SQL合并父子数据列表

Question

我有一个要作为 XML 传递给存储过程的数据列表。数据是Widget 的列表，小部件包含WidgetItem 的列表（父子数据）。我想做一个基于ParentID 的Widget 子集的MERGE。 ParentID 的一些数据已更新，一些已被删除（因此从 xml 中丢失）并且一些数据是新的。

更新的数据永远不需要更新子数据，因为Widget记录只能调整，不能调整其中的项目（子数据）。插入总是会有一个或多个子记录（WidgetItems）。

我似乎无法弄清楚如何在 MERGE 中执行此操作，因为与处理数据层中的合并相比，这似乎是最好的方法。

这是我到目前为止所拥有的……我在卡住的地方放了一个 cmets：

CREATE PROCEDURE dbo.pWidgetsMerge
    @Widgets XML
AS

/*
Assumed XML input @Widgets xml:
<Widgets>
    <Widget>
        <WidgetID>
        <ParentID>
        <StartDate>
        <EndDate>
        <Details>
            <WidgetDetailItem>
                <WidgetDetailItemID>
                <WidgetID>
                <SomeID>                
                <SomeData>
*/

MERGE
    [dbo].[Widget] as w  
USING
    (
        SELECT
            'WidgetID' = P.value('WidgetID[1]', 'INT'),
            'ParentID' = P.value('ParentID[1]', 'INT'),
            'StartDate' = P.value('EffectiveStartDate[1]', 'DATETIME'),
            'EndDate' = P.value('EffectiveEndDate[1]', 'DATETIME')
        FROM
            @Widgets.nodes('/Widgets/Widget') PROPERTYFEED(P)
    ) 
    AS xmlIn
    (
        [WidgetID],
        [StartDate],
        [EndDate]
    )
    ON
        w.[WidgetID] = xmlIn.[WidgetID]
    WHEN
        NOT MATCHED
    THEN
        INSERT 
        (
            [ParentID],
            [StartDate],
            [EndDate]
        ) 
        VALUES
        (
            xmlIn.[ParentID],
            xmlIn.[StartDate],
            xmlIn.[EndDate]
        )


        /*STUCK HERE: After the insert, need to put in the child
            records into a new table [WidgetItems].  Maybe it's another
            operation outside of the merge?*/

    WHEN
        MATCHED AND (
            (w.[StartDate] <> xmlIn.[StartDate]) OR 
            (w.[EndDate] <> xmlIn.[EndDate]))
    THEN
        UPDATE SET
            w.[StartDate] = xmlIn.[StartDate],
            w.[EndDate] = xmlIn.[EndDate]
    WHEN
        NOT MATCHED BY SOURCE AND w.[ParentID] = xmlIn.[ParentID]
    THEN
        UPDATE SET
            w.[DeletedDate] = GETDATE()

另外，如果我遇到了这个错误，我们将不胜感激，或者我确实需要在数据层处理这个问题。

Answer 1

以下是可以回答您问题的更新代码。我添加了 cmets 来解释发生了什么。希望这是有道理的。

正如您所说，ParentID 对于传入的所有小部件都是相同的，因此我将其视为参数而不是 XML 的元素

DECLARE @ParentID INT = 1

DECLARE @Widgets AS XML = 
N'<Widgets>
    <Widget>
        <WidgetID />
        <StartDate />
        <EndDate />
        <Details>
            <WidgetDetailItem>
                <WidgetDetailItemID></WidgetDetailItemID>
                <WidgetID/>
                <SomeID>4</SomeID>             
                <SomeData/>
            </WidgetDetailItem>
            <WidgetDetailItem>
                <WidgetDetailItemID></WidgetDetailItemID>
                <WidgetID/>
                <SomeID>323</SomeID>             
                <SomeData/>
            </WidgetDetailItem>
            <WidgetDetailItem>
                <WidgetDetailItemID></WidgetDetailItemID>
                <WidgetID/>
                <SomeID>1</SomeID>            
                <SomeData/>
            </WidgetDetailItem>
        </Details>
    </Widget>
    <Widget>
        <WidgetID>10</WidgetID>
        <StartDate>January 1, 2015</StartDate>
        <EndDate />
        <Details>
            <WidgetDetailItem>
                <WidgetDetailItemID></WidgetDetailItemID>
                <WidgetID/>
                <SomeID>4</SomeID>         
                <SomeData/>
            </WidgetDetailItem>
            <WidgetDetailItem>
                <WidgetDetailItemID></WidgetDetailItemID>
                <WidgetID/>
                <SomeID>99</SomeID>         
                <SomeData/>
            </WidgetDetailItem>
            <WidgetDetailItem>
                <WidgetDetailItemID></WidgetDetailItemID>
                <WidgetID/>
                <SomeID>6</SomeID>            
                <SomeData/>
            </WidgetDetailItem>
        </Details>
    </Widget>
</Widgets>';

--Used to hold the pseudoID -> WidgetID relationship for inserting the details
DECLARE @WidgetIds AS TABLE ([Action] varchar(10), PseudoID INT, WidgetID INT);

; 
--Use a CTE of the subset of data to be more performant. If we just went straight to the 
--merge we'd be operating on the entire table and that can have some major performance hits
WITH T AS (
              SELECT 
                     w.* 
              FROM
                     [dbo].[Widget] as w 
              WHERE
                     w.[ParentID] = @ParentID
)
MERGE INTO T 
USING (
        SELECT
            --Generate a pseudoid based on the order of the Widget elements so that we have some way of 
            --linking the detail records to the master
            row_number() OVER(ORDER BY PROPERTYFEED.P) PseudoID,
            'WidgetID' = P.value('WidgetID[1]', 'INT'),
            'ParentID' = @ParentID,
            'StartDate' = P.value('StartDate[1]', 'DATETIME'),
            'EndDate' = P.value('EndDate[1]', 'DATETIME')
        FROM
            @Widgets.nodes('/Widgets/Widget') PROPERTYFEED(P)
    ) 
    AS xmlIn
    (
           [PseudoID],
        [WidgetID],
        [ParentID],
        [StartDate],
        [EndDate]
    )
    ON
        T.[WidgetID] = xmlIn.[WidgetID]
    WHEN
        NOT MATCHED
    THEN
        INSERT 
        (
            [ParentID],
            [StartDate],
            [EndDate]
        ) 
        VALUES
        (
            xmlIn.[ParentID],
            xmlIn.[StartDate],
            xmlIn.[EndDate]
        )
    WHEN
        MATCHED AND (
            (T.[StartDate] <> xmlIn.[StartDate]) OR 
            (T.[EndDate] <> xmlIn.[EndDate]))
    THEN
        UPDATE SET
            T.[StartDate] = xmlIn.[StartDate],
            T.[EndDate] = xmlIn.[EndDate]
    WHEN
        NOT MATCHED BY SOURCE AND T.[DeletedDate] IS NULL 
    THEN
        UPDATE SET
            T.[DeletedDate] = GETDATE()         
OUTPUT  $action, xmlIn.PseudoID, INSERTED.WidgetID INTO @WidgetIds

;

--This is some magic to generate a temp table of numbers from 1 to COUNT(Widget)
--This is so we can reference the parent Widget row in the same order as the pseudoid generated above
--http://stackoverflow.com/a/1134379/4375845
;WITH Total(TotalWidgets) AS (SELECT COUNT(1) TotalWidgets FROM @Widgets.nodes('/Widgets/Widget') PROPERTYFEED(P))
       , Numbers(Num) as (
              SELECT 1 AS Num
              UNION ALL 
              SELECT Num+1 
              FROM Numbers
              JOIN Total t ON 1 = 1
              WHERE Num < t.TotalWidgets )
INSERT INTO WidgetDetailItem (WidgetID,SomeID,SomeData)
SELECT 
       w.WidgetID
       ,Details.SomeID
       ,Details.SomeData
FROM 
    (SELECT 
        P.value('WidgetDetailItemID[1]','int')  WidgetDetailItemID           
        , P.value('SomeID[1]','int') SomeID
        , P.value('SomeData[1]','varchar(5)') SomeData
        , n.Num AS PsuedoID
    FROM Numbers n
    --This is what gives us our pseudo ID to link to the row_number() function from the first merge statement
    CROSS APPLY @Widgets.nodes('/Widgets/Widget[sql:column("n.Num")]/Details/WidgetDetailItem') AS M(P)
    ) Details
JOIN @WidgetIds w on Details.PsuedoID = w.PseudoID
WHERE w.Action = 'INSERT' --We only want inserts by your spec

SELECT * FROM Widget;
SELECT * FROM WidgetDetailItem;

Answer 2

我已将传入的 XML 反序列化为一个表。这样就有机会验证 XML 字符串中的数据。

第 1 项：新的反序列化表将允许一种简单的方法来过滤哪些数据包含在 MERGE 中。

第 2 项：将数据插入子表必须在单独的调用中进行。 MERGE 只能处理一个表上的 crud（创建更新删除）操作。

注意：MERGE 将使用 DESTINATION 表中的所有记录。因此，当您与 Source 不匹配时，这将作用于表中未包含的所有记录。