
ORDER BY
    t.Ordinal,
    t.TableName

Answer 2

感谢您使用NXC的解决方案。你让我在正确的轨道上使用递归CTE解决问题。

WITH 
  TablesCTE(TableName, TableID, Ordinal) AS
  (
  SELECT 
    OBJECT_SCHEMA_NAME(so.id) +'.'+ OBJECT_NAME(so.id) AS TableName,
    so.id AS TableID,
    0 AS Ordinal
  FROM dbo.sysobjects so INNER JOIN sys.all_columns ac ON so.ID = ac.object_id
  WHERE
    so.type = 'U'
  AND
    ac.is_rowguidcol = 1
  UNION ALL
  SELECT 
    OBJECT_SCHEMA_NAME(so.id) +'.'+ OBJECT_NAME(so.id) AS TableName,
    so.id AS TableID,
    tt.Ordinal + 1 AS Ordinal
  FROM 
    dbo.sysobjects so 
    INNER JOIN sys.all_columns ac ON so.ID = ac.object_id
    INNER JOIN sys.foreign_keys f 
      ON (f.parent_object_id = so.id AND f.parent_object_id != f.referenced_object_id)
    INNER JOIN TablesCTE tt ON f.referenced_object_id = tt.TableID
  WHERE
    so.type = 'U'
  AND
    ac.is_rowguidcol = 1
)  
SELECT DISTINCT 
  t.Ordinal,
  t.TableName
  FROM TablesCTE t
  INNER JOIN 
    (
    SELECT 
      TableName as TableName,
      Max (Ordinal) as Ordinal
    FROM TablesCTE
    GROUP BY TableName
    ) tt ON (t.TableName = tt.TableName  AND t.Ordinal = tt.Ordinal)
ORDER BY t.Ordinal, t.TableName

对于知道什么是可用的东西：我将使用它来安全地清空数据库而不违反任何外键关系。（按降序截断）通过按升序填写表格，我还可以安全地用其他数据库中的数据填充表格。

Answer 3

您可以使用迭代算法，该算法可能比CTE更少复杂。这是一个根据深度排序的例子：

declare @level int      -- Current depth
       ,@count int      

-- Step 1: Start with tables that have no FK dependencies
--  
if object_id ('tempdb..#Tables') is not null
    drop table #Tables

select s.name + '.' + t.name  as TableName
      ,t.object_id            as TableID
      ,0                      as Ordinal
  into #Tables
  from sys.tables t
  join sys.schemas s
    on t.schema_id = s.schema_id
 where not exists
       (select 1
          from sys.foreign_keys f
         where f.parent_object_id = t.object_id)

set @count = @@rowcount         
set @level = 0


-- Step 2: For a given depth this finds tables joined to 
-- tables at this given depth.  A table can live at multiple 
-- depths if it has more than one join path into it, so we 
-- filter these out in step 3 at the end.
--
while @count > 0 begin

    insert #Tables (
           TableName
          ,TableID
          ,Ordinal
    ) 
    select s.name + '.' + t.name  as TableName
          ,t.object_id            as TableID
          ,@level + 1             as Ordinal
      from sys.tables t
      join sys.schemas s
        on s.schema_id = t.schema_id
     where exists
           (select 1
              from sys.foreign_keys f
              join #Tables tt
                on f.referenced_object_id = tt.TableID
               and tt.Ordinal = @level
               and f.parent_object_id = t.object_id
               and f.parent_object_id != f.referenced_object_id)
                   -- The last line ignores self-joins.  You'll
                   -- need to deal with these separately

   set @count = @@rowcount
   set @level = @level + 1
end

-- Step 3: This filters out the maximum depth an object occurs at
-- and displays the deepest first.
--
select t.Ordinal
      ,t.TableID
      ,t.TableName
  from #Tables t
  join (select TableName     as TableName
              ,Max (Ordinal) as Ordinal
          from #Tables
         group by TableName) tt
    on t.TableName = tt.TableName
   and t.Ordinal = tt.Ordinal
 order by t.Ordinal desc

Answer 4

这会导致自引用表出现问题。您需要手动排除任何指向自引用表的外键。

INNER JOIN sys.foreign_keys  f 
  ON (f.parent_object_id = so.id AND f.parent_object_id != f.referenced_object_id)

  /* Manually exclude self-referencing tables - they cause recursion problems*/      
    and f.object_id  not in /*Below are IDs of foreign keys*/
        ( 1847729685,   
        1863729742,     
        1879729799      
        )   
INNER JOIN TablesCTE tt

SQLServer：如何对按外键依赖关系排序的表名进行排序

4 个答案: