SQLSERVER排查CPU占用高的情况

操作系统是Windows2008R2 ,数据库是SQL2008R2 64位 64G内存，16核CPU 硬件配置还是比较高的，他说服务器运行的是金蝶K3软件，数据库实例里有多个数据库 他说是这几天才出现的，而且在每天的某一个时间段才会出现CPU占用高的情况 内存占用也很高，占用了30个G

操作系统是Windows2008R2 ,数据库是SQL2008R2 64位

64G内存，16核CPU

硬件配置还是比较高的，他说服务器运行的是金蝶K3软件，数据库实例里有多个数据库

他说是这几天才出现的，而且在每天的某一个时间段才会出现CPU占用高的情况

内存占用也很高，占用了30个G

-----------------------------------------------华丽的分割线-------------------------------------------------------

一般排查都是用下面的脚本，一般会用到三个视图sys.sysprocesses ,dm_exec_sessions ,dm_exec_requests

<span>1 <span>USE<span> master
<span>2 <span>GO
<span>3 <span>--<span>如果要指定数据库就把注释去掉
<span>4 <span>SELECT <span>* <span>FROM sys.<span>[<span>sysprocesses<span>] <span>WHERE <span>[<span>spid<span>]<span>><span>50 <span>--<span>AND DB_NAME([dbid])='gposdb'
<span>5 <span>SELECT <span>COUNT(<span>*) <span>FROM <span>[<span>sys<span>].<span>[<span>dm_exec_sessions<span>] <span>WHERE <span>[<span>session_id<span>]<span>><span>50</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

看一下当前的数据库用户连接有多少

然后使用下面语句看一下各项指标是否正常，是否有阻塞，这个语句选取了前10个最耗CPU时间的会话

<span> 1 <span>SELECT <span>TOP <span>10
<span> 2 <span>[<span>session_id<span>]<span>,
<span> 3 <span>[<span>request_id<span>]<span>,
<span> 4 <span>[<span>start_time<span>] <span>AS <span>'<span>开始时间<span>'<span>,
<span> 5 <span>[<span>status<span>] <span>AS <span>'<span>状态<span>'<span>,
<span> 6 <span>[<span>command<span>] <span>AS <span>'<span>命令<span>'<span>,
<span> 7 dest.<span>[<span>text<span>] <span>AS <span>'<span>sql语句<span>'<span>, 
<span> 8 <span>DB_NAME(<span>[<span>database_id<span>]) <span>AS <span>'<span>数据库名<span>'<span>,
<span> 9 <span>[<span>blocking_session_id<span>] <span>AS <span>'<span>正在阻塞其他会话的会话ID<span>'<span>,
<span>10 <span>[<span>wait_type<span>] <span>AS <span>'<span>等待资源类型<span>'<span>,
<span>11 <span>[<span>wait_time<span>] <span>AS <span>'<span>等待时间<span>'<span>,
<span>12 <span>[<span>wait_resource<span>] <span>AS <span>'<span>等待的资源<span>'<span>,
<span>13 <span>[<span>reads<span>] <span>AS <span>'<span>物理读次数<span>'<span>,
<span>14 <span>[<span>writes<span>] <span>AS <span>'<span>写次数<span>'<span>,
<span>15 <span>[<span>logical_reads<span>] <span>AS <span>'<span>逻辑读次数<span>'<span>,
<span>16 <span>[<span>row_count<span>] <span>AS <span>'<span>返回结果行数<span>'
<span>17 <span>FROM sys.<span>[<span>dm_exec_requests<span>] <span>AS<span> der 
<span>18 <span>CROSS<span> APPLY 
<span>19 sys.<span>[<span>dm_exec_sql_text<span>](der.<span>[<span>sql_handle<span>]) <span>AS<span> dest 
<span>20 <span>WHERE <span>[<span>session_id<span>]<span>><span>50 <span>AND <span>DB_NAME(der.<span>[<span>database_id<span>])<span>=<span>'<span>gposdb<span>'  
<span>21 <span>ORDER <span>BY <span>[<span>cpu_time<span>] <span>DESC</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

如果想看具体的SQL语句可以执行下面的SQL语句，记得在SSMS里选择以文本格式显示结果

<span>1 <span>--<span>在SSMS里选择以文本格式显示结果
<span>2 <span>SELECT <span>TOP <span>10 
<span>3 dest.<span>[<span>text<span>] <span>AS <span>'<span>sql语句<span>'
<span>4 <span>FROM sys.<span>[<span>dm_exec_requests<span>] <span>AS<span> der 
<span>5 <span>CROSS<span> APPLY 
<span>6 sys.<span>[<span>dm_exec_sql_text<span>](der.<span>[<span>sql_handle<span>]) <span>AS<span> dest 
<span>7 <span>WHERE <span>[<span>session_id<span>]<span>><span>50  
<span>8 <span>ORDER <span>BY <span>[<span>cpu_time<span>] <span>DESC</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

模拟了一些耗CPU时间的动作

-----------------------------------------华丽的分割线-----------------------------------------------------------

还有查看CPU数和user scheduler数和最大工作线程数，检查worker是否用完也可以排查CPU占用情况

<span>1 <span>--<span>查看CPU数和user scheduler数目
<span>2 <span>SELECT cpu_count,scheduler_count <span>FROM<span> sys.dm_os_sys_info
<span>3 <span>--<span>查看最大工作线程数
<span>4 <span>SELECT max_workers_count <span>FROM sys.dm_os_sys_info</span></span></span></span></span></span></span></span></span></span></span></span></span>

查看机器上的所有schedulers包括user 和system
通过下面语句可以看到worker是否用完，当达到最大线程数的时候就要检查blocking了

对照下面这个表
各种CPU和SQLSERVER版本组合自动配置的最大工作线程数
CPU数                 32位计算机                        64位计算机
  8                        288                                   576
 16                       352                                   704
 32                       480                                   960

<span>1 <span>SELECT
<span>2 <span>scheduler_address,
<span>3 <span>scheduler_id,
<span>4 <span>cpu_id,
<span>5 <span>status,
<span>6 <span>current_tasks_count,
<span>7 <span>current_workers_count,active_workers_count
<span>8 <span>FROM sys.dm_os_schedulers</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

 如果大家有什么需要补充的，或者文章有不正确的，欢迎大家拍砖！！

----------------------------------------------------------------------------------------------

 2013-6-15 做了一下补充，如果SQLSERVER存在要等待的资源，那么执行下面语句就会显示出会话中有多少个worker在等待

结合[sys].[dm_os_wait_stats]视图，如果当前SQLSERVER里面没有任何等待资源，那么下面的SQL语句不会显示任何结果

<span> 1 <span>SELECT <span>TOP <span>10
<span> 2  <span>[<span>session_id<span>]<span>,
<span> 3  <span>[<span>request_id<span>]<span>,
<span> 4  <span>[<span>start_time<span>] <span>AS <span>'<span>开始时间<span>'<span>,
<span> 5  <span>[<span>status<span>] <span>AS <span>'<span>状态<span>'<span>,
<span> 6  <span>[<span>command<span>] <span>AS <span>'<span>命令<span>'<span>,
<span> 7  dest.<span>[<span>text<span>] <span>AS <span>'<span>sql语句<span>'<span>, 
<span> 8  <span>DB_NAME(<span>[<span>database_id<span>]) <span>AS <span>'<span>数据库名<span>'<span>,
<span> 9  <span>[<span>blocking_session_id<span>] <span>AS <span>'<span>正在阻塞其他会话的会话ID<span>'<span>,
<span>10  der.<span>[<span>wait_type<span>] <span>AS <span>'<span>等待资源类型<span>'<span>,
<span>11  <span>[<span>wait_time<span>] <span>AS <span>'<span>等待时间<span>'<span>,
<span>12  <span>[<span>wait_resource<span>] <span>AS <span>'<span>等待的资源<span>'<span>,
<span>13  <span>[<span>dows<span>].<span>[<span>waiting_tasks_count<span>] <span>AS <span>'<span>当前正在进行等待的任务数<span>'<span>,
<span>14  <span>[<span>reads<span>] <span>AS <span>'<span>物理读次数<span>'<span>,
<span>15  <span>[<span>writes<span>] <span>AS <span>'<span>写次数<span>'<span>,
<span>16  <span>[<span>logical_reads<span>] <span>AS <span>'<span>逻辑读次数<span>'<span>,
<span>17  <span>[<span>row_count<span>] <span>AS <span>'<span>返回结果行数<span>'
<span>18  <span>FROM sys.<span>[<span>dm_exec_requests<span>] <span>AS<span> der 
<span>19  <span>INNER <span>JOIN <span>[<span>sys<span>].<span>[<span>dm_os_wait_stats<span>] <span>AS<span> dows 
<span>20  <span>ON der.<span>[<span>wait_type<span>]<span>=<span>[<span>dows<span>].<span>[<span>wait_type<span>]
<span>21  <span>CROSS<span> APPLY 
<span>22  sys.<span>[<span>dm_exec_sql_text<span>](der.<span>[<span>sql_handle<span>]) <span>AS<span> dest 
<span>23  <span>WHERE <span>[<span>session_id<span>]<span>><span>50  
<span>24  <span>ORDER <span>BY <span>[<span>cpu_time<span>] DESC</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

 比如我当前执行了查询SalesOrderDetail_test表100次，由于表数据非常多，所以SSMS需要把SQLSERVER执行的结果慢慢的取走，

造成了ASYNC_NETWORK_IO等待

<span>1 <span>USE <span>[<span>AdventureWorks<span>]
<span>2 <span>GO
<span>3 <span>SELECT <span>* <span>FROM dbo.<span>[<span>SalesOrderDetail_test<span>]
<span>4 <span>GO <span>100</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

------------------------------------------------------------------------------------------------

经过排查和这几天的观察情况，确定是某些表缺失索引导致，现在在这些表上增加了索引，问题解决了

<span>1 <span>select <span>* <span>from t_AccessControl        <span>--<span>权限控制表权限控制
<span>2 <span>select <span>* <span>from t_GroupAccess            <span>--<span>用户组权限表用户组权限
<span>3 <span>select <span>* <span>from t_GroupAccessType        <span>--<span>用户组权限类表用户组权限类
<span>4 <span>select <span>* <span>from t_ObjectAccess        <span>--<span>对象权限表对象权限
<span>5 <span>select <span>* <span>from t_ObjectAccessType    <span>--<span>对象权限类型表对象权限类型
<span>6 <span>select <span>* <span>from t_ObjectType            <span>--<span>对象类型表对象类型</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

查询CPU占用高的语句

<span> 1 <span>SELECT <span>TOP <span>10
<span> 2    total_worker_time<span>/execution_count <span>AS<span> avg_cpu_cost, plan_handle,
<span> 3 <span>   execution_count,
<span> 4    (<span>SELECT <span>SUBSTRING(<span>text, statement_start_offset<span>/<span>2 <span>+ <span>1<span>,
<span> 5       (<span>CASE <span>WHEN statement_end_offset <span>= <span>-<span>1
<span> 6          <span>THEN <span>LEN(<span>CONVERT(<span>nvarchar(<span>max), <span>text)) <span>* <span>2
<span> 7          <span>ELSE<span> statement_end_offset
<span> 8       <span>END <span>- statement_start_offset)<span>/<span>2<span>)
<span> 9    <span>FROM sys.dm_exec_sql_text(sql_handle)) <span>AS<span> query_text
<span>10 <span>FROM<span> sys.dm_exec_query_stats
<span>11 <span>ORDER <span>BY <span>[<span>avg_cpu_cost<span>] <span>DESC</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

查询缺失索引

<span>1 <span>SELECT 
<span>2     DatabaseName <span>= <span>DB_NAME<span>(database_id)
<span>3     ,<span>[<span>Number Indexes Missing<span>] <span>= <span>count(<span>*<span>) 
<span>4 <span>FROM<span> sys.dm_db_missing_index_details
<span>5 <span>GROUP <span>BY <span>DB_NAME<span>(database_id)
<span>6 <span>ORDER <span>BY <span>2 <span>DESC;</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

<span> 1 <span>SELECT  <span>TOP <span>10 
<span> 2         <span>[<span>Total Cost<span>]  <span>= <span>ROUND(avg_total_user_cost <span>* avg_user_impact <span>* (user_seeks <span>+ user_scans),<span>0<span>) 
<span> 3 <span>        , avg_user_impact
<span> 4         , TableName <span>=<span> statement
<span> 5         , <span>[<span>EqualityUsage<span>] <span>=<span> equality_columns 
<span> 6         , <span>[<span>InequalityUsage<span>] <span>=<span> inequality_columns
<span> 7         , <span>[<span>Include Cloumns<span>] <span>=<span> included_columns
<span> 8 <span>FROM<span>        sys.dm_db_missing_index_groups g 
<span> 9 <span>INNER <span>JOIN<span>    sys.dm_db_missing_index_group_stats s 
<span>10        <span>ON s.group_handle <span>=<span> g.index_group_handle 
<span>11 <span>INNER <span>JOIN<span>    sys.dm_db_missing_index_details d 
<span>12        <span>ON d.index_handle <span>=<span> g.index_handle
<span>13 <span>ORDER <span>BY <span>[<span>Total Cost<span>] <span>DESC;</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

定位问题后，新建非聚集索引

<span>1 <span>CREATE <span>NONCLUSTERED <span>INDEX IX_t_AccessControl_F4 <span>ON<span> dbo.t_AccessControl
<span>2 <span>(
<span>3 <span>    FObjectType
<span>4 )include(<span>[<span>FUserID<span>], <span>[<span>FAccessType<span>], <span>[<span>FAccessMask<span>]) <span>WITH( STATISTICS_NORECOMPUTE <span>= <span>OFF, IGNORE_DUP_KEY <span>= <span>OFF, ALLOW_ROW_LOCKS <span>= <span>ON, ALLOW_PAGE_LOCKS <span>= <span>ON) <span>ON <span>[<span>PRIMARY<span>]
<span>5 <span>GO
<span>6 
<span>7 <span>drop <span>index IX_t_AccessControl_F4 <span>on t_AccessControl</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>

 CPU占用恢复正常

跟踪模板和跟踪文件下载，请使用SQL2008R2 版本：files.cnblogs.com/lyhabc/跟踪模板和trace.rar