sql server cpu占用过高优化

SQLSERVER排查CPU占用高的情况

今天中午，有朋友叫我帮他看一下数据库，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

现象

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

内存占用不太高，只占用了30个G

CPU占用100%

---

排查方向

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

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

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

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

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

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

模拟了一些耗CPU时间的动作

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

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

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
<=4                     256                                   512
  8                        288                                   576
 16                       352                                   704
 32                       480                                   960

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

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

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

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

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

---

问题源头

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

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

查询CPU占用高的语句

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

查询缺失索引

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

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

定位问题后，新建非聚集索引

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

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

---

总结

从多次历史经验来看，如果