堆栈内存溢出
  繁体   English   中英

SQL 服务器中每个客户按照交易日期时间顺序的累计净流出流量

[英]Accumulative Net Out Flow as per the Transaction Datetime Sequence each Customer in SQL Server

 原文  2020-11-30 08:55:56  5   1    sql/ sql-server/ tsql

问题描述

我正在尝试根据交易发生的顺序（DateTime ASC）为每个客户获取净流出流量。 为此，我创建了一个循环查询，但不幸的是它需要大量时间来处理。 20K 条记录需要 8 分钟，而我必须在 500 万条记录上运行它。

解释：例如，客户 923xxxxx307 完成了 4 笔交易，这里解释它是如何工作的

  1. 在示例数据表中，第一笔交易是 OUT FLOW，他在账户中使用了 26 卢比，这就是为什么在 Required Output 表中第一行的 NET OUT FLOW 是 26
  2. 在 Sample Data 表中，第二笔交易是 IN FLOW，他在他的账户中存入了 60 卢比，这 60 卢比应存放在他自己的单独存储桶中，这不应反映在 Required Out 表中
  3. 在示例数据表中，第三笔交易是 OUT FLOW，其中他从他的帐户脚本中使用了 62 卢比，应减去他在第 2 点维护的存储桶中的62 卢比，因此在 Required Output 表中，NET OUT FLOW 列应显示 2 即（62- 60 = 2)。 此外，他的存储桶应该变为 0，因为它在第三次交易中被完全消耗
  4. 在 Sample Data 表中，第四笔交易再次 IN FLOW，他在他的账户中存入了 360 卢比，因此他的存储桶应该再次显示 360 卢比，对 Required Output 表没有影响。 等等。

此存储桶将用于下一个 OUT FLOW 事务，或者如果下一个事务再次 IN FLOW，则应在存储桶中的现有 360 中添加新事务的数量

我还为您的测试运行添加了带有小样本数据集的完整 SQL 脚本。

请为此提供一些有效的解决方案。 我附上了 Input & Output 屏幕截图以及我的代码。

-- Creating SAMPLE DATA Table ---
DROP TABLE IF EXISTS #TestData
CREATE TABLE #TestData
(
    [FINANCIAL ID]  BIGINT,
    [DATE]          DATETIME,
    [TRXN DATETIME] DATETIME,
    AMOUNT          FLOAT,
    [CUSTOMER NO]   VARCHAR(20),
    [PRODUCT NAME]  VARCHAR(10)
)
GO

-- Inserting Sample Date in above table ---
INSERT INTO #TestData
VALUES (9442001596,'2020-11-01','2020-11-01 00:05:18',26,'923xxxxx307','OUT FLOW')

INSERT INTO #TestData
VALUES (9442094863,'2020-11-01','2020-11-01 00:15:01',60,'923xxxxx307','IN FLOW')

INSERT INTO #TestData
VALUES (9442106611,'2020-11-01','2020-11-01 00:16:26',62,'923xxxxx307','OUT FLOW')

INSERT INTO #TestData
VALUES (9442198611,'2020-11-01','2020-11-01 00:30:35',360,'923xxxxx307','IN FLOW')

INSERT INTO #TestData
VALUES (9442227548,'2020-11-01','2020-11-01 00:36:07',15000,'923xxxxx864','IN FLOW')

INSERT INTO #TestData
VALUES (9442264685,'2020-11-01','2020-11-01 00:44:03',1660,'923xxxxx864','IN FLOW')

INSERT INTO #TestData
VALUES (9442266137,'2020-11-01','2020-11-01 00:44:24',4540,'923xxxxx864','OUT FLOW')

INSERT INTO #TestData
VALUES (9442358832,'2020-11-01','2020-11-01 01:08:06',200,'923xxxxx864','OUT FLOW')

INSERT INTO #TestData
VALUES (9442434263,'2020-11-01','2020-11-01 01:34:05',190,'923xxxxx864','OUT FLOW')

INSERT INTO #TestData
VALUES (9442517054,'2020-11-01','2020-11-01 02:14:48',5000,'923xxxxx562','IN FLOW')

INSERT INTO #TestData
VALUES (9442525893,'2020-11-01','2020-11-01 02:20:18',5000,'923xxxxx562','IN FLOW')

INSERT INTO #TestData
VALUES (9442533823,'2020-11-01','2020-11-01 02:25:14',10000,'923xxxxx562','IN FLOW')

INSERT INTO #TestData
VALUES (9442541534,'2020-11-01','2020-11-01 02:30:25',10000,'923xxxxx562','IN FLOW')

INSERT INTO #TestData
VALUES (9442545883,'2020-11-01','2020-11-01 02:33:04',25500,'923xxxxx562','OUT FLOW')

INSERT INTO #TestData
VALUES (9442552698,'2020-11-01','2020-11-01 02:37:08',5000,'923xxxxx562','OUT FLOW')

INSERT INTO #TestData
VALUES (9443455472,'2020-11-01','2020-11-01 07:45:10',180,'923xxxxx074','OUT FLOW')

INSERT INTO #TestData
VALUES (9443529884,'2020-11-01','2020-11-01 07:54:41',280,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9443657359,'2020-11-01','2020-11-01 08:10:09',100,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9443670261,'2020-11-01','2020-11-01 08:11:34',100,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9443682756,'2020-11-01','2020-11-01 08:12:59',100,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9443683147,'2020-11-01','2020-11-01 08:13:01',100,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9443872236,'2020-11-01','2020-11-01 08:33:04',100,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9443872617,'2020-11-01','2020-11-01 08:33:06',100,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9443886681,'2020-11-01','2020-11-01 08:34:31',100,'923xxxxx074','IN FLOW')

INSERT INTO #TestData
VALUES (9444185688,'2020-11-01','2020-11-01 09:02:11',300,'923xxxxx074','IN FLOW')

-- Ranked table with ROW NUMBERS --- 
DROP TABLE IF EXISTS #TestData_Ranked
SELECT *, ROWNUMBER = ROW_NUMBER() OVER (ORDER BY td.[FINANCIAL ID])
INTO #TestData_Ranked
FROM #TestData td
GO

CREATE UNIQUE CLUSTERED INDEX [CI_ROWNUMBER] ON #TestData_Ranked (ROWNUMBER ASC)
GO


-- Main Query to get OUTPUT DATA table ---
DECLARE @start BIGINT   = (SELECT MIN(iios.ROWNUMBER) FROM #TestData_Ranked iios)
DECLARE @end BIGINT     = (SELECT MAX(iios.ROWNUMBER) FROM #TestData_Ranked iios)

DROP TABLE IF EXISTS #ibft_in
CREATE TABLE #ibft_in ([CUSTOMER NO] VARCHAR(20), [TRXN DATETIME] DATETIME, AMOUNT FLOAT)

DROP TABLE IF EXISTS #ibft_out
SELECT TOP(0) iios.[FINANCIAL ID], iios.DATE, iios.[TRXN DATETIME], iios.AMOUNT, iios.[CUSTOMER NO], iios.[PRODUCT NAME], 0 AS [NET OUT FLOW]
INTO #ibft_out
FROM #TestData_Ranked iios


WHILE (@start <= @end)
BEGIN

    DECLARE @cur_FinancialId BIGINT = (SELECT [FINANCIAL ID] FROM #TestData_Ranked WHERE ROWNUMBER = @start)
    DECLARE @cur_TrxnDateTime DATETIME = (SELECT [TRXN DATETIME] FROM #TestData_Ranked WHERE ROWNUMBER = @start)
    DECLARE @cur_Date DATE = (SELECT DATE FROM #TestData_Ranked WHERE ROWNUMBER = @start)
    DECLARE @cur_Amount FLOAT = (SELECT AMOUNT FROM #TestData_Ranked WHERE ROWNUMBER = @start)
    DECLARE @cur_Customer_No VARCHAR(20) = (SELECT [CUSTOMER NO] FROM #TestData_Ranked WHERE ROWNUMBER = @start)
    DECLARE @cur_ProductName VARCHAR(10) = (SELECT [PRODUCT NAME] FROM #TestData_Ranked WHERE ROWNUMBER = @start)

    DECLARE @IN_IfExists INT    = (SELECT COUNT(*) FROM #ibft_in ii WHERE ii.[CUSTOMER NO] = @cur_Customer_No)
    DECLARE @IN_Amount FLOAT    = ISNULL((SELECT ISNULL(ii.AMOUNT, 0) FROM #ibft_in ii WHERE ii.[CUSTOMER NO] = @cur_Customer_No), 0)

    DECLARE @Remaining_Amount FLOAT = ISNULL(CASE WHEN (@cur_Amount - @IN_Amount)<0 THEN 0 ELSE (@cur_Amount - @IN_Amount) END, 0)

    PRINT @start
    PRINT @cur_Amount
    PRINT @IN_Amount

    IF (@cur_ProductName = 'IN FLOW' AND @IN_IfExists = 0)
    BEGIN
        INSERT INTO #ibft_in
        VALUES (@cur_Customer_No, @cur_TrxnDateTime, @cur_Amount)
    END
    ELSE
    BEGIN
        IF (@cur_ProductName = 'IN FLOW' AND @IN_IfExists > 0)
        BEGIN
            UPDATE #ibft_in
            SET 
                [TRXN DATETIME] = @cur_TrxnDateTime,
                AMOUNT          = @IN_Amount + @cur_Amount
            WHERE 
                [CUSTOMER NO]           = @cur_Customer_No
        END
        ELSE
        BEGIN
            IF (@cur_ProductName = 'OUT FLOW')
            BEGIN
                INSERT INTO #ibft_out
                VALUES (@cur_FinancialId, @cur_Date, @cur_TrxnDateTime, @cur_Amount, @cur_Customer_No, @cur_ProductName, @Remaining_Amount)

                UPDATE #ibft_in
                SET 
                    AMOUNT = CASE WHEN (@IN_Amount - @cur_Amount)<0 THEN 0 ELSE (@IN_Amount - @cur_Amount) END
                WHERE
                    [CUSTOMER NO] = @cur_Customer_No
            END
        END
    END

SET @start = @start + 1
END
--------------------------------------------

样本数据

在此处输入图像描述

需要 OUTPUT

在此处输入图像描述

1 个解决方案

解决方案1
 1 已采纳  2020-12-01 01:31:10

编辑：

这在技术上仍然是一个 RBAR 查询，只是重写了。 我没有足够的样本数据来知道性能是什么样的，所以显然请对你的 20k 数据集运行它来比较速度，但这使用 CTE 而不是 CURSOR。 请注意，这从您的-- Ranked table with ROW NUMBERS ---行开始：

-- Ranked table with ROW NUMBERS --- 
DROP TABLE IF EXISTS #TestData_Ranked
SELECT *, 
    --ROWNUMBER = ROW_NUMBER() OVER (ORDER BY td.[FINANCIAL ID]), 
    TrxTypeSEQ = ROW_NUMBER() over (PARTITION BY [customer no] ORDER BY [trxn datetime])
INTO #TestData_Ranked
FROM #TestData td
GO

-- not unique but still an index
CREATE CLUSTERED INDEX [CI_ROWNUMBER] ON #TestData_Ranked (TrxTypeSEQ ASC)
GO


DROP TABLE IF EXISTS #ibft_out;

WiTH SortedTrx AS (
  SELECT 
    [Financial ID],
    [Date],
    [Trxn Datetime],
    [Amount],
    [Customer No],
    [Product Name],
    TrxTypeSEQ,
    BUCKET = case when [Product Name] = 'IN FLOW' then Amount ELSE 0 END,
    NETOUT = case when [Product Name] = 'OUT FLOW' then Amount ELSE 0 END
  FROM #TestData_Ranked  
  WHERE TrxTypeSEQ = 1

  UNION ALL 

  SELECT 
    AllTrx.[Financial ID],
    AllTrx.[Date],
    AllTrx.[Trxn Datetime],
    AllTrx.[Amount],
    AllTrx.[Customer No],
    AllTrx.[Product Name],
    AllTrx.TrxTypeSEQ,
    BUCKET = case when AllTrx.[Product Name] = 'IN FLOW' then SortedTrx.Bucket + AllTrx.Amount 
                    ELSE 
                        case when AllTrx.[Amount] > SortedTrx.Bucket then 0 else SortedTrx.Bucket - AllTrx.[Amount] end
                    END,
    NETOUT = case when AllTrx.[Product Name] = 'IN FLOW' then 0 
                    else 
                        case when AllTrx.[Amount] < SortedTrx.Bucket then 0 else AllTrx.[Amount] - SortedTrx.Bucket end
                    END
  FROM #TestData_Ranked  AllTrx 
  INNER JOIN SortedTrx ON AllTrx.[customer no] = SortedTrx.[customer no] AND SortedTrx.TrxTypeSEQ + 1 = AllTrx.TrxTypeSEQ
)
SELECT 
    [Financial ID],
    [Date],
    [Trxn Datetime],
    [Amount],
    [Customer No],
    [Product Name],
    NETOUT
INTO #ibft_out
FROM SortedTrx 
WHERE [Product Name] = 'OUT FLOW'
ORDER BY [Trxn Datetime]

SELECT * FROM #ibft_out

原件：

将您的WHILE循环重写为CURSOR将减少您查询#TestData_Ranked的次数从 6 到 1； 而不是每次查询#ibft_in 3x，您只需查询一次或两次，具体取决于事务类型。 我很想知道它给你什么样的表现； 使用执行计划分析的示例数据在 7 秒内运行，而使用您的代码则在 40 秒内运行。

在您的-- Main Query to get OUTPUT DATA table ---符合：

-- Create our useful tables for IN & OUT calculations 
DROP TABLE IF EXISTS #ibft_in
CREATE TABLE #ibft_in ([CUSTOMER NO] VARCHAR(20), [TRXN DATETIME] DATETIME, AMOUNT FLOAT)

DROP TABLE IF EXISTS #ibft_out
SELECT TOP(0) iios.[FINANCIAL ID], iios.DATE, iios.[TRXN DATETIME], iios.AMOUNT, iios.[CUSTOMER NO], iios.[PRODUCT NAME], 0 AS [NET OUT FLOW]
INTO #ibft_out
FROM #TestData_Ranked iios

-- Declare these once, outside a loop, to cut down on wasted work
DECLARE @currRow INT = 0,
        @currFinancialID BIGINT = 0,
        @currDateTime DATETIME,
        @currAmt FLOAT,
        @currCustomer VARCHAR(20),
        @currTrxType VARCHAR(10),
        @fundsAvailable FLOAT;
        
-- We want a one-way cursor as fast as we can
DECLARE trx_cursor CURSOR LOCAL FAST_FORWARD FOR
   SELECT 
      ROWNUMBER,
      [Financial ID],
      [Trxn Datetime],
      [Amount],
      [Customer No],
      [Product Name]
    FROM #TestData_Ranked
    ORDER BY ROWNUMBER ASC

OPEN trx_cursor
FETCH NEXT FROM trx_cursor INTO @currRow, 
                                @currFinancialID,
                                @currDateTime,
                                @currAmt,
                                @currCustomer,
                                @currTrxType
WHILE @@FETCH_STATUS = 0
BEGIN
    -- If this is an IN transaction, INSERT or UPDATE our bucket of rupees
    IF @currTrxType = 'IN FLOW' 
        MERGE #ibft_in as Tgt
        USING (select @currCustomer as cno, @currDateTime as dt, @currAmt as amt) as Src
        ON tgt.[CUSTOMER NO] = cno
        WHEN NOT MATCHED BY TARGET THEN INSERT 
            VALUES (cno, dt, amt) 
        WHEN MATCHED THEN UPDATE
            SET Tgt.[TRXN DATETIME] = dt,
                Tgt.AMOUNT = tgt.amount + amt;

    -- OTHERWISE, calculate our NET OUT
    ELSE 
    BEGIN
        SELECT @fundsAvailable = ISNULL((SELECT ISNULL(ii.AMOUNT, 0) FROM #ibft_in ii WHERE ii.[CUSTOMER NO] = @currCustomer), 0)

        INSERT INTO #ibft_out 
        VALUES ( 
            @currFinancialID, 
            CAST(@currDateTime as DATE), 
            @currDateTime, 
            @currAmt, 
            @currCustomer, 
            @currTrxType, 
            ISNULL(CASE WHEN (@currAmt - @fundsAvailable)<0 THEN 0 ELSE (@currAmt - @fundsAvailable) END, 0)
        )

        UPDATE #ibft_in
        SET AMOUNT = CASE WHEN (@fundsAvailable - @currAmt)<0 THEN 0 ELSE (@fundsAvailable - @currAmt) END
        WHERE [CUSTOMER NO] = @currCustomer
    END
        

FETCH NEXT FROM trx_cursor INTO @currRow, 
                                @currFinancialID,
                                @currDateTime,
                                @currAmt,
                                @currCustomer,
                                @currTrxType
END
CLOSE trx_cursor
DEALLOCATE trx_cursor

select * from #ibft_in
select * from #ibft_out

暂无
暂无

声明:本站的技术帖子网页，遵循CC BY-SA 4.0协议，如果您需要转载，请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.

相关问题 在 SQL Server 中对每个段的客户事务进行排名 SELECT 3 sql 中的客户交易 在SQL Server中为每个客户生成行项目的顺序ID 在SQL Server 2008中查找每个客户的记录总数 客户账户余额如何影响 SQL 服务器中的客户交易表？ Hive sql 找出每个国家有多少普通客户 asp.net和SQL Server +事务回滚 SQL：计算每个客户的操作 ADO.NET事务和SQL Server 2008事务 SQL Server日期时间超出范围问题
相关标签
 
粤ICP备18138465号  © 2020-2024 STACKOOM.COM