我昨天回答了一个递归的CTE,它暴露了SQL Server中实现这些问题的方式的问题(也可能在其他RDBMS中?)。基本上,当我尝试对当前的递归级别使用ROW_NUMBER
时,它会针对当前递归级别的每一行子集运行。我希望这可以在真正的SET逻辑中工作,并且针对整个当前递归级别运行。
看来from this MSDN article,我发现的问题是预期的功能:
CTE的递归部分中的分析和聚合函数是 应用于当前递归级别的集合而不是集合 对于CTE。 像ROW_NUMBER这样的函数只对子集有效 数据通过当前递归级别而不是整个传递给它们 数据集用于CTE的递归部分。更多 信息,参见J.在递归CTE中使用分析函数。
在我的挖掘中,我找不到任何地方可以解释为什么选择按照它的方式工作?这在基于集合的语言中更像是一种过程方法,因此这对我的SQL思维过程起作用,在我看来非常混乱。 是否有人知道和/或任何人都可以解释为什么递归CTE以递归方式处理递归级别的分析函数?
以下是帮助可视化的代码:
注意,每个代码输出中的RowNumber
列。
Here is the SQLFiddle for the CTE (only showing the 2nd level of the recursion)
WITH myCTE
AS
(
SELECT *, ROW_NUMBER() OVER (ORDER BY Score desc) AS RowNumber, 1 AS RecurseLevel
FROM tblGroups
WHERE ParentId IS NULL
UNION ALL
SELECT tblGroups.*,
ROW_NUMBER() OVER (ORDER BY myCTE.RowNumber , tblGroups.Score desc) AS RowNumber,
RecurseLevel + 1 AS RecurseLevel
FROM tblGroups
JOIN myCTE
ON myCTE.GroupID = tblGroups.ParentID
)
SELECT *
FROM myCTE
WHERE RecurseLevel = 2;
WITH myCTE
AS
(
SELECT *, ROW_NUMBER() OVER (ORDER BY Score desc) AS RowNumber, 1 AS RecurseLevel
FROM tblGroups
WHERE ParentId IS NULL
)
SELECT tblGroups.*,
ROW_NUMBER() OVER (ORDER BY myCTE.RowNumber , tblGroups.Score desc) AS RowNumber,
RecurseLevel + 1 AS RecurseLevel
FROM tblGroups
JOIN myCTE
ON myCTE.GroupID = tblGroups.ParentID;
我总是设想SQL递归CTE更像this while loop
DECLARE @RecursionLevel INT
SET @RecursionLevel = 0
SELECT *, ROW_NUMBER() OVER (ORDER BY Score desc) AS RowNumber, @RecursionLevel AS recurseLevel
INTO #RecursiveTable
FROM tblGroups
WHERE ParentId IS NULL
WHILE EXISTS( SELECT tblGroups.* FROM tblGroups JOIN #RecursiveTable ON #RecursiveTable.GroupID = tblGroups.ParentID WHERE recurseLevel = @RecursionLevel)
BEGIN
INSERT INTO #RecursiveTable
SELECT tblGroups.*,
ROW_NUMBER() OVER (ORDER BY #RecursiveTable.RowNumber , tblGroups.Score desc) AS RowNumber,
recurseLevel + 1 AS recurseLevel
FROM tblGroups
JOIN #RecursiveTable
ON #RecursiveTable.GroupID = tblGroups.ParentID
WHERE recurseLevel = @RecursionLevel
SET @RecursionLevel = @RecursionLevel + 1
END
SELECT * FROM #RecursiveTable ORDER BY RecurseLevel;
答案 0 :(得分:1)
分析函数在需要已知结果集来解析时非常特殊。 它们依赖于以下,前置或完整结果集来计算当前值。 也就是说,在包含分析函数的视图上永远不允许合并视图。为什么? 这将改变结果。
例如:
Select * from (
select row_number() over (partition by c1 order by c2) rw, c3 from t) z
where c3=123
与
不同 select row_number() over (partition by c1 order by c2) rw, c3 from t
where c3=123
这两个将返回rw的不同值。 这就是为什么包含分析函数的子查询将始终完全解析,并且永远不会与其余部分合并。
<强>更新强>
查看第二个查询:
WITH myCTE
AS
(
SELECT *, ROW_NUMBER() OVER (ORDER BY Score desc) AS RowNumber, 1 AS RecurseLevel
FROM tblGroups
WHERE ParentId IS NULL
)
SELECT tblGroups.*,
ROW_NUMBER() OVER (ORDER BY myCTE.RowNumber , tblGroups.Score desc) AS RowNumber,
RecurseLevel + 1 AS RecurseLevel
FROM tblGroups
JOIN myCTE
ON myCTE.GroupID = tblGroups.ParentID;
它的工作原理就像是(相同的执行计划和结果):
SELECT tblGroups.*,
ROW_NUMBER() OVER (ORDER BY myCTE.RowNumber , tblGroups.Score desc) AS RowNumber,
RecurseLevel + 1 AS RecurseLevel
FROM tblGroups
JOIN (
SELECT *, ROW_NUMBER() OVER (ORDER BY Score desc) AS RowNumber, 1 AS RecurseLevel
FROM tblGroups
WHERE ParentId IS NULL
)myCTE ON myCTE.GroupID = tblGroups.ParentID;
需要对此进行分区以重置rownumber。
递归查询在while循环中不起作用,它们不是程序性的。在基础上,它们像递归函数一样工作,但根据表,查询,索引,它们可以被优化为以某种方式运行。
如果我们遵循以下概念:在使用分析函数时查看无法合并,并查看查询1.它只能运行一次,并且它是嵌套循环。
WITH myCTE
AS
( /*Cannot be merged*/
SELECT *, ROW_NUMBER() OVER (ORDER BY Score desc) AS RowNumber, 1 AS RecurseLevel,
cast(0 as bigint) n
FROM tblGroups
WHERE ParentId IS NULL
UNION ALL
/*Cannot be merged*/
SELECT tblGroups.*,
ROW_NUMBER() OVER (ORDER BY myCTE.RowNumber, tblGroups.Score desc) AS RowNumber, RecurseLevel + 1 AS RecurseLevel,
myCTE.RowNumber
FROM tblGroups
JOIN myCTE
ON myCTE.GroupID = tblGroups.ParentID
)
SELECT *
FROM myCTE;
所以第一次选择,不能合并第二次,也不是。运行此查询的唯一方法是在每个级别返回的每个项目的嵌套循环中,因此重置。同样,这不是程序问题,也不是可能的执行计划问题。
希望这能回答你的问题,如果没有,请告诉我。)
ý