Question

没有太多细节，我有一个查询使用所有聚集和非聚集索引搜索产生执行计划（听起来很有希望）。不幸的是，查询表现不佳，我很难理解为什么。

我正在使用set statistics io on并且可以看到其中一个表产生了大量扫描和逻辑/物理读取：

SET statistics io ON

go

    SELECT order_number,
           audit_id,
           orderadmission_net_paid_delta / 100.00,
           'Admission',
           orderadmission_net_paid_delta / 100.00,
           performance_gl_description1,
           section_data1,
           performance_gl_code,
           price_type_data1,
           year(performance_start_date),
           month(performance_start_date),
           paymentmethod_type,
           paymentmethod_name,
           ''
    FROM   JCRProdReplication.dbo.ts_audit WITH (NOLOCK)
           JOIN JCRProdReplication.dbo.ts_order_admission WITH (NOLOCK)
             ON orderadmission_audit_id = audit_id
           LEFT JOIN JCRProdReplication.dbo.ts_order WITH (NOLOCK)
             ON order_id = orderadmission_order_id
           LEFT JOIN JCRProdReplication.dbo.ts_performance WITH (NOLOCK)
             ON performance_id = orderadmission_performance_id
           LEFT JOIN JCRProdReplication.dbo.ts_seat WITH (NOLOCK)
             ON seat_id = orderadmission_seat_id
           LEFT JOIN JCRProdReplication.dbo.ts_section WITH (NOLOCK)
             ON section_id = seat_section_id
           LEFT JOIN JCRProdReplication.dbo.ts_price_type WITH (NOLOCK)
             ON price_type_id = orderadmission_price_type_id
           LEFT JOIN JCRProdReplication.dbo.ts_order_payment WITH (NOLOCK)
             ON orderpayment_audit_id = audit_id
           LEFT JOIN JCRProdReplication.dbo.ts_payment_method WITH (NOLOCK)
             ON paymentmethod_id = orderpayment_paymentmethod_id
    WHERE  audit_time >= '20140107'
           AND audit_time < '20140108'

(72174 row(s) affected)
Table 'ts_payment_method'. Scan count 0, logical reads 4180, physical reads 1, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'ts_price_type'. Scan count 0, logical reads 4184, physical reads 26, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'ts_section'. Scan count 0, logical reads 4184, physical reads 28, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'ts_seat'. Scan count 0, logical reads 6276, physical reads 2240, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'ts_performance'. Scan count 0, logical reads 4184, physical reads 50, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'ts_order'. Scan count 0, logical reads 8368, physical reads 820, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'ts_order_admission'. Scan count 71877, logical reads 288490, physical reads 44104, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'ts_audit'. Scan count 1, logical reads 252, physical reads 5, read-ahead reads 246, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

我理解为什么索引搜索会在统计中显示大量扫描/读取的下一步是什么？（我一整天都在谷歌搜索，没有找到解释）。

Answer 1

STATISTICS IO报告的扫描计数可能与您的想法不符。

实施例

USE tempdb;
SET NOCOUNT ON;

CREATE TABLE Num1 (N INT PRIMARY KEY);

CREATE TABLE Num2 (N INT);
CREATE CLUSTERED INDEX IX ON Num2(N);

INSERT INTO Num1
OUTPUT      inserted.N
INTO Num2
SELECT number
FROM   master..spt_values
WHERE  type = 'P'
       AND number BETWEEN 1 AND 1000

SET STATISTICS IO ON;

SELECT *, sys.fn_PhysLocFormatter(N2.%%physloc%%)
FROM   Num1 N1
       INNER LOOP JOIN Num2 N2
         ON N1.N = N2.N

SET STATISTICS IO OFF;

DROP TABLE Num1, Num2

提供计划

enter image description here

输出

Table 'Num2'. Scan count 1000, logical reads 2002
Table 'Num1'. Scan count 1, logical reads 3

尽管该计划显示有1000次扫描报告。

由于索引未声明为唯一，因此每次搜索都是部分扫描。索引被搜索到1000次，然后在每次搜索时需要扫描索引，直到找到第一行与该数字不匹配为止。

Num2.IX索引有一个根页面和两个叶子页面。

2,002个逻辑读取分解如下。

大多数搜索都需要两次读取（单个根页面和单个叶子页面）。其中两个搜索（对于我的情况下的数字622和623）在读取两个叶页时都进行了三次逻辑读取。

这些是第一页叶子页面的最后一行，第一页面分别是第一行。

对于数字622，它需要读取下一页的下一行以查看它是否重复。对于数字623，解释是索引的结构如下。

enter image description here

根页面具有可能包含在每个较低页面上的最低值。由于索引未声明为唯一，因此如果跨越页面边界的多个匹配行，则需要检查上一页。

以下博客文章提供了有关解释STATISTICS IO输出的其他说明。

为什么索引要在统计中显示高扫描计数？

1 个答案: