首页 > 代码库 > 人人都是 DBA(XIII)索引信息收集脚本汇编

人人都是 DBA(XIII)索引信息收集脚本汇编

原文:人人都是 DBA(XIII)索引信息收集脚本汇编

什么?有个 SQL 执行了 8 秒!

哪里出了问题?臣妾不知道啊,得找 DBA 啊。

DBA 人呢?离职了!!擦!!!

程序员在无处寻求帮助时,就得想办法自救,努力让自己变成 "伪 DBA"。

索引

  1. 找出哪些表的 Index 需要改进
  2. 在指定数据库中查找哪些表的 Index 需要改进
  3. 根据缓存的查询计划判断 SP 是否需要优化
  4. 发现那些 Index 的写远多于读的表 
  5. 查看 Index 的 Statistics 最后更新时间
  6. 查看哪些 Index 被修改的最频繁
  7. 查看 Index 碎片化指数
  8. 哪个 Index 上的读操作最活跃
  9. 哪个 Index 上的写操作最活跃
  10. 查看 Index 所使用的 Buffer 数量
  11. 按照 IO Latch 等待请求对索引进行排行

找出哪些表的 Index 需要改进

SELECT CONVERT(DECIMAL(18, 2), user_seeks * avg_total_user_cost * (avg_user_impact * 0.01)) AS [index_advantage]    ,migs.last_user_seek    ,mid.[statement] AS [Database.Schema.Table]    ,mid.equality_columns    ,mid.inequality_columns    ,mid.included_columns    ,migs.unique_compiles    ,migs.user_seeks    ,migs.avg_total_user_cost    ,migs.avg_user_impactFROM sys.dm_db_missing_index_group_stats AS migs WITH (NOLOCK)INNER JOIN sys.dm_db_missing_index_groups AS mig WITH (NOLOCK) ON migs.group_handle = mig.index_group_handleINNER JOIN sys.dm_db_missing_index_details AS mid WITH (NOLOCK) ON mig.index_handle = mid.index_handleORDER BY index_advantage DESCOPTION (RECOMPILE);

这里查询出的数据,只是说明数据寻址时间有点儿长,不一定就是缺少索引所引起的。

在指定数据库中查找哪些表的 Index 需要改进

SELECT DISTINCT CONVERT(DECIMAL(18, 2), user_seeks * avg_total_user_cost * (avg_user_impact * 0.01)) AS [index_advantage]    ,migs.last_user_seek    ,mid.[statement] AS [Database.Schema.Table]    ,mid.equality_columns    ,mid.inequality_columns    ,mid.included_columns    ,migs.unique_compiles    ,migs.user_seeks    ,migs.avg_total_user_cost    ,migs.avg_user_impact    ,OBJECT_NAME(mid.[object_id]) AS [Table Name]    ,p.rows AS [Table Rows]FROM sys.dm_db_missing_index_group_stats AS migs WITH (NOLOCK)INNER JOIN sys.dm_db_missing_index_groups AS mig WITH (NOLOCK) ON migs.group_handle = mig.index_group_handleINNER JOIN sys.dm_db_missing_index_details AS mid WITH (NOLOCK) ON mig.index_handle = mid.index_handleINNER JOIN sys.partitions AS p WITH (NOLOCK) ON p.[object_id] = mid.[object_id]WHERE mid.database_id = DB_ID()ORDER BY index_advantage DESCOPTION (RECOMPILE);

根据缓存的查询计划判断 SP 是否需要优化

SELECT TOP (25) OBJECT_NAME(objectid) AS [ObjectName]    ,query_plan    ,cp.objtype    ,cp.usecountsFROM sys.dm_exec_cached_plans AS cp WITH (NOLOCK)CROSS APPLY sys.dm_exec_query_plan(cp.plan_handle) AS qpWHERE CAST(query_plan AS NVARCHAR(MAX)) LIKE N%MissingIndex%    AND dbid = DB_ID()ORDER BY cp.usecounts DESCOPTION (RECOMPILE);

发现那些 Index 的写远多于读的表

SELECT OBJECT_NAME(s.[object_id]) AS [Table Name]    ,i.[name] AS [Index Name]    ,i.index_id    ,i.is_disabled    ,i.is_hypothetical    ,i.has_filter    ,i.fill_factor    ,user_updates AS [Total Writes]    ,user_seeks + user_scans + user_lookups AS [Total Reads]    ,user_updates - (user_seeks + user_scans + user_lookups) AS [Difference]FROM sys.dm_db_index_usage_stats AS s WITH (NOLOCK)INNER JOIN sys.indexes AS i WITH (NOLOCK) ON s.[object_id] = i.[object_id]    AND i.index_id = s.index_idWHERE OBJECTPROPERTY(s.[object_id], IsUserTable) = 1    AND s.database_id = DB_ID()    AND user_updates > (user_seeks + user_scans + user_lookups)    AND i.index_id > 1ORDER BY [Difference] DESC    ,[Total Writes] DESC    ,[Total Reads] ASCOPTION (RECOMPILE);

由于对索引的写操作远多于读操作,看起来 Index 的帮助不大,但需要根据业务需求来判断是否能够 Drop 掉该索引。

查看 Index 的 Statistics 最后更新时间

SELECT SCHEMA_NAME(o.[schema_id]) + N. + o.[name] AS [Object Name]    ,o.type_desc AS [Object Type]    ,i.[name] AS [Index Name]    ,STATS_DATE(i.[object_id], i.index_id) AS [Statistics Date]    ,s.auto_created    ,s.no_recompute    ,s.user_created    ,st.row_count    ,st.used_page_countFROM sys.objects AS o WITH (NOLOCK)INNER JOIN sys.indexes AS i WITH (NOLOCK) ON o.[object_id] = i.[object_id]INNER JOIN sys.stats AS s WITH (NOLOCK) ON i.[object_id] = s.[object_id]    AND i.index_id = s.stats_idINNER JOIN sys.dm_db_partition_stats AS st WITH (NOLOCK) ON o.[object_id] = st.[object_id]    AND i.[index_id] = st.[index_id]WHERE o.[type] IN (        U        ,V        )    AND st.row_count > 0ORDER BY STATS_DATE(i.[object_id], i.index_id) DESCOPTION (RECOMPILE);

参考资料:

  • Statistics
  • UPDATE STATISTICS (Transact-SQL)
  • sp_updatestats (Transact-SQL)
  • Rebuilding Indexes vs. Updating Statistics
  • Does a re-index update statistics?
  • SQL Server Index and Statistics Maintenance

查看哪些 Index 被修改的最频繁

SQL Server 2008 R2

SELECT TableName = OBJECT_NAME(s.[object_id])    ,SchemaName = SCHEMA_NAME(o.[schema_id])    ,IndexName = i.[name]    ,user_updates    ,i.is_primary_keyFROM sys.dm_db_index_usage_stats sJOIN sys.objects O ON s.[object_id] = O.[object_id]JOIN sys.indexes i ON s.[object_id] = i.[object_id]    AND s.index_id = i.index_idWHERE OBJECTPROPERTY(s.[object_id], IsMsShipped) = 0    AND user_seeks = 0    AND user_scans = 0    AND user_lookups = 0    AND i.NAME IS NOT NULL -- Ignore HEAP indexes.ORDER BY user_updates DESC

The user_updates counter indicates the level of maintenance on the index caused by insert, update, or delete operations on the underlying table or view.

SQL Server 2012

SELECT o.[name] AS [Object Name]    ,o.[object_id]    ,o.type_desc    ,s.[name] AS [Statistics Name]    ,s.stats_id    ,s.no_recompute    ,s.auto_created    ,sp.modification_counter    ,sp.rows    ,sp.rows_sampled    ,sp.last_updatedFROM sys.objects AS o WITH (NOLOCK)INNER JOIN sys.stats AS s WITH (NOLOCK) ON s.object_id = o.object_idCROSS APPLY sys.dm_db_stats_properties(s.object_id, s.stats_id) AS spWHERE o.type_desc NOT IN (        NSYSTEM_TABLE        ,NINTERNAL_TABLE        )    AND sp.modification_counter > 0ORDER BY sp.modification_counter DESC    ,o.[name]OPTION (RECOMPILE);

查看 Index 碎片化指数

SELECT DB_NAME(ps.database_id) AS [Database Name]    ,OBJECT_NAME(ps.[object_id]) AS [Object Name]    ,i.[name] AS [Index Name]    ,ps.index_id    ,ps.index_type_desc    ,ps.avg_fragmentation_in_percent    ,ps.fragment_count    ,ps.page_count    ,i.fill_factor    ,i.has_filter    ,i.filter_definitionFROM sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, NLIMITED) AS psINNER JOIN sys.indexes AS i WITH (NOLOCK) ON ps.[object_id] = i.[object_id]    AND ps.index_id = i.index_idWHERE ps.database_id = DB_ID()    AND ps.page_count > 2500ORDER BY ps.avg_fragmentation_in_percent DESCOPTION (RECOMPILE);

参考资料:

  • Stop Worrying About SQL Server Fragmentation
  • Importance of index maintenance
  • Reorganize and Rebuild Indexes
  • Fragmentation and Index Maintenance Tips
  • Index Fragmentation–"If it isn’t broken, don’t fix it"

哪个 Index 上的读操作最活跃

SELECT OBJECT_NAME(s.[object_id]) AS [ObjectName]    ,i.[name] AS [IndexName]    ,i.index_id    ,user_seeks + user_scans + user_lookups AS [Reads]    ,s.user_updates AS [Writes]    ,i.type_desc AS [IndexType]    ,i.fill_factor AS [FillFactor]    ,i.has_filter    ,i.filter_definition    ,s.last_user_scan    ,s.last_user_lookup    ,s.last_user_seekFROM sys.dm_db_index_usage_stats AS s WITH (NOLOCK)INNER JOIN sys.indexes AS i WITH (NOLOCK) ON s.[object_id] = i.[object_id]WHERE OBJECTPROPERTY(s.[object_id], IsUserTable) = 1    AND i.index_id = s.index_id    AND s.database_id = DB_ID()ORDER BY user_seeks + user_scans + user_lookups DESCOPTION (RECOMPILE);

哪个 Index 上的写操作最活跃

SELECT OBJECT_NAME(s.[object_id]) AS [ObjectName]    ,i.[name] AS [IndexName]    ,i.index_id    ,s.user_updates AS [Writes]    ,user_seeks + user_scans + user_lookups AS [Reads]    ,i.type_desc AS [IndexType]    ,i.fill_factor AS [FillFactor]    ,i.has_filter    ,i.filter_definition    ,s.last_system_update    ,s.last_user_updateFROM sys.dm_db_index_usage_stats AS s WITH (NOLOCK)INNER JOIN sys.indexes AS i WITH (NOLOCK) ON s.[object_id] = i.[object_id]WHERE OBJECTPROPERTY(s.[object_id], IsUserTable) = 1    AND i.index_id = s.index_id    AND s.database_id = DB_ID()ORDER BY s.user_updates DESCOPTION (RECOMPILE);

查看 Index 所使用的 Buffer 数量

SELECT TOP 25 obj.[name] AS TableName    ,i.[name] AS IndexName    ,i.[type_desc] AS IndexType    ,count(*) AS Buffered_Page_Count    ,count(*) * 8192 / (1024 * 1024) AS Buffer_MB    ,obj.index_idFROM sys.dm_os_buffer_descriptors AS bdINNER JOIN (    SELECT object_name(object_id) AS NAME        ,index_id        ,allocation_unit_id        ,object_id    FROM sys.allocation_units AS au    INNER JOIN sys.partitions AS p ON au.container_id = p.hobt_id        AND (            au.type = 1            OR au.type = 3            )        UNION ALL        SELECT object_name(object_id) AS NAME        ,index_id        ,allocation_unit_id        ,object_id    FROM sys.allocation_units AS au    INNER JOIN sys.partitions AS p ON au.container_id = p.hobt_id        AND au.type = 2    ) AS obj ON bd.allocation_unit_id = obj.allocation_unit_idLEFT JOIN sys.indexes i ON i.object_id = obj.object_id    AND i.index_id = obj.index_idWHERE database_id = db_id()GROUP BY obj.NAME    ,obj.index_id    ,i.[name]    ,i.[type_desc]ORDER BY Buffered_Page_Count DESC

按照 IO Latch 等待请求对索引进行排行

SELECT OBJECT_SCHEMA_NAME(ios.object_id) + . + OBJECT_NAME(ios.object_id) AS table_name    ,i.[name] AS index_name    ,page_io_latch_wait_count    ,page_io_latch_wait_in_ms    ,CAST(1. * page_io_latch_wait_in_ms / NULLIF(page_io_latch_wait_count, 0) AS DECIMAL(12, 2)) AS page_io_avg_lock_wait_ms    ,page_latch_wait_count    ,page_latch_wait_in_ms    ,CAST(1. * page_latch_wait_in_ms / NULLIF(page_latch_wait_count, 0) AS DECIMAL(12, 2)) AS page_avg_lock_wait_msFROM sys.dm_db_index_operational_stats(DB_ID(), NULL, NULL, NULL) iosINNER JOIN sys.indexes i ON i.object_id = ios.object_id    AND i.index_id = ios.index_idWHERE OBJECTPROPERTY(ios.object_id, IsUserTable) = 1ORDER BY 3 DESC

 

《人人都是 DBA》系列文章索引:

 序号 

 名称 

1

 人人都是 DBA(I)SQL Server 体系结构

2

 人人都是 DBA(II)SQL Server 元数据

3

 人人都是 DBA(III)SQL Server 调度器

4

 人人都是 DBA(IV)SQL Server 内存管理

5

 人人都是 DBA(V)SQL Server 数据库文件

6

 人人都是 DBA(VI)SQL Server 事务日志

7

 人人都是 DBA(VII)B 树和 B+ 树

8

 人人都是 DBA(VIII)SQL Server 页存储结构

9

 人人都是 DBA(IX)服务器信息收集脚本汇编

10

 人人都是 DBA(X)资源信息收集脚本汇编

11

 人人都是 DBA(XI)I/O 信息收集脚本汇编

12

 人人都是 DBA(XII)查询信息收集脚本汇编

13

 人人都是 DBA(XIII)索引信息收集脚本汇编

14

 人人都是 DBA(XIV)存储过程信息收集脚本汇编 

15

 人人都是 DBA(XV)锁信息收集脚本汇编

本系列文章《人人都是 DBA》由 Dennis Gao 发表自博客园,未经作者本人同意禁止任何形式的转载,任何自动或人为的爬虫转载行为均为耍流氓。

人人都是 DBA(XIII)索引信息收集脚本汇编