SQLServer 通过DMV实现低影响的自动监控和历史场景追溯
一. 背景: 我们管理和维护的数据库系统基本都是7*24小时运转的,运转时会出现什么情况谁都无法估计,往往一个平时运行很正常的系统,某天晚上突然就抽风了,而此刻并没有系统负责人在旁边,当值班同事的电话打到正在酣睡的DBA手机上时,DBA不得不朦朦胧胧的
一. 背景:
我们管理和维护的数据库系统基本都是7*24小时运转的,运转时会出现什么情况谁都无法估计,往往一个平时运行很正常的系统,某天晚上突然就抽风了,而此刻并没有系统负责人在旁边,当值班同事的电话打到正在酣睡的DBA手机上时,DBA不得不朦朦胧胧的,条件反射要上数据库看看;要在家里连到公司的内外一般都需要DBA通知运维人员(也可能是个在做梦的哥们)先给自己开个VPN(某些管控严格的公司还需要DBA打车去公司才能处理),然后通过一大堆的验证才通过慢如蜗牛(往往也和此刻的心态有关)的网络,远程连接到了公司内网出问题的数据库上,此刻DBA才开始真正开始排查起问题来;如果说等待DBA登录到问题数据库上,问题依然在出现,那还算好,可以很快找到问题并处理,但是往往这种抽风的现象是短暂的,可能持续几分钟后,系统又恢复了正常,等到DBA"跑山涉水,翻山越岭" 的好不容易登录到数据库上检查时,数据库里的进程、锁、日志一起都正常,系统运行的很Hai;于是第二天上班时老大们问起昨天晚上事故的原因时,DBA只能凭空猜测,可能是网络、数据库阻塞、抑或是程序方面的问题吧......,此刻不同部门的人都猜是其他部门管理的东西出了问题,于是大家都把手头上的数据、监控图拿出来,证明自己这里是没有问题的;如果监控网络和系统的图和APP的Log都没用出现啥问题(说实话监控也不一定准确的),那就基本要把问题推到数据库身上了(DBA往往成为炮灰)。
数据库有没有出问题,如果出了问题又是啥问题呢?windows Log和DB Log都没异常,如果我们没有把Profile持续的开启(基于性能的考虑,一般都不会持续开启),就很难说清楚在系统出问题时数据库究竟有没有出问题,出了问题又是因为什么原因引起的;如果说此类问题出现了一次,就消失了,那还算好,成为一个无头案,悲剧的是这类问题无规律,反复的出现,如果DBA不能找出问题的原因,也不能证明数据库当时是正常的话,估计在公司里面就没用立足之地了。
二 . 对策:
其实对这种灵异的系统抽风事件,有两种比较好的解决方案:
1. 开启Trace 跟踪,就是数据库的Profile功能 ,这个还可以结合windows的性能计数器一同使用,能够直观的了解到特殊时间点上运行了什么语句,资源消耗情况是怎样的;不过这个方法比较消耗系统的资源,对访问压力比较大的数据库需要慎重;
2. 收集数据库DMV当时的情况,使得DBA在故障时间过后,还能通过这些数据了解到事故发生时,数据库里面运行的语句,以及锁和资源的情况;这种方法只是访问系统性能视图,对数据库其他业务影响比较小,也是我接下来要介绍的方案。
三. 实施方案:
思路:根据数据库中某些动态指标,触发收集DMV的过程,必要时发送报警邮件
1. 创建存储系统性能视图的表:
--创建四个记录表
USE[master]
GO
/****** Object: Table [dbo].[dc_block_info] Script Date: 05/23/2011 11:35:48 ******/
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
SET ANSI_PADDING ON
GO
CREATETABLE[dbo].[dc_block_info](
[spid][smallint]NULL,
[status][nchar](30) NULL,
[SQLBuffer][nvarchar](max) NULL,
[hostname][nchar](128) NULL,
[BlkBy][varchar](10) NULL,
[BlockedSQLBuffer][nvarchar](max) NULL,
[LoginName][varchar](100) NULL,
[DBName][varchar](50) NULL,
[CPUTime][int]NULL,
[DiskIO][int]NULL,
[LastBatch][datetime]NULL,
[program_name][nchar](128) NULL,
[Command][varchar](100) NULL,
[batch_id][int]NULL
) ON[PRIMARY]
GO
SET ANSI_PADDING OFF
GO
USE[master]
GO
/****** Object: Table [dbo].[dc_Blocked_Resource_Info] Script Date: 05/23/2011 11:35:48 ******/
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
CREATETABLE[dbo].[dc_Blocked_Resource_Info](
[spid][smallint]NULL,
[dbid][smallint]NOTNULL,
[ObjId][int]NOTNULL,
[IndId][smallint]NOTNULL,
[Type][nvarchar](4) NULL,
[Resource][nvarchar](32) NULL,
[Mode][nvarchar](8) NULL,
[Status][nvarchar](5) NULL,
[batch_id][int]NULL
) ON[PRIMARY]
GO
USE[master]
GO
/****** Object: Table [dbo].[dc_info_BlockedInfo] Script Date: 05/23/2011 11:35:48 ******/
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
SET ANSI_PADDING ON
GO
CREATETABLE[dbo].[dc_info_BlockedInfo](
[batch_id][int]NULL,
[lock_type][varchar](100) NULL,
[database_name][varchar](20) NULL,
[blk_object][varchar](100) NULL,
[lock_req][varchar](100) NULL,
[waiter_sid][int]NULL,
[wait_time][int]NULL,
[waiter_batch][varchar](max) NULL,
[waiter_stmt][varchar](max) NULL,
[blocker_sid][int]NULL,
[blocker_stmt][varchar](max) NULL,
[create_date][datetime]NULL
) ON[PRIMARY]
GO
SET ANSI_PADDING OFF
GO
USE[master]
GO
/****** Object: Table [dbo].[dc_info_SessionConn] Script Date: 05/23/2011 11:35:48 ******/
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
SET ANSI_PADDING ON
GO
CREATETABLE[dbo].[dc_info_SessionConn](
[batch_id][int]NULL,
[session_id][int]NULL,
[blocking_session_id][int]NULL,
[textdata][varchar](max) NULL,
[login_name][varchar](30) NULL,
[host_name][varchar](100) NULL,
[database_name][varchar](30) NULL,
[program_name][varchar](200) NULL,
[command][varchar](100) NULL,
[status][varchar](20) NULL,
[cpu_time][int]NULL,
[memory_usage_kb][int]NULL,
[reads][int]NULL,
[writes][int]NULL,
[transaction_isolation_level][int]NULL,
[connect_time][datetime]NULL,
[last_read][datetime]NULL,
[last_write][datetime]NULL,
[net_transport][varchar](20) NULL,
[client_net_address][varchar](30) NULL,
[client_tcp_port][int]NULL,
[local_tcp_port][int]NULL,
[start_time][datetime]NULL,
[wait_time][int]NULL,
[wait_type][varchar](100) NULL,
[last_wait_type][varchar](100) NULL,
[wait_resource][varchar](1000) NULL,
[open_transaction_count][int]NULL,
[create_date][datetime]NULL
) ON[PRIMARY]
GO
SET ANSI_PADDING OFF
GO
2. 定义性能收集的存储过程:
USE[master]
GO
--数据库的阻塞和锁信息,该SP可以记录session中显示不出的信息
createprocedure[dbo].[usp_blocker_info](@batch_idint)
as
begin
set nocount on
declare@spidsmallint,@blockedsmallint
declare@c_SQLnvarchar(4000),@b_SQLnvarchar(4000)
declare@idint
--保存Sysprocesses 的内容
createtable #Temp(spid smallint,status nchar(30),hostname nchar(128),program_name nchar(128)
,cmd nchar(16),cpu int,physical_io int,blocked smallint,dbid smallint
,loginame nchar(128),last_batch datetime
,SQLBuffer nvarchar(4000),BlockedSQLBuffer nvarchar(4000))
--保存DBCC InputBuffer 的结果
createtable #Temp1(id intidentity(1,1),eventtype varchar(20),parameters int,eventinfo nvarchar(4000))
createtable #Temp_b(id intidentity(1,1),eventtype varchar(20),parameters int,eventinfo nvarchar(4000))
select*into #Temp2
from master..sysprocesses (nolock)
--保存被阻塞的进程信息
insertinto #Temp(spid ,status ,hostname ,program_name ,cmd ,cpu ,physical_io ,blocked ,dbid
,loginame ,last_batch )
SELECT spid ,status ,hostname ,program_name ,cmd ,cpu ,physical_io ,blocked ,dbid
,convert(sysname, rtrim(loginame)) ,last_batch
from #Temp2
where blocked >0
--保存阻塞的源头
insertinto #Temp(spid ,status ,hostname ,program_name ,cmd ,cpu ,physical_io ,blocked ,dbid
,loginame ,last_batch )
SELECT spid ,status ,hostname ,program_name ,cmd ,cpu ,physical_io ,blocked ,dbid
,convert(sysname, rtrim(loginame)) ,last_batch
from #Temp2
where spid in (select blocked from #Temp)
and spid notin (select spid from #Temp)
select@spid=min(spid) from #Temp
while@spidisnotnull
begin
set@c_SQL='dbcc inputbuffer('+convert(varchar(5), @spid) +')'
select@blocked=isnull(blocked,0) from #Tempwhere spid=@spid
if(@blocked0)
begin
set@b_SQL='dbcc inputbuffer('+convert(varchar(5), @blocked) +')'
insertinto #Temp_b
exec (@b_SQL)
select@id=@@identity
update #Temp
set BlockedSQLBuffer = #Temp_b.eventinfo
from #Temp,#Temp_b
where #Temp_b.id =@id
and #Temp.blocked =@blocked
end
insertinto #Temp1
exec (@c_SQL)
select@id=@@identity
update #Temp
set SQLBuffer = #Temp1.eventinfo
from #Temp,#Temp1
where #Temp1.id =@id
and #Temp.spid =@spid
select@spid=min(spid) from #Tempwhere spid >@spid
end
insertinto dc_block_info(batch_id,spid,status,SQLBuffer,hostname,BlkBy,BlockedSQLBuffer,LoginName,DBName,
CPUTime,DiskIO,LastBatch,program_name,Command)
SELECT@batch_id,convert(char(5),spid) SPID,CASElower(status) When'sleeping'Thenlower(status) Elseupper(status) END Status
,SQLBuffer
, CASE hostname WhenNullThen' .'When''Then' .'Else hostname END HostName
,CASEisnull(convert(char(5),blocked),'0') When'0'Then' .'
Elseisnull(convert(char(5),blocked),'0') END BlkBy,BlockedSQLBuffer
,loginame Login
,db_name(dbid) DBName,convert(varchar,cpu) CPUTime
,convert(varchar,physical_io) DiskIO,Last_Batch LastBatch
,program_name ProgramName, cmd Command
from #Temp
orderby BlkBy, spid
set nocount off
end
Go
USE[master]
GO
--预警SP
CREATEproc[dbo].[RecodeAndAlertInfo]
as
begin
declare@banch_idint
select@banch_id=isnull(MAX(batch_id),0)+1from dc_info_SessionConn
--记录当前所有会话信息
insertinto dc_info_SessionConn
SELECT@banch_id,S.session_id, R.blocking_session_id,
current_execute_sql =SUBSTRING(T.text,
R.statement_start_offset /2+1,
CASE
WHEN statement_end_offset =-1THENLEN(T.text)
ELSE (R.statement_end_offset - statement_start_offset) /2+1
END),
S.login_name,S.host_name,databaseName=DB_NAME(R.database_id),S.program_name,R.command,
S.status,S.cpu_time, memory_usage_kb = S.memory_usage *8, S.reads, S.writes,
S.transaction_isolation_level,C.connect_time, C.last_read, C.last_write,
C.net_transport, C.client_net_address, C.client_tcp_port, C.local_tcp_port,
R.start_time, R.wait_time, R.wait_type, R.last_wait_type, R.wait_resource,
R.open_transaction_count,GETDATE()
FROM sys.dm_exec_sessions S
LEFTJOIN sys.dm_exec_connections C
ON S.session_id = C.session_id
LEFTJOIN sys.dm_exec_requests R
ON S.session_id = R.session_id
AND C.connection_id = R.connection_id
OUTER APPLY sys.dm_exec_sql_text(R.sql_handle) T
WHERE S.is_user_process =1-- 如果不限制此条件,则查询所有进程(系统和用户进程)
AND command isnotnull
--记录当前阻塞信息
insertinto dc_info_BlockedInfo
select@banch_id,t1.resource_type as[lock type] ,db_name(resource_database_id) as[database]
,t1.resource_associated_entity_id as[blk object]
,t1.request_mode as[lock req]-- lock requested
,t1.request_session_id as[waiter sid]-- spid of waiter
,t2.wait_duration_ms as[wait time]
,(selecttextfrom sys.dm_exec_requests as r --- get sql for waiter
cross apply sys.dm_exec_sql_text(r.sql_handle)
where r.session_id = t1.request_session_id) as waiter_batch
,(selectsubstring(qt.text,r.statement_start_offset/2,
(casewhen r.statement_end_offset =-1thenlen(convert(nvarchar(max), qt.text)) *2
else r.statement_end_offset end- r.statement_start_offset)/2)
from sys.dm_exec_requests as r
cross apply sys.dm_exec_sql_text(r.sql_handle) as qt
where r.session_id = t1.request_session_id) as waiter_stmt --- statement executing now
,t2.blocking_session_id as[blocker sid]--- spid of blocker
,(selecttextfrom sys.sysprocesses as p --- get sql for blocker
cross apply sys.dm_exec_sql_text(p.sql_handle)
where p.spid = t2.blocking_session_id) as blocker_stmt,getdate()
from sys.dm_tran_locks as t1, sys.dm_os_waiting_tasks as t2
where t1.lock_owner_address = t2.resource_address
--记录资源信息
insertinto dc_Blocked_Resource_Info
selectconvert (smallint, req_spid) As spid,
rsc_dbid As dbid,
rsc_objid As ObjId,
rsc_indid As IndId,
substring (v.name, 1, 4) As Type,
substring (rsc_text, 1, 32) as Resource,
substring (u.name, 1, 8) As Mode,
substring (x.name, 1, 5) As Status ,@banch_id
from master.dbo.syslockinfo,
master.dbo.spt_values v,
master.dbo.spt_values x,
master.dbo.spt_values u
where master.dbo.syslockinfo.rsc_type = v.number
and v.type ='LR'
and master.dbo.syslockinfo.req_status = x.number
and x.type ='LS'
and master.dbo.syslockinfo.req_mode +1= u.number
and u.type ='L'
andsubstring (x.name, 1, 5) ='WAIT'
orderby spid
exec usp_blocker_info @banch_id
end
GO
3. 创建信息收集和邮件报警的SP:
说明:
SP定义了一个邮件发送的过程,需要先配置好数据库的邮件发送(google一下很多的);
SP需要做到JOB里面,一分钟运行一次收集信息;
SP收集信息时的条件(不同的系统触发条件不一样):
a. 用户链接数大于550,活动链接数大于40;
b. 阻塞进程比率大于10%;
USE[master]
GO
--创建触发监控条件,并发邮件
CREATEproc[dbo].[ConAlert]
as
begin
DECLARE@connfloat ,@activeconnfloat ,@blockedcountfloat,@spidcountfloat,@spidblockedfloat ,@countint
DECLARE@bodyNVARCHAR(MAX),@subjectnvarchar(200)
-- User Connections
Select@conn=ISNULL(cntr_value,0) from sys.dm_os_performance_counters with(nolock)
where counter_name='User Connections'
if@conn>=550--根据用户链接数来收集信息和触发报警
begin
--active requests
Select@activeconn=isnull(SUM(cntr_value),0) from sys.dm_os_performance_counters with(nolock)
where counter_name ='Active requests'
groupby counter_name
set@count=1
if@activeconn>=40--根据活动链接数来判断
begin
while1=1--循环收集信息
begin
exec RecodeAndAlertInfo
if(@count%5=0) --连续次就发邮件
begin
set@subject='Server:'+@@SERVERNAME+' Connections Alert'
SET@body='Server:'+@@SERVERNAME+CHAR(13)+'; UserConnections:'+cast(@connasvarchar)+CHAR(13)+'; ActiveRequests:'+cast(@activeconnasvarchar)
EXEC msdb.dbo.sp_send_dbmail
@recipients=N'Ken@xxx.com;itmon@xxx.com',
@body=@body,
@body_format='TEXT',
@subject=@subject,
@profile_name='dba_profile'--需要配置该模板
end
Select@activeconn=isnull(SUM(cntr_value),0) from sys.dm_os_performance_counters with(nolock)
where counter_name ='Active requests'
groupby counter_name
if@activeconn>=40
waitfor delay '00:00:05'
else
break;
if@count>50
break;
set@count=@count+1;
end
end
end
else--根据Blocked 百分比来收集信息和触发报警
begin
select@spidcount=count(0) from sys.sysprocesses with(nolock) where spid>50
select@blockedcount=count(0) from sys.sysprocesses with(nolock) where spid>50and blocked0
--计算百分比
set@spidblocked=@blockedcount/@spidcount
set@count=1
if(@spidblocked>=0.1) --10%时,记录信息
begin
while1=1--循环收集信息
begin
exec RecodeAndAlertInfo
if(@count%5=0or@spidblocked>=0.15) --连续5次阻塞比率大于10%或者阻塞比率大于等于15%时,记录信息并发生报警邮件
begin
set@subject='Server:'+@@SERVERNAME+' Blocked Alert'
SET@body='Server:'+@@SERVERNAME+CHAR(13)+'; BlockedCount:'+cast(@blockedcountasvarchar)+CHAR(13)+'; SPIDCount:'+cast(@spidcountasvarchar)
EXEC msdb.dbo.sp_send_dbmail
@recipients=N'Ken@xxx.com',
@body=@body,
@body_format='TEXT',
@subject=@subject,
@profile_name='dba_profile'
end
select@spidcount=count(0) from sys.sysprocesses with(nolock) where spid>50
select@blockedcount=count(0) from sys.sysprocesses with(nolock) where spid>50and blocked0
set@spidblocked=@blockedcount/@spidcount
if@spidblocked>=0.1
waitfor delay '00:00:05'
else
break;
if@count>50
break;
set@count=@count+1;
end
end
end
end
4. 查看信息:
USE[master]
GO
--通过以下查询来分析信息
declare@batch_idint
select@batch_id=isnull(MAX(batch_id),1) from dc_info_SessionConn
select*from dc_info_SessionConn where batch_id=@batch_id
select*from dc_info_BlockedInfo where batch_id=@batch_id
select*from dc_block_info where batch_id=@batch_id
select*from dc_Blocked_Resource_Info where batch_id=@batch_id
有了这些阻塞、进程和资源的信息,我们就可以了解系统出问题时,数据库是否出现异常,以及数据库异常时都有些什么语句在系统中运行,它们是否造成了大量的阻塞或消耗了大量的资源;这样DBA就方便定位问题了。
上一篇: Linux下安装Informix数据库
下一篇: PHP函数getenv使用实例介绍