SQLSERVER排查CPU佔用高的情況

SQLSERVER排查CPU佔用高的情況

今天中午，有朋友叫我幫他看一下數據庫，操作系統是Windows2008R2 ,數據庫是SQL2008R2 64位

64G內存，16核CPU

硬件配置還是比較高的，他說服務器運行的是金蝶K3軟件，數據庫實例裏有多個數據庫

他說是這幾天纔出現的，而且在每天的某一個時間段纔會出現CPU佔用高的情況

內存佔用也很高，佔用了30個G

-----------------------------------------------華麗的分割線-------------------------------------------------------

一般排查都是用下面的腳本，一般會用到三個視圖sys.sysprocesses ,dm_exec_sessions ,dm_exec_requests

1 USE master
2 GO
3 --如果要指定數據庫就把註釋去掉
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

看一下當前的數據庫用戶連接有多少

然後使用下面語句看一下各項指標是否正常，是否有阻塞，這個語句選取了前10個最耗CPU時間的會話

 1 SELECT TOP 10
 2 [session_id],
 3 [request_id],
 4 [start_time] AS '開始時間',
 5 [status] AS '狀態',
 6 [command] AS '命令',
 7 dest.[text] AS 'sql語句', 
 8 DB_NAME([database_id]) AS '數據庫名',
 9 [blocking_session_id] AS '正在阻塞其他會話的會話ID',
10 [wait_type] AS '等待資源類型',
11 [wait_time] AS '等待時間',
12 [wait_resource] AS '等待的資源',
13 [reads] AS '物理讀次數',
14 [writes] AS '寫次數',
15 [logical_reads] AS '邏輯讀次數',
16 [row_count] AS '返回結果行數'
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

如果想看具體的SQL語句可以執行下面的SQL語句，記得在SSMS裏選擇以文本格式顯示結果

1 --在SSMS裏選擇以文本格式顯示結果
2 SELECT TOP 10 
3 dest.[text] AS 'sql語句'
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

模擬了一些耗CPU時間的動作

-----------------------------------------華麗的分割線-----------------------------------------------------------

還有查看CPU數和user scheduler數和最大工作線程數，檢查worker是否用完也可以排查CPU佔用情況

1 --查看CPU數和user scheduler數目
2 SELECT cpu_count,scheduler_count FROM sys.dm_os_sys_info
3 --查看最大工作線程數
4 SELECT max_workers_count FROM sys.dm_os_sys_info

查看機器上的所有schedulers包括user 和system
通過下面語句可以看到worker是否用完，當達到最大線程數的時候就要檢查blocking了

對照下面這個表
各種CPU和SQLSERVER版本組合自動配置的最大工作線程數
CPU數                 32位計算機                        64位計算機
<=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

 如果大家有什麼需要補充的，或者文章有不正確的，歡迎大家拍磚！！

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

 2013-6-15 做了一下補充，如果SQLSERVER存在要等待的資源，那麼執行下面語句就會顯示出會話中有多少個worker在等待

結合[sys].[dm_os_wait_stats]視圖，如果當前SQLSERVER裏面沒有任何等待資源，那麼下面的SQL語句不會顯示任何結果

 1 SELECT TOP 10
 2  [session_id],
 3  [request_id],
 4  [start_time] AS '開始時間',
 5  [status] AS '狀態',
 6  [command] AS '命令',
 7  dest.[text] AS 'sql語句', 
 8  DB_NAME([database_id]) AS '數據庫名',
 9  [blocking_session_id] AS '正在阻塞其他會話的會話ID',
10  der.[wait_type] AS '等待資源類型',
11  [wait_time] AS '等待時間',
12  [wait_resource] AS '等待的資源',
13  [dows].[waiting_tasks_count] AS '當前正在進行等待的任務數',
14  [reads] AS '物理讀次數',
15  [writes] AS '寫次數',
16  [logical_reads] AS '邏輯讀次數',
17  [row_count] AS '返回結果行數'
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

 比如我當前執行了查詢SalesOrderDetail_test表100次，由於表數據非常多，所以SSMS需要把SQLSERVER執行的結果慢慢的取走，

造成了ASYNC_NETWORK_IO等待

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

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

經過排查和這幾天的觀察情況，確定是某些表缺失索引導致，現在在這些表上增加了索引，問題解決了

1 select * from t_AccessControl        --權限控制表權限控制
2 select * from t_GroupAccess            --用戶組權限表用戶組權限
3 select * from t_GroupAccessType        --用戶組權限類表用戶組權限類
4 select * from t_ObjectAccess        --對象權限表對象權限
5 select * from t_ObjectAccessType    --對象權限類型表對象權限類型
6 select * from t_ObjectType            --對象類型表對象類型

查詢CPU佔用高的語句

 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

查詢缺失索引

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;

定位問題後，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佔用恢復正常

跟蹤模板和跟蹤文件下載，請使用SQL2008R2 版本：files.cnblogs.com/lyhabc/跟蹤模板和trace.rar