• 利用扩展事件(Xevents)捕捉高消耗查询


    生产环境中有时需要使用者抓取一些特定的语句分析,如超超长查询,或高IO查询等.一般来说大家对跟踪比较熟悉,主要因为其有完善的UI支持.由于扩展事件在sql2012才提供UI支持,所以虽然在08时就已经存在的更轻量级的扩展事件没有被广泛使用.这里和大家分享一个扩展事件脚本,用来捕捉高消耗的查询语句.

    注意:使用完毕后应停止扩展事件捕捉.

           应根据自身实际需求设置过滤条件,如CPU,duration,Reads等

           在sql2012中会提供更为丰富的内容(如query_hash等)

    code 1 将扩展事件追踪保存到本地文件中

              

    CREATE EVENT SESSION L_QUERIES ON SERVER 
    ADD EVENT sqlserver.sql_statement_completed(
        ACTION(sqlserver.username,sqlserver.client_hostname,sqlserver.database_id,sqlserver.session_id,sqlserver.sql_text,sqlserver.plan_handle)
        WHERE (cpu>1000 AND duration>=5000000 and Reads>=10000 and sqlserver.database_id=17)) 
    ADD TARGET package0.asynchronous_file_target
       (SET FILENAME = N'D:L_QUERIES.xel', METADATAFILE = N'D:L_QUERIES.xem')
    WITH (max_dispatch_latency = 1 seconds);
    --cpu 1s
    --duration 5s
    --reads io 10000
    --database_id 17 if you do not need it ,cut out
    GO
    
    ALTER EVENT SESSION L_QUERIES ON SERVER STATE = START;------start the event
    GO 
    
    
    SELECT
       data.value (
          '(/event[@name=''sql_statement_completed'']/@timestamp)[1]', 'DATETIME') AS [Time],
       data.value (
          '(/event/data[@name=''cpu'']/value)[1]', 'INT') AS [CPU (ms)],
       data.value (
          '(/event/data[@name=''reads'']/value)[1]', 'BIGINT') AS [Reads],   
       data.value (
          '(/event/action[@name=''session_id'']/value)[1]', 'smallint') AS [session_id],   
       data.value (
          '(/event/action[@name=''database_id'']/value)[1]', 'smallint') AS [database_id],
       CONVERT (FLOAT, data.value ('(/event/data[@name=''duration'']/value)[1]', 'BIGINT')) / 1000000
          AS [Duration (s)],
       data.value (
            '(/event/action[@name=''username'']/value)[1]', 'VARCHAR(100)') AS [username],
       data.value  (   
            '(/event/action[@name=''client_hostname'']/value)[1]', 'VARCHAR(100)') AS [client_hostname],
       data.value  (      
            '(/event/action[@name=''sql_text'']/value)[1]', 'VARCHAR(MAX)') AS [SQL Statement],
       SUBSTRING (data.value ('(/event/action[@name=''plan_handle'']/value)[1]', 'VARCHAR(100)'), 15, 50)
          AS [Plan Handle]
    FROM
       (SELECT CONVERT (XML, event_data) AS data FROM sys.fn_xe_file_target_read_file
          ('D:L_QUERIES*.xel', 'D:L_QUERIES*.xem', null, null)
    ) entries
    ORDER BY [Time] DESC;----------query the event info what you need
    
    
    --ALTER EVENT SESSION L_QUERIES ON SERVER STATE = STOP;---stop the event
    
    --DROP EVENT SESSION L_QUERIES ON SERVER;------------drop the event

    code2 将扩展事件保存到ring buffer中

    CREATE EVENT SESSION L_QUERIES ON SERVER 
    ADD EVENT sqlserver.sql_statement_completed(
        ACTION(sqlserver.username,sqlserver.client_hostname,sqlserver.database_id,sqlserver.session_id,sqlserver.sql_text,sqlserver.plan_handle)
        WHERE (cpu>1000 AND duration>=5000000 and Reads>=10000 and sqlserver.database_id=17)) 
        ADD TARGET package0.ring_buffer
            (SET max_memory = 4000) -----not the default 4096kb
       WITH (max_dispatch_latency = 1 seconds) 
    --cpu 1s
    --duration 5s
    --reads io 10000
    --database_id 17 if you do not need it ,cut out
    GO
    
    ALTER EVENT SESSION L_QUERIES ON SERVER STATE = START;------start the event
    GO 
    
    
    
    DECLARE @target_data XML
    SELECT  @target_data=CAST(target_data AS XML)
    FROM sys.dm_xe_sessions AS s 
    JOIN sys.dm_xe_session_targets AS t 
        ON t.event_session_address = s.address
    WHERE s.name = 'L_QUERIES'
      AND t.target_name = 'ring_buffer'
       
    SELECT
       t.value('(@timestamp)[1]', 'datetime2') AS [time],
       t.value ('(data[@name=''cpu'']/value)[1]', 'INT') AS [CPU (ms)],
       t.value ('(data[@name=''reads'']/value)[1]', 'BIGINT') AS [Reads],   
       t.value ('(action[@name=''session_id'']/value)[1]', 'smallint') AS [session_id],   
       t.value ('(action[@name=''database_id'']/value)[1]', 'smallint') AS [database_id],
       CONVERT (FLOAT, t.value ('(data[@name=''duration'']/value)[1]', 'BIGINT')) / 1000000
          AS [Duration (s)],
       t.value ('(action[@name=''username'']/value)[1]', 'VARCHAR(100)') AS [username],
       t.value ('(action[@name=''client_hostname'']/value)[1]', 'VARCHAR(100)') AS [client_hostname],
       t.value ('(action[@name=''sql_text'']/value)[1]', 'VARCHAR(MAX)') AS [SQL Statement],
       SUBSTRING (t.value ('(action[@name=''plan_handle'']/value)[1]', 'VARCHAR(100)'), 15, 50)
          AS [Plan Handle]
        FROM @target_data.nodes('RingBufferTarget/event') AS q(t)
    ORDER BY [Time] DESC;----------query the event info what you need
    
    --ALTER EVENT SESSION L_QUERIES ON SERVER STATE = STOP;---stop the event
    
    --DROP EVENT SESSION L_QUERIES ON SERVER-----------drop the event
  • 相关阅读:
    面试题21 包含min函数的栈
    面试题20 顺时针打印矩阵
    基于熵的方法计算query与docs相似度
    使用信息检索和深度学习方法的智能对话
    常用的激活函数
    spark实现smote近邻采样
    wide&deep用于ltr排序
    deepfm用于ltr排序
    lightgbm用于排序
    静态工厂方法+服务提供者框架模板
  • 原文地址:https://www.cnblogs.com/shanksgao/p/3962133.html
Copyright © 2020-2023  润新知