为什么我的 SQL Server 查询在 UPDATE 和 SELECT 上的表现不同？

Question

我编写了一个 SQL Server 查询，它在对字段进行分区后更新记录以具有序列号。当我将它作为 SELECT 语句运行时，一切看起来都很棒：

DECLARE @RunDetailID INT = 448
DECLARE @JobDetailID INT

SELECT @JobDetailID = [JobDetailID] FROM [RunDetails] WHERE [RunDetailID] = @RunDetailID

SELECT
    [OrderedRecords].[NewSeq9],
    RIGHT([OrderedRecords].[NewSeq9], 4)
FROM
    (
        SELECT
            [Records].*,
            [Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9
        FROM
        (
            SELECT
                [MRDFStorageID], 
                [RunDetailID], 
                [SortField], 
                [PieceID], 
                [Seq9], 
                [BallotType]
            FROM
                [MRDFStorage]
                    JOIN [BallotStyles] ON [MRDFStorage].[SortField] = [BallotStyles].[Style] and [BallotStyles].[JobDetailID] = @JobDetailID
            WHERE
                [RunDetailID] IN (SELECT [RunDetailID] FROM [RunDetails] WHERE [JobDetailID] = @JobDetailID AND [RunStatusID] <> 0)
        ) Records
    ) OrderedRecords 
        JOIN MRDFStorage ON [OrderedRecords].[MRDFStorageID] = [MRDFStorage].[MRDFStorageID]
WHERE
    [MRDFStorage].[RunDetailID] = @RunDetailID

但是，当我将查询转换为 UPDATE 命令时，它开始跳过偶数：

DECLARE @RunDetailID INT = 448
DECLARE @JobDetailID INT 

SELECT @JobDetailID = [JobDetailID] FROM [RunDetails] WHERE [RunDetailID] = @RunDetailID

UPDATE
    [MRDFStorage]
SET
    [Seq9] = [OrderedRecords].[NewSeq9],
    [Overlay1] = [OrderedRecords].[NewSeq9],
    [Overlay10] = RIGHT([OrderedRecords].[NewSeq9], 4)
FROM
    (
        SELECT
            [Records].*,
            [Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9
        FROM
        (
            SELECT
                [MRDFStorageID], 
                [RunDetailID], 
                [SortField], 
                [PieceID], 
                [Seq9], 
                [BallotType], 
                CAST([SpecialProcessing] as Int) StartCount
            FROM
                [MRDFStorage]
                    JOIN [BallotStyles] ON [MRDFStorage].[SortField] = [BallotStyles].[Style] and [BallotStyles].[JobDetailID] = @JobDetailID
            WHERE
                [RunDetailID] IN (SELECT [RunDetailID] FROM [RunDetails] WHERE [JobDetailID] = @JobDetailID AND [RunStatusID] <> 0)
        ) Records
    ) OrderedRecords 
        JOIN MRDFStorage ON [OrderedRecords].[MRDFStorageID] = [MRDFStorage].[MRDFStorageID]
WHERE
    [MRDFStorage].[RunDetailID] = @RunDetailID

我试过专门关注这部分：

[Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9

有什么我不知道的副作用吗？

更新表定义

CREATE TABLE [dbo].[MRDFStorage] (
    [MRDFStorageID]           INT            IDENTITY (1, 1) NOT NULL,
    [RunDetailID]             INT            NOT NULL,
    [PieceID]                 VARCHAR (15)   NULL,
    [SortField]               VARCHAR (20)   NULL,
    [BallotType]              VARCHAR (100)  NULL,
    [Seq9]                    VARCHAR (15)   NULL,
    CONSTRAINT [PK_MRDFStorage] PRIMARY KEY CLUSTERED ([MRDFStorageID] ASC),
    CONSTRAINT [FK_MRDFStorage_RunDetails] FOREIGN KEY ([RunDetailID]) REFERENCES [dbo].[RunDetails] ([RunDetailID])
);

CREATE TABLE [dbo].[BallotStyles] (
    [BallotStyleID]     INT           IDENTITY (1, 1) NOT NULL,
    [JobDetailID]       INT           NOT NULL,
    [Style]             VARCHAR (20)  NOT NULL,
    CONSTRAINT [PK_BallotStyles] PRIMARY KEY CLUSTERED ([BallotStyleID] ASC)
);

CREATE TABLE [dbo].[RunDetails] (
    [RunDetailID]        INT            IDENTITY (1, 1) NOT NULL,
    [JobDetailID]        INT            NOT NULL,
    CONSTRAINT [PK_RunDetails] PRIMARY KEY CLUSTERED ([RunDetailID] ASC)
);

Answer 1

看不到您所拥有的数据会使这变得有点困难，但我设法重现了您所看到的内容：

insert into RunDetails(RunDetailID, JobDetailID) values(448, 1)
insert into MRDFStorage(RunDetailID, SortField) values(448, 'S1')
insert into MRDFStorage(RunDetailID, SortField) values(448, 'S1')
insert into MRDFStorage(RunDetailID, SortField) values(448, 'S1')
insert into BallotStyles(JobDetailID, Style) values(1, 'S1')
insert into BallotStyles(JobDetailID, Style) values(1, 'S1')

这里的关键因素是 中的两行BallotStyles。

您正在加入BallotStyles最内部的查询，并且使用上面的数据，您将获得 中每一行的重复行MRDFStorage。由于您没有使用其中的任何列，因此BallotStyles您只是检查行是否存在，并且可以使用子句来完成exists，这当然不会创建重复的行。

UPDATE
    [MRDFStorage]
SET
    [Seq9] = [OrderedRecords].[NewSeq9]
FROM
    (
        SELECT
            MRDFStorageID,
            [Records].[SortField] + RIGHT('0000' + CAST(ROW_NUMBER() OVER(PARTITION BY [Records].[SortField] ORDER BY [Records].[RunDetailID], [Records].[SortField], [Records].[PieceID]) AS VARCHAR), 4) NewSeq9
        FROM
        (
            SELECT
                M.[MRDFStorageID], 
                M.[RunDetailID], 
                M.[SortField], 
                M.[PieceID], 
                M.[BallotType]
            FROM
                [MRDFStorage] as M
-- Remove this join
--                    JOIN [BallotStyles] ON M.[SortField] = [BallotStyles].[Style] and [BallotStyles].[JobDetailID] = @JobDetailID
            WHERE
                [RunDetailID] IN (SELECT [RunDetailID] FROM [RunDetails] WHERE [JobDetailID] = @JobDetailID) and
-- Add this exists check instead of the join
                EXISTS (
                       SELECT *
                       FROM BallotStyles AS BS
                       WHERE M.SortField = BS.Style and
                             BS.JobDetailID = @JobDetailID
                       )
        ) Records
    ) OrderedRecords 
        JOIN MRDFStorage ON [OrderedRecords].[MRDFStorageID] = [MRDFStorage].[MRDFStorageID]
WHERE
    [MRDFStorage].[RunDetailID] = @RunDetailID