如果正确读取SQL Server中的扩展事件？

    SQL Server中使用扩展事件捕捉所需的信息后，可以选择存放的位置。比如说内存或文件中，但无论存在哪里，其本质都是一个大XML。因此在SQL Server中读取该XML就是解析扩展事件结果的方式。

    微软官方或者一些SQL Server论坛提供了使用SQL XML解析扩展事件的脚本，如代码清单1所示。

   1: WITH    events_cte

   2:           AS ( SELECT   DATEADD(mi,

   3:                                 DATEDIFF(mi, GETUTCDATE(), CURRENT_TIMESTAMP),

   4:                                 xevents.event_data.value('(event/@timestamp)[1]',

   5:                                                          'datetime2')) AS [event time] ,

   6:                                                             xevents.event_data.value('(event/@name)[1]',

   7:                                                  'nvarchar(128)') AS [Event Name],

   8:                         xevents.event_data.value('(event/action[@name="client_app_name"]/value)[1]',

   9:                                                  'nvarchar(128)') AS [client app name] ,

  10:                         xevents.event_data.value('(event/action[@name="client_hostname"]/value)[1]',

  11:                                                  'nvarchar(max)') AS [client host name] ,

  12:                         xevents.event_data.value('(event/action[@name="sql_text"]/value)[1]',

  13:                                                  'nvarchar(max)') AS [sql_text] ,

  14:           

  15:                         xevents.event_data.value('(event/action[@name="database_name"]/value)[1]',

  16:                                                  'nvarchar(max)') AS [database name] ,

  17:                         xevents.event_data.value('(event/action[@name="username"]/value)[1]',

  18:                                                  'nvarchar(max)') AS [username] ,

  19:                         xevents.event_data.value('(event/action[@name="duration"]/value)[1]',

  20:                                                  'bigint') AS [duration (ms)] ,

  21:                         xevents.event_data.value('(event/action[@name="cpu_time"]/value)[1]',

  22:                                                  'bigint') AS [cpu time (ms)] ,

  23:                         xevents.event_data.value('(event/data[@name="object_name"]/value)[1]',

  24:                                                  'nvarchar(max)') AS [OBJECT_NAME]

  25:                FROM     sys.fn_xe_file_target_read_file('D:\XeventResutl\DDLAudit*.xel',

  26:                                                         NULL, NULL, NULL)

  27:                         CROSS APPLY ( SELECT    CAST(event_data AS XML) AS event_data

  28:                                     ) AS xevents

  29:              )

  30:     SELECT  *

  31:     FROM    events_cte

  32:     ORDER BY [event time] DESC;

代码清单1.读取扩展事件文件的脚本

    但代码清单1的脚本使用的是XQuery，XQuery在使用Xml的节点属性作为删选条件时，数据上千以后就会变得非常慢。因此我对上述脚本进行了改写，将XML读取出来后，变为节点的集合以关系数据格式存放，再用子查询进行筛选，这种方式读取数据基本上是秒出，如代码清单2所示。

   1: WITH   tt

   2:          AS ( SELECT   MIN(event_name) AS event_name ,

   3:               DATEADD(hh,DATEDIFF(hh, GETUTCDATE(), CURRENT_TIMESTAMP),

   4:                                CONVERT(DATETIME, MIN(CASE WHEN d_name = 'collect_system_time'

   5:                                                          AND d_package IS NOT NULL THEN d_value

   6:                                                      END))) AS [event_timestamp] ,

   7:                        CONVERT 

   8:        (VARCHAR(MAX), MIN(CASE WHEN  d_name = 'client_hostname'

   9:                                     AND d_package IS NOT NULL THEN d_value

  10:                           END)) AS [Client_hostname] ,

  11:                        CONVERT 

  12:        (VARCHAR(MAX), MIN(CASE WHEN --event_name = 'sql_batch_completed'

  13:                                d_name = 'client_app_name'

  14:                               THEN d_value

  15:                     END)) AS [Client_app_name] ,

  16:                        CONVERT 

  17:        (VARCHAR(MAX), MIN(CASE WHEN  d_name = 'database_name'

  18:                                     AND d_package IS NOT NULL THEN d_value

  19:                           END)) AS [database_name] ,

  20:                           CONVERT

  21:                                   (VARCHAR(MAX), MIN(CASE WHEN  d_name = 'object_name'

  22:                                      THEN d_value

  23:                           END)) AS [object_name] ,

  24:                        CONVERT 

  25:        (BIGINT, MIN(CASE WHEN event_name = 'sql_batch_completed'

  26:                               AND d_name = 'duration'

  27:                               AND d_package IS NULL THEN d_value

  28:                     END)) AS [sql_statement_completed.duration] ,

  29:             

  30:                        CONVERT 

  31:        (VARCHAR(MAX), MIN(CASE WHEN d_name = 'sql_text'

  32:                                      THEN d_value

  33:                           END)) AS [sql_statement_completed.sql_text] ,

  34:                        CONVERT 

  35:        (VARCHAR(MAX), MIN(CASE WHEN d_name = 'username'

  36:                                     AND d_package IS NOT NULL THEN d_value

  37:                           END)) AS [username] 

  38:               FROM     ( SELECT    * ,

  39:                                    CONVERT(VARCHAR(400), NULL) AS attach_activity_id

  40:                          FROM      ( SELECT    event.value('(@name)[1]',

  41:                                                            'VARCHAR(400)') AS event_name ,

  42:                                                DENSE_RANK() OVER ( ORDER BY event ) AS unique_event_id ,

  43:                                                n.value('(@name)[1]',

  44:                                                        'VARCHAR(400)') AS d_name ,

  45:                                                n.value('(@package)[1]',

  46:                                                        'VARCHAR(400)') AS d_package ,

  47:                                                n.value('((value)[1]/text())[1]',

  48:                                                        'VARCHAR(MAX)') AS d_value ,

  49:                                                n.value('((text)[1]/text())[1]',

  50:                                                        'VARCHAR(MAX)') AS d_text

  51:                                      FROM      ( SELECT    ( SELECT

  52:                                                              CONVERT(XML, target_data)

  53:                                                              FROM

  54:                                                              sys.dm_xe_session_targets st

  55:                                                              JOIN sys.dm_xe_sessions s ON s.address = st.event_session_address

  56:                                                              WHERE

  57:                                                              s.name = 'DDL'

  58:                                                              AND st.target_name = 'ring_buffer'

  59:                                                            ) AS [x]

  60:                                                FOR

  61:                                                  XML PATH('') ,

  62:                                                      TYPE

  63:                                                ) AS the_xml ( x )

  64:                                                CROSS APPLY x.nodes('//event') e ( event )

  65:                                                CROSS APPLY event.nodes('*')

  66:                                                AS q ( n )

  67:                                    ) AS data_data

  68:                        ) AS activity_data

  69:               GROUP BY unique_event_id

  70:             )

  71:    SELECT  *

  72:    FROM    tt

  73:  

代码清单2.对扩展事件结果的优化读取方式

参考资料：http://blog.wharton.com.au/2011/06/13/part-5-openxml-and-xquery-optimisation-tips/