1

是的，所以我有这个笨重的查询需要优化，我将它修剪了很多，以使其更具可读性，同时仍然可以理解这一点。

我基本上看到在顶级查询和所有三个子查询中都存在相同的“分组依据”逻辑，这些列也是“内部连接”逻辑的参数。问题是，我不确定如何优化它，尽管我可以想象必须有一些更简单的方法来实现相同的结果。

涉及的表：Invoice、InvoiceLine、ProductType

InvoiceLine 通过外键与 Invoice 和 ProductType 相关联

此查询应该将单个发票的总和 invoiceline.click 与所有其他发票的总和 invoiceline.click 进行比较，按 producttype.name 和 invoiceline.origin 分组，另外还按 invoice.final 拆分。所以结果应该是这样的：

产品类型 | 产地 | 点击参考发票 | 点击所有其他最终发票 | 点击所有其他未完成的发票

让我强调一下，查询确实有效，但它太慢了。关于优化这个东西的任何提示？

DECLARE @startDate datetime;
DECLARE @endDate datetime;
DECLARE @refInvoiceGuid uniqueidentifier;

SET @startDate='2013-09-01 00:00:00';
SET @endDate='2013-09-30 23:59:59';
SET @refInvoiceGuid='34d03903-a2ad-49ae-bd72-e98b47cdbc52';

SELECT
    ProductType.Name,
    InvoiceLine.Origin,
    invRef.ClicksRef,
    invFinal.ClicksFinal,
    invNotFinal.ClicksNotFinal
FROM InvoiceLine
INNER JOIN ProductType ON InvoiceLine.ProductType_Ref = ProductType.Id
INNER JOIN (
        SELECT
            ProductType.Name AS ProductName,
            InvoiceLine.Origin AS Origin,
            SUM(InvoiceLine.Clicks) AS ClicksRef
        FROM InvoiceLine
        INNER JOIN ProductType ON InvoiceLine.ProductType_Ref = ProductType.Id
        INNER JOIN Invoice ON Invoice.Id = InvoiceLine.Invoice_Ref
        WHERE
            InvoiceLine.BillingDate >= @startDate
            AND InvoiceLine.BillingDate <= @endDate
            AND Invoice.Guid = @refInvoiceGuid
        GROUP BY
            ProductType.Name, InvoiceLine.Origin
    ) invRef ON ProductType.Name = invRef.ProductName AND InvoiceLine.Origin = invRef.Origin
INNER JOIN (
        SELECT 
            ProductType.Name AS ProductName,
            InvoiceLine.Origin AS Origin,
            SUM(InvoiceLine.Clicks) AS ClicksFinal
        FROM InvoiceLine
        INNER JOIN ProductType ON InvoiceLine.ProductType_Ref = ProductType.Id
        INNER JOIN Invoice ON Invoice.Id=InvoiceLine.Invoice_Ref AND Invoice.Final=1
        WHERE
            InvoiceLine.BillingDate >= @startDate
            AND InvoiceLine.BillingDate <= @endDate
            AND Invoice.Guid != @refInvoiceGuid
        GROUP BY
            ProductType.Name, InvoiceLine.Origin
    ) invFinal ON ProductType.Name = invFinal.ProductName AND InvoiceLine.Origin = invFinal.Origin
INNER JOIN (
        SELECT 
            ProductType.Name AS ProductName,
            InvoiceLine.Origin AS Origin,
            SUM(InvoiceLine.Clicks) AS ClicksNotFinal
        FROM InvoiceLine
        INNER JOIN ProductType ON InvoiceLine.ProductType_Ref = ProductType.Id
        INNER JOIN Invoice ON Invoice.Id=InvoiceLine.Invoice_Ref AND Invoice.Final=0
        WHERE
            InvoiceLine.BillingDate >= @startDate
            AND InvoiceLine.BillingDate <= @endDate
            AND Invoice.Guid != @refInvoiceGuid
        GROUP BY
            ProductType.Name, InvoiceLine.Origin
    ) invNotFinal ON ProductType.Name = invNotFinal.ProductName AND InvoiceLine.Origin = invNotFinal.Origin
WHERE
    InvoiceLine.BillingDate >= @startDate
    AND InvoiceLine.BillingDate <= @endDate
GROUP BY
    ProductType.Name,
    InvoiceLine.Origin,
    invRef.ClicksRef,
    invFinal.ClicksFinal,
    invNotFinal.ClicksNotFinal

更新 1

我添加了一个索引：

CREATE NONCLUSTERED INDEX [IX_ProductOrigin] ON InvoiceLine (Invoice_Ref,BillingDate) INCLUDE (Origin,Clicks,ProductType_Ref);

而且我已经将查询重写为更紧凑（显然这具有相同的性能）：

SELECT
    ProductType.Name,
    InvoiceLine.Origin, 
    SUM(CASE WHEN Invoice.Guid = @refInvoiceGuid THEN InvoiceLine.Clicks ELSE 0 END) AS ClicksRef,
    SUM(CASE WHEN Invoice.Guid <> @refInvoiceGuid AND Invoice.Final = 1 THEN InvoiceLine.Clicks ELSE 0 END) AS ClicksFinal,
    SUM(CASE WHEN Invoice.Guid <> @refInvoiceGuid AND Invoice.Final = 0 THEN InvoiceLine.Clicks ELSE 0 END) AS ClicksNotFinal
FROM InvoiceLine  
INNER JOIN ProductType ON InvoiceLine.ProductType_Ref = ProductType.Id
INNER JOIN Invoice ON Invoice.id = InvoiceLine.Invoice_Ref
WHERE InvoiceLine.BillingDate >= @startDate AND InvoiceLine.BillingDate <= @endDate
GROUP BY ProductType.Name, InvoiceLine.Origin

它看起来速度提高了，可读性也确实提高了，但实际上它仍然需要 3 分钟才能执行。以下是统计数据和执行计划：

统计和计划

因此，如果我正确理解了统计数据，那么索引查找只需要很长时间吗？关于如何改进这一点的任何想法？还是我刚刚达到数据过多的地步？

以下是该指数的一些统计数据：

指数

和表：

桌子

运行'set statistics io on'的结果：

(170 row(s) affected)
Table 'ProductType'. Scan count 0, logical reads 340, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'Worktable'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'InvoiceLine'. Scan count 2741, logical reads 37444, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'Invoice'. Scan count 1, logical reads 115, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

更新 2

反转索引列的顺序后：

CREATE NONCLUSTERED INDEX [IX_ProductOrigin] ON InvoiceLine (BillingDate,Invoice_Ref) INCLUDE (Origin,Clicks,ProductType_Ref);

(170 row(s) affected)
Table 'ProductType'. Scan count 0, logical reads 340, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'Worktable'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'InvoiceLine'. Scan count 1, logical reads 28371, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'Invoice'. Scan count 1, logical reads 115, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

带有索引查找统计信息的执行计划：

执行计划2

这显着提高了性能（我们已经从 170 秒变为 15 秒）。感谢您迄今为止的帮助。还有其他建议吗？

4

1 回答 1

0 
DECLARE @startDate datetime;
DECLARE @endDate datetime;
DECLARE @refInvoiceGuid uniqueidentifier;

SET @startDate='2013-09-01 00:00:00';
SET @endDate='2013-09-30 23:59:59';
SET @refInvoiceGuid='34d03903-a2ad-49ae-bd72-e98b47cdbc52';
select p.Name,il1.origin,SUM(il1.Clicks) as ClicksRef,SUM(il2.Clicks) as ClicksFinal,
SUM(il3.Clicks) as ClicksNotFinal
from Invoice i 
inner join InvoiceLine il1 on il1.Invoice_Ref = i.Id
and il1.BillingDate >= @startDate AND il1.BillingDate <= @endDate AND i.Guid =       
@refInvoiceGuid
left join InvoiceLine il2 on il2.Invoice_Ref = il1.Invoice_Ref  and il2.Final=1 and    
il2.BillingDate = il1.BillingDate 
left join InvoiceLine il3 on il3.Invoice_Ref = il2.Invoice_Ref  and il3.Final=0 and    
il3.BillingDate = il1.BillingDate 
INNER JOIN ProductType p on il1.ProductType_Ref = p.Id
group by p.Name,il1.origin
于 2013-10-25T10:49:13.363 回答