Question

我在SQL Server 2008数据库中有一个非常大的表，包含4000万行。

CREATE TABLE [dbo].[myTable](
    [ID] [bigint] NOT NULL,
    [CONTRACT_NUMBER] [varchar](50) NULL,
    [CUSTOMER_NAME] [varchar](200) NULL,
    [INVOICE_NUMBER] [varchar](50) NULL,
    [AGENCY] [varchar](50) NULL,
    [AMOUNT] [varchar](50) NULL,
    [INVOICE_MONTH] [int] NULL,
    [INVOICE_YEAR] [int] NULL,
    [Unique_ID] [bigint] NULL,
    [bar_code] [varchar](50) NOT NULL,
 CONSTRAINT [PK_MyTable] PRIMARY KEY CLUSTERED 
(
    [ID] ASC,
    [bar_code] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]

我正在尝试优化以下查询的效果：

SELECT top 35  ID,
            CONTRACT_NR,
            CUSTOMER_NAME,
            INVOICE_NUMBER,
            AMOUNT,
            AGENCY,
            CONTRACT_NUMBER,
            ISNULL([INVOICE_MONTH], 1) as [INVOICE_MONTH],
            ISNULL([INVOICE_YEAR], 1) as [INVOICE_YEAR],
            bar_code, 
            Unique_ID
            from MyTable 
WHERE 
CONTRACT_NUMBER like @CONTRACT_NUMBER and
INVOICE_NUMBER like @INVOICE_NUMBER and 
CUSTOMER_NAME like @CUSTOMER_NAME 
ORDER BY Unique_ID desc

为了做到这一点，我在列CONTRACT_NUMBER，INVOICE_NUMBER和CUSTOMER_NAME上构建了一个包含的索引。

CREATE NONCLUSTERED INDEX [ix_search_columns_without_uniqueid] ON [dbo].[MyTable] 
(
    [CONTRACT_NUMBER] ASC,
    [CUSTOMER_NAME] ASC,
    [INVOICE_NUMBER] ASC
)
INCLUDE ( [ID],
[AGENCY],
[AMOUNT],
[INVOICE_MONTH],
[INVOICE_YEAR],
[Unique_ID],
[Contract_nr],
[bar_code]) WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]

仍然需要执行3秒到10秒的查询。从查询执行计划中我看到索引搜索操作正在消耗大约30％的总工作负载而不是排序（前N）操作消耗其他70％。任何想法如何优化此查询，响应时间小于1秒是首选？注意：我还尝试在索引列中包含dhe列[Unique_ID]。在这种情况下，查询执行计划正在进行索引扫描，但是由于许多用户查询数据库，我遇到了同样的问题。

Answer 1

查看此page了解更多详情。

使用完整扫描更新统计信息，以使优化程序更轻松。

UPDATE STATISTICS tablename WITH fullscan GO

设置统计时间并执行以下查询

SET STATISTICS time ON GO SELECT num_of_reads, num_of_bytes_read, num_of_writes, num_of_bytes_written FROM sys.dm_io_virtual_file_stats(DB_ID('tempdb'), 1) GO SELECT TOP 100 c1, c2,c3 FROM yourtablename WHERE c1<30000 ORDER BY c2 GO SELECT num_of_reads, num_of_bytes_read, num_of_writes, num_of_bytes_written FROM sys.dm_io_virtual_file_stats(DB_ID('tempdb'), 1) GO

<强>结果

CPU time = 124 ms,  elapsed time = 91 ms
Before Query execution 
num_of_reads         num_of_bytes_read    num_of_writes     num_of_bytes_written
-------------------- -------------------- -------------------- --------------------
725864               46824931328          793589               51814416384
After Query execution  
num_of_reads         num_of_bytes_read    num_of_writes        num_of_bytes_written
-------------------- -------------------- -------------------- --------------------
725864               46824931328          793589               51814416384

来源：https://www.mssqltips.com/sqlservertip/2053/trick-to-optimize-top-clause-in-sql-server/

Answer 2

尝试将您的聚集索引（当前在两列上）替换为unique_id上的一个（假设它确实是唯一的）。这有助于您的分类。然后在WHERE中使用的三个列上添加第二个覆盖索引 - 正如您所尝试的那样。检查您的统计信息是否是最新的。我有一种感觉，PK中的bar_code列阻止你的排序尽可能快地运行。

您的变量是否包含通配符？如果它们包含通配符，则无法使用WHERE列上的索引。如果他们不是通配符，请尝试直接"="，假设区分大小写不是问题。

更新：由于您拥有前导通配符，因此您将无法利用CONTRACT_NUMBER，INVOICE_NUMBER或CUSTOMER_NAME上的索引：作为GriGrim建议，这里唯一的选择是使用全文搜索（CONTAINS关键字等。）。

如何优化选择前N个查询

2 个答案: