我有一个树形分层结构代表我的产品结构。
在每个产品级别（6个级别），我都有一个销售价格链接到它。我正在使用两个相互连接的桌子将较低级别的价格与上面的价格联系起来。
我想这样做，这样我就不会考虑更多的价格了不止一次。这是通过以下代码完成的（注意我只使用0级，1级和2级来表示想法）：
SELECT L0_SALESPRICE
,L1_SALESPRICE
,L2_SALESPRICE
FROM
(SELECT DISTINCT A.*
FROM BCT A
JOIN QuotationLine QL ON A.PRICECALCID = QL.PRICECALCID
WHERE A.Levels = 0) AS L0
JOIN
(SELECT DISTINCT A.*
FROM BCT A
JOIN QuotationLine QL ON A.PRICECALCID = QL.PRICECALCID
WHERE A.Levels = 1) AS L1 ON L0.ItemId = L1.ParentItemId
JOIN
(SELECT DISTINCT A.*
FROM BCT A
JOIN QuotationLine QL ON A.PRICECALCID = QL.PRICECALCID
WHERE A.Levels = 2) AS L2 ON L1.ItemId = L2.ParentItemId
问题是查询永远不会执行，我得到内存不足错误。
表BCT是750 000行，表QuotationLine是22000行。
任何建议表示赞赏。
答案 0 :(得分：0)
为了演示如何解决此问题，以下是一些示例表定义。
CREATE TABLE [Description]
(
[DescriptionId] INT IDENTITY NOT NULL CONSTRAINT [PK_Description] PRIMARY KEY,
[DescriptionText] NVARCHAR(50) NOT NULL
)
GO
CREATE TABLE [Hierarchy]
(
[HierarchyId] INT NOT NULL CONSTRAINT [PK_Hierarchy] PRIMARY KEY,
[ParentHierarchyId] INT NULL CONSTRAINT [FK_Hierarchy_ParentHierarchyId] REFERENCES [Hierarchy] ([HierarchyId]) ON DELETE NO ACTION ON UPDATE NO ACTION,
[Price] MONEY NOT NULL,
[DescriptionId] INT NULL CONSTRAINT [FK_Hierarchy_Description] REFERENCES [Description] ([DescriptionId]) ON DELETE SET NULL ON UPDATE CASCADE
)
GO
CREATE INDEX [IX_Hierarchy_ParentHierarchyId] ON [Hierarchy]
([ParentHierarchyId]) INCLUDE ([HierarchyId], [Price], [DescriptionId])
GO
获得层次结构的一种天真的方式 - 即不太可能解决您的性能问题的方法 - 可能是：
;WITH RowEnds AS
(
SELECT h.[HierarchyId], h.[ParentHierarchyId], h.[Price], h.[DescriptionId], h.[HierarchyId] AS [RowEndHierarchyId], 0 AS [ReverseLevel]
FROM [Hierarchy] h
WHERE NOT EXISTS (SELECT 1 FROM [Hierarchy] i WHERE i.[ParentHierarchyId] = h.[HierarchyId])
UNION ALL
SELECT h.[HierarchyId], h.[ParentHierarchyId], h.[Price], h.[DescriptionId], r.[RowEndHierarchyId], r.[ReverseLevel] + 1 AS [ReverseLevel]
FROM [Hierarchy] h
INNER JOIN RowEnds r ON h.[HierarchyId] = r.[ParentHierarchyId]
),
InOrder AS
(
SELECT r.RowEndHierarchyId, r.[HierarchyId], r.Price, d.DescriptionText, RANK() OVER (PARTITION BY r.[RowEndHierarchyId] ORDER BY r.[ReverseLevel] DESC) AS [Level]
FROM RowEnds r
LEFT JOIN [Description] d ON r.DescriptionId = d.DescriptionId
)
SELECT DISTINCT o.RowEndHierarchyId, p.[1] AS Price1, d.[1] AS Description1, p.[2] AS Price2, d.[2] AS Description2, p.[3] AS Price3, d.[3] AS Description3,
p.[4] AS Price4, d.[4] AS Description4, p.[5] AS Price5, d.[5] AS Description5, p.[6] AS Price6, d.[6] AS Description6,
p.[7] AS Price7, d.[7] AS Description7
FROM InOrder o
INNER JOIN
(SELECT projp.RowEndHierarchyId, projp.[Level], projp.[Price]
FROM InOrder projp) ppre
PIVOT (MIN([Price]) FOR [Level] IN ([1], [2], [3], [4], [5], [6], [7])) p
ON o.RowEndHierarchyId = p.RowEndHierarchyId
LEFT JOIN
(SELECT projd.RowEndHierarchyId, projd.[Level], projd.DescriptionText
FROM INOrder projd) dpre
PIVOT (MIN(DescriptionText) FOR [Level] IN ([1], [2], [3], [4], [5], [6], [7])) d
ON o.RowEndHierarchyId = d.RowEndHierarchyId
ORDER BY o.RowEndHierarchyId
当然，此示例使用递归公用表表达式来获取层次结构。查询采用相反的方法，而不是从树的根开始并朝向叶子工作。这样做的好处是输出中的每一行都对应于树中的叶节点。
但是，使用此方法，性能可能仍然不能令人满意，因为您没有机会索引公共表表达式，其结果集可能非常大。如果tempdb具有足够的空间和性能，则以下更详细的查询可以提高性能。
CREATE TABLE #RowEnd
(
[RowEndHierarchyId] INT NOT NULL,
[HierarchyId] INT NOT NULL,
[ParentHierarchyId] INT NULL,
[Price] MONEY NOT NULL,
[DescriptionId] INT NULL,
[ReverseLevel] INT NOT NULL,
PRIMARY KEY ([RowEndHierarchyId], [ReverseLevel] DESC)
)
CREATE INDEX [IX_RowEnd_ParentHierarchyId] ON #RowEnd
([ParentHierarchyId], [RowEndHierarchyId], [ReverseLevel])
CREATE INDEX [IX_RowEnd_ReverseLevel] ON #RowEnd
([ReverseLevel] DESC, [ParentHierarchyId], [RowEndHierarchyId])
INSERT #RowEnd ([HierarchyId], [ParentHierarchyId], [Price], [DescriptionId], [RowEndHierarchyId], [ReverseLevel])
SELECT h.[HierarchyId], h.[ParentHierarchyId], h.[Price], h.[DescriptionId], h.[HierarchyId], 1
FROM [Hierarchy] h
WHERE NOT EXISTS (SELECT 1 FROM [Hierarchy] i WHERE i.ParentHierarchyId = h.[HierarchyId])
DECLARE @ReverseLevel INT
SET @ReverseLevel = 0
WHILE EXISTS (SELECT 1 FROM #RowEnd re WHERE re.ReverseLevel > @ReverseLevel)
BEGIN
SET @ReverseLevel = @ReverseLevel + 1
INSERT #RowEnd ([HierarchyId], [ParentHierarchyId], [Price], [DescriptionId], [RowEndHierarchyId], [ReverseLevel])
SELECT h.[HierarchyId], h.[ParentHierarchyId], h.[Price], h.[DescriptionId], re.[RowEndHierarchyId], @ReverseLevel + 1
FROM [Hierarchy] h
INNER JOIN #RowEnd re ON re.ParentHierarchyId = h.[HierarchyId] AND re.ReverseLevel = @ReverseLevel
END
CREATE TABLE #Price
(
RowEndHierarchyId INT NOT NULL PRIMARY KEY,
[1] MONEY NULL,
[2] MONEY NULL,
[3] MONEY NULL,
[4] MONEY NULL,
[5] MONEY NULL,
[6] MONEY NULL,
[7] MONEY NULL
)
INSERT #Price (RowEndHierarchyId, [1], [2], [3], [4], [5], [6], [7])
SELECT p.RowEndHierarchyId, p.[1], p.[2], p.[3], p.[4], p.[5], p.[6], p.[7]
FROM (SELECT re.RowEndHierarchyId, re.Price, RANK() OVER (PARTITION BY re.RowEndHierarchyId ORDER BY re.ReverseLevel DESC) AS [Level]
FROM #RowEnd re) ppre
PIVOT (MIN([Price]) FOR [Level] IN ([1], [2], [3], [4], [5], [6], [7])) p
CREATE TABLE #Description
(
RowEndHierarchyId INT NOT NULL PRIMARY KEY,
[1] NVARCHAR(50) NULL,
[2] NVARCHAR(50) NULL,
[3] NVARCHAR(50) NULL,
[4] NVARCHAR(50) NULL,
[5] NVARCHAR(50) NULL,
[6] NVARCHAR(50) NULL,
[7] NVARCHAR(50) NULL
)
INSERT #Description (RowEndHierarchyId, [1], [2], [3], [4], [5], [6], [7])
SELECT d.RowEndHierarchyId, d.[1], d.[2], d.[3], d.[4], d.[5], d.[6], d.[7]
FROM (SELECT re.RowEndHierarchyId, dt.DescriptionText, RANK() OVER (PARTITION BY re.RowEndHierarchyId ORDER BY re.ReverseLevel DESC) AS [Level]
FROM #RowEnd re
LEFT JOIN [Description] dt ON re.DescriptionId = dt.DescriptionId) dpre
PIVOT (MIN([DescriptionText]) FOR [Level] IN ([1], [2], [3], [4], [5], [6], [7])) d
SELECT p.RowEndHierarchyId,
p.[1] AS Price1, d.[1] AS Description1,
p.[2] AS Price2, d.[2] AS Description2,
p.[3] AS Price3, d.[3] AS Description3,
p.[4] AS Price4, d.[4] AS Description4,
p.[5] AS Price5, d.[5] AS Description5,
p.[6] AS Price6, d.[6] AS Description6,
p.[7] AS Price7, d.[7] AS Description7
FROM #Price p
INNER JOIN #Description d ON p.RowEndHierarchyId = d.RowEndHierarchyId
ORDER BY p.RowEndHierarchyId
DROP TABLE #Description
DROP TABLE #Price
DROP TABLE #RowEnd
获取层次结构的基本逻辑与先前版本类似。但是，以这种方式索引临时表可能会大大提高查询性能。