Question

以下查询返回正确的结果，但如何更快地获得相同的结果？

目标是通过总结今天，本周，月和季度的销售情况，输出一个跟踪卖家进度的表格。

SellerID    Today                 ThisWeek              ThisMonth             ThisQuarter
----------- --------------------- --------------------- --------------------- ---------------------
1           400,00                700,00                900,00                900,00
2           950,00                1850,00               2650,00               2650,00

我的查询：

CREATE TABLE #sales(
    [Price] MONEY,
    [Date] DATE,
    [SellerID] INT
)

INSERT INTO #sales VALUES 
(100, '2012-01-01', 1),
(200, '2012-04-01',1),
(300, '2012-04-23',1),
(400, '2012-04-27',1),
(700, '2012-01-01', 2),
(700, '2012-01-02', 2),
(800, '2012-04-01',2),
(900, '2012-04-23',2),
(950, '2012-04-27',2)


SELECT 
SellerID AS SellerID,

SUM(CASE WHEN [Date] >= DATEADD(DAY, DATEDIFF(DAY, 0, GETDATE()),0) THEN [Price] END) AS Today,
SUM(CASE WHEN [Date] >= DATEADD(WEEK, DATEDIFF(WEEK, 0, GETDATE()), 0) THEN [Price] END) AS ThisWeek,
SUM(CASE WHEN [Date] >= DATEADD(MONTH, DATEDIFF(MONTH, 0, GETDATE()), 0) THEN [Price] END) AS ThisMonth,
SUM(CASE WHEN [Date] >= DATEADD(QUARTER, DATEDIFF(QUARTER, 0, GETDATE()), 0) THEN [Price] END) AS ThisQuarter

FROM #sales
WHERE DATEPART(YEAR, [Date]) = DATEPART(YEAR, GETDATE()) 
GROUP BY SellerID

在较大的表上执行相同的查询时，这会非常慢。只需删除CASE语句就可以将执行时间缩短近50％。

如何以更快，更有效的方式获得相同的结果？

Answer 1

由于是星期五下午，我想我会扩展我对仓储的评论。即使您无法使用SSAS或任何其他OLAP完全探索多维数据集，您仍然可以执行自己的报表特定仓库。在你的情况下，我会建立一个新的数据库（我总是称我的DW，但世界是你的牡蛎），并创建2个模式Fact和Dim（代表事实和维度）。在您的情况下，它需要2个表，但您可能需要为“SellerID”添加另一个维度，具体取决于是否需要进一步报告。

CREATE TABLE Dim.Date
(       DateKey     DATE NOT NULL,
        DayOfWeek   VARCHAR(20) NOT NULL,
        Day         TINYINT NOT NULL,
        Week        TINYINT NOT NULL,
        Quarter     TINYINT NOT NULL,
        Month       TINYINT NOT NULL,
        Year        SMALLINT NOT NULL
    CONSTRAINT PK_Dim_Date_DateKey PRIMARY KEY (DateKey)
)
CREATE TABLE Fact.Sales
(       DateKey     DATE NOT NULL,
        SellerID    INT NOT NULL,
        Sales       INT NOT NULL,
        Amount      MONEY NOT NULL,
    CONSTRAINT PK_Fact_Sales PRIMARY KEY (DateKey, SellerID),
    CONSTRAINT FK_Fact_Sales_DateKey FOREIGN KEY (DateKey) REFERENCES Dim.Date
)

假设数据不会过期，您可以使用这样的程序在预定的工作中填满您的仓库：

DECLARE @MaxDate DATE
SELECT  @MaxDate = DATEADD(DAY, 1, MAX(DateKey))
FROM    Fact.Sales

INSERT INTO Dim.Date
SELECT  DATEADD(DAY, Increment, @MaxDate), 
        DATENAME(WEEKDAY, DATEADD(DAY, Increment, @MaxDate)), 
        DATEPART(DAY, DATEADD(DAY, Increment, @MaxDate)),
        DATEPART(WEEK, DATEADD(DAY, Increment, @MaxDate)),
        DATEPART(MONTH, DATEADD(DAY, Increment, @MaxDate)),
        DATEPART(QUARTER, DATEADD(DAY, Increment, @MaxDate)),
        DATEPART(YEAR, DATEADD(DAY, Increment, @MaxDate))
FROM    (   SELECT  ROW_NUMBER() OVER(ORDER BY Object_ID) - 1 [Increment]
            FROM    Sys.Objects
        ) obj
WHERE   NOT EXISTS
        (   SELECT  1
            FROM    Dim.Date
            WHERE   Date.DateKey = DATEADD(DAY, Increment, @MaxDate)
        )


INSERT INTO Fact.Sales
SELECT  [Date], SellerID, COUNT(*), SUM(Price)
FROM    LiveDatabase..Sales
WHERE   [Date] >= @MaxDate
GROUP BY [Date], SellerID

这将为您提供以下查询以生成报告

SELECT  SellerID,
        SUM(CASE WHEN Today.DateKey = Date.DateKey THEN Amount ELSE O END) [Today],
        SUM(CASE WHEN Today.Week = Date.Week THEN Amount ELSE O END) [ThisWeek],
        SUM(CASE WHEN Today.Month = Date.Month THEN Amount ELSE O END) [ThisMonth],
        SUM(CASE WHEN Today.Quarter = Date.Quarter THEN Amount ELSE O END) [ThisQuarter],
        SUM(CASE WHEN Today.Year = Date.Year THEN Amount ELSE O END) [ThisYear]
FROM    Fact.Sales
        INNER JOIN Dim.Date
            ON Date.DateKey = Sales.DateKey
        INNER JOIN Dim.Date Today
            ON Today.DateKey = CAST(GETDATE() AS DATE)
            AND Today.Year = Date.Year
GROUP BY SellerID

它看起来（如果有的话）比原始查询更复杂，但是在线数据库越多，您就会看到越多的好处。我已经完成了SQL Fiddle来展示优势，它用10000个随机销售记录填充实时数据，然后创建一个仓库（可能需要几秒钟来构建模式）。您应该注意到仓库上查询的执行时间明显更快（c.20x）。它在第一次运行时可能不会快20倍，但是一旦为两个查询缓存了查询计划，仓库查询的速度一直快20倍（无论如何都是为了我）。

Answer 2

保留数据的非规范化版本吗？

例如：http://sqlfiddle.com/#!3/300a5/2

select 
    *
    ,DATENAME(day, [date]) as day
    ,DATENAME(month, [date]) as month
    , DATENAME(year, [date])  as year
    ,DATENAME(quarter, [date]) as quarter
into deNormalised 
from #sales

然后您可以运行以下查询：

select 
    year
    ,sum(price)
from 
    deNormalised
where 
    quarter = 1
group by 
    year

获得多年来第一季度的比较

显然，这意味着您必须制定一个时间表来维护数据的非规范化版本。你可以在更新时或每小时触发一次。

您还可以尝试将最新数据添加到非规范化结果中。这样，您只对当前创建的行进行慢速处理。

编辑：我不知道使用DATENAME函数是否会改善使用现有结构的性能。

Answer 3

select 
     SellerID
    ,sum(case when [Date]=getdate() then [Price] else 0 end) as Today
    ,sum(case when datepart(week,[Date])=datepart(week,getdate()) then [Price] else 0 end) as ThisWeek
    ,sum(case when datepart(MONTH,[Date])=datepart(month,getdate()) then [Price] else 0 end) as ThisMonth
    ,sum(case when datepart(QUARTER,[Date])=datepart(QUARTER,getdate()) then [Price] else 0 end) as ThisQUARTER
from #sales
Group by SellerID

基于今天，本周，本月，今年第四季度的最快销售总结方式？

3 个答案: