Question

对于具有350,000条记录的最大表PremiseProviderBillings的数据库，以下SQL需要5秒。但是在具有150万条记录的相同数据库上，它需要一分钟

SELECT
   n.CustomerInvoiceNumberId as InvoiceNo,C.CustomerBillId,c.customerid, S.Volumetric, S.Fixed, S.VAT, S.Discount, C.Debit,c.EffectiveDate,c.TransactionDateTime,s.Consumption,r.CustomerCreditNoteId--,s.Volumetric + s.Fixed + s.Vat - s.discount - c.debit as variance
FROM
  CustomerPayments C 
INNER JOIN
  (SELECT
     CustomerBillId, SUM(a.VolumetricCharge) as Volumetric,SUM(a.FixedCharge) as Fixed,
     SUM(a.VAT) as VAT,SUM(a.Discount) as Discount,sum(a.EstimatedConsumption) as Consumption
   FROM
     PremiseProviderBillings a, PremiseProviderBills b
    WHERE a.PremiseProviderBillId = b.PremiseProviderBillId
   GROUP BY
     CustomerBillId) S
ON
  C.CustomerBillId = S.CustomerBillId 
  and debit <> 0 -- hide credit note lines, we mark these results with customerCreditNoteId to show they have been credited
INNER JOIN dbo.CustomerInvoiceNumbers n on c.CustomerBillId = n.CustomerBillId
left OUTER JOIN
           dbo.CustomerCreditNotes AS r ON c.CustomerPaymentId = r.CustomerPaymentId
where isnull(c.transactionDateTimeEnd,'')=''

如果我然后运行内部SQL部分，它将较小数据库上的值相加，则需要2秒。在较大的数据库上需要34秒，内部SQL低于......

SELECT
     CustomerBillId, SUM(a.VolumetricCharge) as Volumetric,SUM(a.FixedCharge) as Fixed,
     SUM(a.VAT) as VAT,SUM(a.Discount) as Discount,sum(a.EstimatedConsumption) as Consumption
   FROM
     PremiseProviderBillings a, PremiseProviderBills b
    WHERE a.PremiseProviderBillId = b.PremiseProviderBillId
   GROUP BY
     CustomerBillId

很明显，这个SQL根本无法扩展。鉴于数据库将增长应该采用哪种技术来改善这一点？

我检查了所有连接，以确保没有丢失的索引，以确保所有连接都是基于密钥的，确定

我原本以为这种方法没问题，但是我应该改变SQL的结构，这是不可扩展且效率低的吗？

此致

Answer 1

尝试对内部查询使用公用表表达式，这可能会加快一些速度。

WITH CTE AS
(
    SELECT
         CustomerBillId, SUM(a.VolumetricCharge) as Volumetric,SUM(a.FixedCharge) as Fixed,
         SUM(a.VAT) as VAT,SUM(a.Discount) as Discount,sum(a.EstimatedConsumption) as Consumption
       FROM
         PremiseProviderBillings a, PremiseProviderBills b
        WHERE a.PremiseProviderBillId = b.PremiseProviderBillId
       GROUP BY
         CustomerBillId
)   

SELECT
   n.CustomerInvoiceNumberId as InvoiceNo,C.CustomerBillId,c.customerid, S.Volumetric, S.Fixed, S.VAT, S.Discount, C.Debit,c.EffectiveDate,c.TransactionDateTime,s.Consumption,r.CustomerCreditNoteId--,s.Volumetric + s.Fixed + s.Vat - s.discount - c.debit as variance
FROM
  CustomerPayments C 
INNER JOIN
  CTE S
ON
  C.CustomerBillId = S.CustomerBillId 
  and debit <> 0 -- hide credit note lines, we mark these results with customerCreditNoteId to show they have been credited
INNER JOIN dbo.CustomerInvoiceNumbers n on c.CustomerBillId = n.CustomerBillId
left OUTER JOIN
           dbo.CustomerCreditNotes AS r ON c.CustomerPaymentId = r.CustomerPaymentId
where isnull(c.transactionDateTimeEnd,'')=''

Answer 2

如果您经常使用该查询，并且根据您写入表的频率，可能值得为此创建indexed view。然而值得注意的是，这是猜测，并且索引视图确实带来了折衷，您的读取速度会更快但写入速度会更慢。

CREATE VIEW dbo.CustomerBillingView
WITH SCHEMABINDING
AS
    SELECT  b.CustomerBillId,
            SUM(a.VolumetricCharge) AS Volumetric,
            SUM(a.FixedCharge) AS Fixed,
            SUM(a.VAT) AS VAT,
            SUM(a.Discount) AS Discount,
            SUM(a.EstimatedConsumption) AS Consumption,
            COUNT_BIG(*) AS Records -- REQUIRED TO CREATE INDEX
    FROM    dbo.PremiseProviderBillings a
            INNER JOIN dbo.PremiseProviderBills b
                ON a.PremiseProviderBillId = b.PremiseProviderBillId
    GROUP BY b.CustomerBillId;
GO

CREATE UNIQUE CLUSTERED INDEX UQ_CustomerBillingView__CustomerBillId
    ON dbo.CustomerBillingView (CustomerBillId);

GO

然后，您只需要使用提示NOEXPAND的视图来确保使用索引。

SELECT  n.CustomerInvoiceNumberId as InvoiceNo,
        c.CustomerBillId,
        c.customerid, 
        s.Volumetric, 
        s.Fixed, 
        s.VAT, 
        s.Discount, 
        c.Debit,
        c.EffectiveDate,
        c.TransactionDateTime,
        s.Consumption,
        r.CustomerCreditNoteId
        --,s.Volumetric + s.Fixed + s.Vat - s.discount - c.debit as variance
FROM    CustomerPayments AS c 
        INNER JOIN dbo.CustomerBillingView AS s WITH (NOEXPAND)
            ON c.CustomerBillId = s.CustomerBillId 
            AND c.Debit <> 0 
            -- hide credit note lines, we mark these results with customerCreditNoteId to show they have been credited
        INNER JOIN dbo.CustomerInvoiceNumbers n 
            ON c.CustomerBillId = n.CustomerBillId
        LEFT OUTER JOIN dbo.CustomerCreditNotes AS r 
            ON c.CustomerPaymentId = r.CustomerPaymentId
WHERE   ISNULL(c.transactionDateTimeEnd,'') = '';

与每个查询调优问题一样，您是唯一拥有正确回答所需信息的人。根据我的经验（主要在计费系统中），像这样的索引视图通常可以使用结算数据，因为大多数发票运行都是周期性的，所以写入是批量而不是连续的，并且读取也往往超过写入，因为数据是静态的，一旦创建了发票然后它很少更新。

通过缓慢运行来调整组

2 个答案: