这天我有一个非常奇怪的行为。我正在尝试使用 SQL Server (2017) 内存表和列存储索引。但是在创建了一个非常小的数据库（只有 2 个表），里面的数据非常少之后，我注意到我的一个查询返回了错误的数据。
一些背景：
第一个表 [Attribute]包含一些分层数据（Id、ParentId）。该表有一个触发器，它计算嵌套集模型的一些列。
基于此，我创建了一个视图 [vw_Attribute_Hierarchy]，它返回所有属性及其子节点的列表。我想用它来总结每个分层节点的数据（因此包括子值）。
第二个表 [FactImpacts] 仅包含一些要在列 [PlanValue] 和 [ActualValue] 上汇总的数据。
我试图创建一个仅包含所需列/数据的架构。但似乎查询结果取决于给定的数据集。
这是目前我最小的脚本（请在 SQL CMD 模式下执行）来重现我的问题，它应该返回两个结果，每行包含三行，但我只得到一个用于第一个查询和三个用于第二个查询。
用于创建数据库和一些内容的脚本：
SET ANSI_NULLS, ANSI_PADDING, ANSI_WARNINGS, ARITHABORT, CONCAT_NULL_YIELDS_NULL, QUOTED_IDENTIFIER ON;
SET NUMERIC_ROUNDABORT OFF;
GO
:setvar DatabaseName "NEW-DB-823474692347"
:setvar DefaultFilePrefix "NEW-DB-823474692347"
:setvar DefaultDataPath "C:\Program Files\Microsoft SQL Server\MSSQL14.SQL2017\MSSQL\DATA\"
:setvar DefaultLogPath "C:\Program Files\Microsoft SQL Server\MSSQL14.SQL2017\MSSQL\DATA\"
GO
:on error exit
GO
/*
Detect SQLCMD mode and disable script execution if SQLCMD mode is not supported.
To re-enable the script after enabling SQLCMD mode, execute the following:
SET NOEXEC OFF;
*/
:setvar __IsSqlCmdEnabled "True"
GO
IF N'$(__IsSqlCmdEnabled)' NOT LIKE N'True'
BEGIN
PRINT N'SQLCMD mode must be enabled to successfully execute this script.';
SET NOEXEC ON;
END
GO
USE [master];
GO
IF (DB_ID(N'$(DatabaseName)') IS NOT NULL)
BEGIN
ALTER DATABASE [$(DatabaseName)]
SET SINGLE_USER WITH ROLLBACK IMMEDIATE;
DROP DATABASE [$(DatabaseName)];
END
GO
PRINT N'Creating $(DatabaseName)...'
GO
CREATE DATABASE [$(DatabaseName)]
ON
PRIMARY(NAME = [$(DatabaseName)], FILENAME = N'$(DefaultDataPath)$(DefaultFilePrefix)_Primary.mdf')
LOG ON (NAME = [$(DatabaseName)_log], FILENAME = N'$(DefaultLogPath)$(DefaultFilePrefix)_Primary.ldf') COLLATE Latin1_General_CI_AS
GO
IF EXISTS (SELECT 1
FROM [master].[dbo].[sysdatabases]
WHERE [name] = N'$(DatabaseName)')
BEGIN
ALTER DATABASE [$(DatabaseName)]
SET AUTO_CLOSE OFF
WITH ROLLBACK IMMEDIATE;
END
GO
PRINT N'Creating [FG_MemoryOptimizedData]...';
GO
ALTER DATABASE [$(DatabaseName)]
ADD FILEGROUP [FG_MemoryOptimizedData] CONTAINS MEMORY_OPTIMIZED_DATA;
GO
ALTER DATABASE [$(DatabaseName)]
ADD FILE (NAME = [FG_MemoryOptimizedData_76BAE8A0], FILENAME = N'$(DefaultDataPath)$(DefaultFilePrefix)_FG_MemoryOptimizedData_76BAE8A0.mdf') TO FILEGROUP [FG_MemoryOptimizedData];
GO
USE [$(DatabaseName)];
GO
PRINT N'Creating [dbo].[Attribute]...';
GO
CREATE TABLE [dbo].[Attribute] (
[TenantId] UNIQUEIDENTIFIER NOT NULL,
[AttributeId] UNIQUEIDENTIFIER NOT NULL,
[AttributeParentId] UNIQUEIDENTIFIER NULL,
[AttributeName] NVARCHAR (128) NOT NULL,
[LeftExtent] INT NOT NULL,
[RightExtent] INT NOT NULL,
[HierarchyLevel] INT NOT NULL,
[NodeCount] INT NOT NULL,
CONSTRAINT [PK_Attribute] PRIMARY KEY NONCLUSTERED ([AttributeId] ASC)
);
GO
PRINT N'Creating [dbo].[Attribute].[IX_Attribute_Rn]...';
GO
CREATE UNIQUE CLUSTERED INDEX [IX_Attribute_Rn]
ON [dbo].[Attribute]([TenantId] ASC, [AttributeName] ASC);
GO
PRINT N'Creating [dbo].[Attribute].[IX_Attribute_Name]...';
GO
CREATE UNIQUE NONCLUSTERED INDEX [IX_Attribute_Name]
ON [dbo].[Attribute]([AttributeName] ASC);
GO
PRINT N'Creating [dbo].[FactImpact]...';
GO
CREATE TABLE [dbo].[FactImpact] (
[Rn] BIGINT IDENTITY (1, 1) NOT NULL,
[TenantId] UNIQUEIDENTIFIER CONSTRAINT [DF_FactImpact_TenantId] DEFAULT '10000000-0000-0000-0000-000000000000' NOT NULL,
[ImpactId] UNIQUEIDENTIFIER CONSTRAINT [DF_FactImpact_ImpactId] DEFAULT NEWID() NOT NULL,
[AttributeId] UNIQUEIDENTIFIER NOT NULL,
[Date] DATE NOT NULL,
[PlanValue] DECIMAL (38, 10) NULL,
[ActualValue] DECIMAL (38, 10) NULL,
CONSTRAINT [PK_FactImpact_Rn] PRIMARY KEY NONCLUSTERED HASH ([Rn]) WITH (BUCKET_COUNT = 100000),
INDEX [COLIX_FactImpact] CLUSTERED COLUMNSTORE
)
WITH (MEMORY_OPTIMIZED = ON);
GO
PRINT N'Creating [dbo].[DF_Attribute_TenantId]...';
GO
ALTER TABLE [dbo].[Attribute]
ADD CONSTRAINT [DF_Attribute_TenantId] DEFAULT '10000000-0000-0000-0000-000000000000' FOR [TenantId];
GO
PRINT N'Creating [dbo].[DF_Attribute_Id]...';
GO
ALTER TABLE [dbo].[Attribute]
ADD CONSTRAINT [DF_Attribute_Id] DEFAULT NEWSEQUENTIALID() FOR [AttributeId];
GO
PRINT N'Creating [dbo].[DF_Attribute_LeftExtend]...';
GO
ALTER TABLE [dbo].[Attribute]
ADD CONSTRAINT [DF_Attribute_LeftExtend] DEFAULT 0 FOR [LeftExtent];
GO
PRINT N'Creating [dbo].[DF_Attribute_RightExtend]...';
GO
ALTER TABLE [dbo].[Attribute]
ADD CONSTRAINT [DF_Attribute_RightExtend] DEFAULT 0 FOR [RightExtent];
GO
PRINT N'Creating [dbo].[DF_Attribute_HierarchyLevel]...';
GO
ALTER TABLE [dbo].[Attribute]
ADD CONSTRAINT [DF_Attribute_HierarchyLevel] DEFAULT 0 FOR [HierarchyLevel];
GO
PRINT N'Creating [dbo].[DF_Attribute_ChildCount]...';
GO
ALTER TABLE [dbo].[Attribute]
ADD CONSTRAINT [DF_Attribute_ChildCount] DEFAULT 0 FOR [NodeCount];
GO
PRINT N'Creating [dbo].[FK_Attribute_ParentId]...';
GO
ALTER TABLE [dbo].[Attribute]
ADD CONSTRAINT [FK_Attribute_ParentId] FOREIGN KEY ([AttributeParentId]) REFERENCES [dbo].[Attribute] ([AttributeId]);
GO
PRINT N'Creating [dbo].[vw_Attribute_Hierarchy]...';
GO
CREATE VIEW [dbo].[vw_Attribute_Hierarchy]
AS
SELECT
[Root].[TenantId],
[Root].[AttributeId] AS [AttributeRootId],
[Root].[AttributeName] AS [AttributeRootName],
[Node].[AttributeId] AS [AttributeNodeId],
[Node].[AttributeName] AS [AttributeNodeName]
FROM [dbo].[Attribute] [Root]
JOIN [dbo].[Attribute] [Node] ON [Node].[LeftExtent] BETWEEN [Root].[LeftExtent] AND [Root].[RightExtent]
GO
PRINT N'Creating [dbo].[iusp_CalculateNestedSets]...';
GO
CREATE PROCEDURE [dbo].[iusp_CalculateNestedSets]
(
@SchemaName SYSNAME,
@TableName SYSNAME,
@GroupColumnName SYSNAME,
@IdColumnName SYSNAME = 'Id',
@ParentIdColumnName SYSNAME = 'ParentId'
)
AS
-- http://www.sqlservercentral.com/articles/Hierarchy/94040/
SET NOCOUNT ON
CREATE TABLE [#HTally]([N] INT NOT NULL PRIMARY KEY)
INSERT INTO [#HTally]([N])
SELECT TOP 500 --(16 * 500 = VARBINARY(84000) in length)
N = ISNULL(CAST(
(ROW_NUMBER() OVER (ORDER BY (SELECT NULL))-1)*16+1
AS INT),0)
FROM sys.all_columns ac1
CROSS JOIN sys.all_columns ac2
--===== Build the new table on-the-fly including some place holders
;WITH cteBuildPath AS
(
--=== This is the "anchor" part of the recursive CTE.
-- The only thing it does is load the Root Node.
SELECT
[ANCHOR].[TenantId],
[ANCHOR].[AttributeId],
[ANCHOR].[AttributeParentId],
HLevel = 1,
[SortPath] = CAST(
CAST([ANCHOR].[AttributeId] AS BINARY(16))
AS VARBINARY(8000)) --Up to 1000 levels deep.
FROM [dbo].[Attribute] AS anchor
WHERE [AttributeParentId] IS NULL --Only the Root Node has a NULL [ParentId]
UNION ALL
--==== This is the "recursive" part of the CTE that adds 1 for each level
-- and concatenates each level of [Id]'s to the [SortPath] column.
SELECT
[RECUR].[TenantId],
[RECUR].[AttributeId],
[RECUR].[AttributeParentId],
HLevel = cte.HLevel + 1,
[SortPath] = CAST( --This does the concatenation to build [SortPath]
cte.[SortPath] + CAST([RECUR].[AttributeId] AS BINARY(16))
AS VARBINARY(8000))
FROM [dbo].[Attribute] AS recur WITH (TABLOCK)
INNER JOIN cteBuildPath AS cte ON cte.[AttributeId] = [RECUR].[AttributeParentId]
)
--=== This final INSERT/SELECT creates the Node # in the same order as a
-- push-stack would. It also creates the final table with some
-- "reserved" columns on the fly. We'll leave the [SortPath] column in
-- place because we're still going to need it later.
-- The ISNULLs make NOT NULL columns
SELECT
[AttributeId] = ISNULL([SORTED].[AttributeId], '00000000-0000-0000-0000-000000000000'),
[SORTED].[AttributeParentId],
[HLevel] = ISNULL([SORTED].HLevel,0),
[LeftExtent] = ISNULL(CAST(0 AS INT),0), --Place holder
[RightExtent] = ISNULL(CAST(0 AS INT),0), --Place holder
[NodeNumber] = ROW_NUMBER() OVER (PARTITION BY [SORTED].[TenantId] ORDER BY [SORTED].[SortPath]),
[NodeCount] = ISNULL(CAST(0 AS INT),0), --Place holder
[SortPath] = ISNULL([SORTED].[SortPath],[SORTED].[SortPath])
INTO [#Hierarchy]
FROM cteBuildPath AS sorted
OPTION (MAXRECURSION 500) --Change this IF necessary
;
--===== Declare a working variable to hold the result of the calculation
-- of the LeftExtent so that it may be easily used to create the
-- RightExtent in a single scan of the final table.
DECLARE @LeftExtent INT
--===== Create the Nested Sets from the information available in the table
-- and in the following CTE. This uses the proprietary form of UPDATE
-- available in SQL Serrver for extra performance.
;WITH cteCountDownlines AS
(
--=== Count each occurance of [Id] in the sort path
SELECT
[AttributeId] = CAST(SUBSTRING(h.[SortPath],[T].N,16) AS UNIQUEIDENTIFIER),
[NodeCount] = COUNT(*) --Includes current node
FROM [#Hierarchy] h, [#HTally] t
WHERE [T].N BETWEEN 1 AND DATALENGTH([SortPath])
GROUP BY SUBSTRING(h.[SortPath],[T].N, 16)
) --=== Update the NodeCount and calculate both Bowers
UPDATE [SourceTable]
SET @LeftExtent = [SourceTable].[LeftExtent] = 2 * h.[NodeNumber] - h.[HLevel],
[SourceTable].[RightExtent] = ([downline].[NodeCount] - 1) * 2 + @LeftExtent + 1,
[SourceTable].[NodeCount] = [downline].[NodeCount],
[SourceTable].[HierarchyLevel] = h.[HLevel]
FROM [dbo].[Attribute] AS [SourceTable]
JOIN [#Hierarchy] h ON h.[AttributeId] = [SourceTable].[AttributeId]
JOIN cteCountDownlines downline ON h.[AttributeId] = [downline].[AttributeId]
WHERE NOT (
[SourceTable].[LeftExtent] = 2 * h.[NodeNumber] - h.[HLevel]
AND [SourceTable].[RightExtent] = ([downline].[NodeCount] - 1) * 2 + @LeftExtent + 1
AND [SourceTable].[NodeCount] = [downline].[NodeCount]
AND [SourceTable].[HierarchyLevel] = h.[HLevel]
)
;
RETURN 0
GO
PRINT N'Creating [dbo].[tr_Attribute_IUD]...';
GO
CREATE TRIGGER [dbo].[tr_Attribute_IUD]
ON [dbo].[Attribute]
FOR INSERT, UPDATE, DELETE
AS
BEGIN
SET NOCOUNT ON
IF UPDATE([AttributeId]) OR UPDATE([AttributeParentId]) BEGIN
EXEC [dbo].[iusp_CalculateNestedSets] @SchemaName = 'dbo', @TableName = 'Attribute', @GroupColumnName = 'TenantId', @IdColumnName = 'AttributeId', @ParentIdColumnName = 'AttributeParentId'
END
END
GO
/*
Post-Deployment Script Template
--------------------------------------------------------------------------------------
This file contains SQL statements that will be appended to the build script.
Use SQLCMD syntax to include a file in the post-deployment script.
Example: :r .\myfile.sql
Use SQLCMD syntax to reference a variable in the post-deployment script.
Example: :setvar TableName MyTable
SELECT * FROM [$(TableName)]
--------------------------------------------------------------------------------------
*/
DECLARE @ItemsPerHierarchyLevel INT = 2
DECLARE @HierarchyLevels INT = 3
-- ATTRIBUTE DATA
;WITH [Source]
AS
(
SELECT TOP (@ItemsPerHierarchyLevel)
ROW_NUMBER() OVER(ORDER BY (SELECT 1)) AS [RN]
FROM [sys].[syscolumns]
),
[CTE]
AS
(
SELECT
NEWID() AS [AttributeId],
CAST(NULL AS UNIQUEIDENTIFIER) AS [AttributeParentId],
CAST('Attribute-' + CAST([Source].[RN] AS NVARCHAR(128)) AS NVARCHAR(128)) AS [AttributeName],
1 AS [Level]
FROM [Source]
UNION ALL
SELECT
NEWID() AS [AttributeId],
[CTE].[AttributeId] AS [AttributeParentId],
CAST([CTE].[AttributeName] + '.' + CAST([Source].[RN] AS NVARCHAR(128)) AS NVARCHAR(128)) AS [AttributeName],
[CTE].[Level] + 1 AS [Level]
FROM [Source]
JOIN [CTE] ON [CTE].[Level] < @HierarchyLevels
)
INSERT INTO [dbo].[Attribute]([AttributeId], [AttributeParentId], [AttributeName])
SELECT
[CTE].[AttributeId],
[CTE].[AttributeParentId],
[CTE].[AttributeName]
FROM [CTE]
ORDER BY [AttributeName]
;
-- IMPACT DATA
;WITH [RN]
AS
(
SELECT TOP (24)
ROW_NUMBER() OVER(ORDER BY (SELECT 1)) AS [RN]
FROM [sys].[syscolumns]
),
[Date]
AS
(
SELECT DATEADD(MONTH, [RN].[RN] - 1, DATEFROMPARTS(2019, 1, 1)) AS [Date]
FROM [RN]
)
INSERT INTO [dbo].[FactImpact]([AttributeId], [Date], [PlanValue], [ActualValue])
SELECT
[Attribute].[AttributeId],
[Date].[Date],
1 AS [PlanValue],
1 AS [ActualValue]
FROM [dbo].[Attribute],
[dbo].[Attribute] [Attribute2], -- ONLY TO DUPLICATE DATA
[dbo].[Attribute] [Attribute3], -- ONLY TO DUPLICATE DATA
[Date]
;
GO
PRINT N'Update complete.';
GO
USE [NEW-DB-823474692347]
GO
以下两个查询应始终返回相同的结果：
USE [NEW-DB-823474692347]
GO
SELECT
[vw_Attribute_Hierarchy].[AttributeRootId],
[vw_Attribute_Hierarchy].[AttributeRootName],
SUM([FactImpact].[PlanValue]) AS [SumFactImpacts]
FROM
[dbo].[FactImpact] INNER JOIN [vw_Attribute_Hierarchy] [vw_Attribute_Hierarchy] ON [vw_Attribute_Hierarchy].[AttributeNodeId] = [FactImpact].[AttributeId]
WHERE [vw_Attribute_Hierarchy].[AttributeNodeName] = 'Attribute-1.1.1'
GROUP BY [vw_Attribute_Hierarchy].[AttributeRootId]
, [vw_Attribute_Hierarchy].[AttributeRootName]
GO
SELECT
[vw_Attribute_Hierarchy].[AttributeRootId],
[vw_Attribute_Hierarchy].[AttributeRootName],
SUM([FactImpact].[PlanValue]) AS [SumFactImpacts]
FROM
-- SWITCH LEFT AND RIGHT TABLE ONLY
[vw_Attribute_Hierarchy] [vw_Attribute_Hierarchy] INNER JOIN [dbo].[FactImpact] ON [vw_Attribute_Hierarchy].[AttributeNodeId] = [FactImpact].[AttributeId]
WHERE [vw_Attribute_Hierarchy].[AttributeNodeName] = 'Attribute-1.1.1'
GROUP BY [vw_Attribute_Hierarchy].[AttributeRootId]
, [vw_Attribute_Hierarchy].[AttributeRootName]
GO
奇怪的是，如果您进行以下更改之一，两个查询总是返回相同的正确数据：
1) 我为最后两个 SELECT 语句切换了 INNER JOIN 中的表。如果 [FactImpact] 在第二位，则结果是正确的。
2) 删除 [FactImpacts] 上的 In-Memory 选项
3) 删除 [FactImpacts] 上的聚集列存储索引
4）减少或增加[FactImpacts]中的数据量
DELETE [dbo].[FactImpact] WHERE [AttributeId] IN (SELECT [AttributeId] FROM [Attribute] WHERE [Attribute].[AttributeName] LIKE 'Attribute-2%')
5）在表[属性]上再添加一个索引
CREATE INDEX [IX_Attribute_Nested_Set] ON [dbo].[Attribute] ([LeftExtent], [RightExtent])
我希望你能听从我的解释！
我能够在 SQL Server 2016 / 2017 中本地重现这种奇怪的行为。