The Problem

索引一直是优化查询性能的不二法门。其中一个最直接的问题便是当审查一个低性能查询语句时，检查索引是否在正确的地方或者加索引没有。运行一个batchjob查看索引碎片，必要时采取步骤优化索引碎片是日常维护程序中不可缺少的。

今天的主题便是如何判定数据库中的索引碎片和优化措施

我们经常会用到sys.dm_db_index_physical_stats表来查看索引信息

示例：

USE AdventureWorks2014
GO

DECLARE @database_name VARCHAR(100) = ‘AdventureWorks2014‘;
SELECT  SD.name AS database_name ,
        SO.name AS object_name ,
        SI.name AS index_name ,
        IPS.index_type_desc ,
        IPS.page_count ,
        IPS.avg_fragmentation_in_percent
FROM    sys.dm_db_index_physical_stats(NULL, NULL, NULL, NULL, NULL) IPS
        INNER JOIN sys.databases SD ON SD.database_id = IPS.database_id
        INNER JOIN sys.indexes SI ON SI.index_id = IPS.index_id
        INNER JOIN sys.objects SO ON SO.object_id = SI.object_id
                                     AND IPS.object_id = SO.object_id
WHERE   alloc_unit_type_desc = ‘IN_ROW_DATA‘
        AND index_level = 0
        AND SD.name = @database_name
ORDER BY IPS.avg_fragmentation_in_percent DESC;

当知道索引碎片信息以后，我们应该如何优化呢？或者在任何数据库中查看索引碎片和优化呢？

The Solution

优化索引有两种方案

• Index Rebuild：

  When an index is rebuilt , it is completely replaced with a new copy of the index, built from scratch as though
  it were just newly created. In SQL Server Standard edition, this is an offline operation, meaning that it can
  cause contention while running.

• Index Reorganization： 
  
  Reorganizing an index results in cleanup at the leaf level, reordering pages and reapplying the fill factor as
  necessary. This operation is always online, regardless of the edition of SQL Server you are running and can
  be interrupted at any time with no ill effects.

现在我们通过百分比来判定索引是rebulid还是reorganization

IF OBJECT_ID(‘dbo.index_maintenance_demo‘,‘P‘) IS NOT NULL
BEGIN
    DROP PROCEDURE dbo.index_maintenance_demo;
END

SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO

CREATE PROCEDURE dbo.index_maintenance_demo
    @reorganization_percentage TINYINT = 10 ,
    @rebuild_percentage TINYINT = 35 ,
    @print_results_only BIT = 1
AS
    BEGIN
        DECLARE @sql_command NVARCHAR(MAX) = ‘‘;
        DECLARE @parameter_list NVARCHAR(MAX) = ‘@reorganization_percentage TINYINT, @rebuild_percentage TINYINT‘
        DECLARE @database_name NVARCHAR(MAX);
        DECLARE @database_list TABLE
            (
              database_name NVARCHAR(MAX) NOT NULL
            );
        INSERT  INTO @database_list
                ( database_name
                )
                SELECT  name
                FROM    sys.databases
                WHERE   databases.name NOT IN ( ‘msdb‘, ‘master‘, ‘TempDB‘,
                                                ‘model‘,‘ReportServer$SQL2014‘ );
        CREATE TABLE #index_maintenance
            (
              database_name NVARCHAR(MAX) ,
              schema_name NVARCHAR(MAX) ,
              object_name NVARCHAR(MAX) ,
              index_name NVARCHAR(MAX) ,
              index_type_desc NVARCHAR(MAX) ,
              page_count BIGINT ,
              avg_fragmentation_in_percent FLOAT ,
              index_operation NVARCHAR(MAX)
            );

        SELECT  @sql_command = @sql_command + ‘
USE [‘ + database_name + ‘]

INSERT  INTO #index_maintenance
        ( database_name ,
          schema_name ,
          object_name ,
          index_name ,
          index_type_desc ,
          page_count ,
          avg_fragmentation_in_percent ,
          index_operation
        )
        SELECT  CAST(SD.name AS NVARCHAR(MAX)) AS database_name ,
                CAST(SS.name AS NVARCHAR(MAX)) AS schema_name ,
                CAST(SO.name AS NVARCHAR(MAX)) AS object_name ,
                CAST(SI.name AS NVARCHAR(MAX)) AS index_name ,
                IPS.index_type_desc ,
                IPS.page_count ,
                -- Be sure to filter as much as possible...this can return a lot of data if you dont filter by database and table.
                IPS.avg_fragmentation_in_percent ,
                CAST(CASE WHEN IPS.avg_fragmentation_in_percent >= @rebuild_percentage
                          THEN ‘‘REBUILD‘‘
                          WHEN IPS.avg_fragmentation_in_percent >= @reorganization_percentage
                          THEN ‘‘REORGANIZE‘‘
                     END AS NVARCHAR(MAX)) AS index_operation
        FROM    sys.dm_db_index_physical_stats(NULL, NULL, NULL, NULL, NULL) IPS
                INNER JOIN sys.databases SD ON SD.database_id = IPS.database_id
                INNER JOIN sys.indexes SI ON SI.index_id = IPS.index_id
                INNER JOIN sys.objects SO ON SO.object_id = SI.object_id
                                             AND IPS.object_id = SO.object_id
                INNER JOIN sys.schemas SS ON SS.schema_id = SO.schema_id
        WHERE   alloc_unit_type_desc = ‘‘IN_ROW_DATA‘‘
                AND index_level = 0
                AND SD.name = ‘‘‘ + database_name + ‘‘‘
                AND IPS.avg_fragmentation_in_percent >= @reorganization_percentage
                AND SI.name IS NOT NULL -- Only review index, not heap data.
                AND SO.is_ms_shipped = 0 -- Do not perform maintenance on system objects
                ORDER BY SD.name ASC;‘
        FROM    @database_list
        WHERE   database_name IN ( SELECT    name    FROM    sys.databases );

        EXEC sp_executesql @sql_command, @parameter_list,
            @reorganization_percentage, @rebuild_percentage;

        SELECT  @sql_command = ‘‘;
        SELECT  @sql_command = @sql_command + ‘
            USE ‘ +  QUOTENAME(database_name) + ‘
            ALTER INDEX ‘ + QUOTENAME(index_name) + ‘ ON ‘ + QUOTENAME(schema_name) + ‘.‘ + QUOTENAME(object_name) + ‘ ‘
                + index_operation + ‘;‘
        FROM    #index_maintenance;

        SELECT  *
        FROM    #index_maintenance
        ORDER BY avg_fragmentation_in_percent DESC;

        IF @print_results_only = 1
        BEGIN
            PRINT @sql_command;
        END
        ELSE
        BEGIN
            EXEC sp_executesql @sql_command;
        END

        DROP TABLE #index_maintenance;
    END
GO

然后运行SP得到以下结果

EXEC dbo.index_maintenance_demo @reorganization_percentage = 10, @rebuild_percentage = 35,
@print_results_only = 1;

打印出来的SQL如下：

USE [AdventureWorks2014]
ALTER INDEX [PK_ProductCostHistory_ProductID_StartDate] ON [Production].[ProductCostHistory] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [AK_ProductDescription_rowguid] ON [Production].[ProductDescription] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [PK_DatabaseLog_DatabaseLogID] ON [dbo].[DatabaseLog] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [PK_ProductInventory_ProductID_LocationID] ON [Production].[ProductInventory] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [PK_ProductListPriceHistory_ProductID_StartDate] ON [Production].[ProductListPriceHistory] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [PK_SpecialOfferProduct_SpecialOfferID_ProductID] ON [Sales].[SpecialOfferProduct] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [AK_SpecialOfferProduct_rowguid] ON [Sales].[SpecialOfferProduct] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [PK_StateProvince_StateProvinceID] ON [Person].[StateProvince] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [PK_ProductModelProductDescriptionCulture_ProductModelID_ProductDescriptionID_CultureID] ON [Production].[ProductModelProductDescriptionCulture] REBUILD;
USE [AdventureWorks2014]
ALTER INDEX [AK_BillOfMaterials_ProductAssemblyID_ComponentID_StartDate] ON [Production].[BillOfMaterials] REORGANIZE;

有了这个SP，以后我们的维护工作就轻便了许多。