原文:如何有效抓取SQL Server的BLOCKING信息
转自:微软亚太区数据库技术支持组 官方博客
http://blogs.msdn.com/b/apgcdsd/archive/2011/12/12/sql-server-blocking.aspx
SQL Server允许并发操作,BLOCKING是指在某一操作没有完成之前,其他操作必须等待,以便于保证数据的完整性。BLOCKING的解决方法要查看BLOCKING的头是什么,为什么BLOCKING头上的语句执行的很慢。通常来讲只要我们能找到BLOCKING头上的语句,我们总能够想出各种各种的办法,来提升性能,缓解或解决BLOCKING的问题。
但是问题的关键是,我们不知道BLOCKING什么时候会发生。用户跟我们抱怨数据库性能很差,等我们连上数据库去查看的时候,那时候有可能BLOCKING可能就已经过去了。性能又变好了。或者由于问题的紧急性,我们直接重新启动服务器以恢复运营。但是问题并没有最终解决,我们不知道下次问题会在什么时候发生。
BLOCKING问题的后果比较严重。因为终端用户能直接体验到。他们提交一个订单的时候,无论如何提交不上去,通常几秒之内能完成的一个订单提交,甚至要等待十几分钟,才能提交完成。更有甚者,极严重的BLOCKING能导致SQL Server停止工作。如下面的SQL ERRORLOG所表示, 在短短的几分钟之内,SPID数据从158增长到694, 并马上导致SQL Server打了一个dump, 停止工作。我们很容易推断出问题的原因是由于BLOCKING导致的,但是我们无法得知BLOCKING HEADER是什么,我们必须要等下次问题重现时,辅之以工具,才能得知BLOCKING
HEADER在做什么事情。如果信息抓取时机不对,我们可能要等问题发生好几次,才能抓到。这时候,客户和经理就会有抱怨了。因为我们的系统是生产系统,问题每发生一次,都会对客户带来损失。
2011-06-01 16:22:30.98 spid1931 Alert There are 158 Active database sessions which is too high.
2011-06-01 16:23:31.16 spid3248 Alert There are 342 Active database sessions which is too high.
2011-06-01 16:24:31.06 spid3884 Alert There are 517 Active database sessions which is too high.
2011-06-01 16:25:31.08 spid3688 Alert There are 694 Active database sessions which is too high.
2011-06-01 16:26:50.93 Server Using 'dbghelp.dll' version '4.0.5'
2011-06-01 16:26:50.97 Server **Dump thread - spid = 0, EC = 0x0000000000000000
2011-06-01 16:26:50.97 Server ***Stack Dump being sent to D:\MSSQL10.INSTANCE\MSSQL\LOG\SQLDump0004.txt
2011-06-01 16:26:50.97 Server * *******************************************************************************
2011-06-01 16:26:50.97 Server *
2011-06-01 16:26:50.97 Server * BEGIN STACK DUMP:
2011-06-01 16:26:50.97 Server * 06/01/11 16:26:50 spid 4124
2011-06-01 16:26:50.97 Server *
2011-06-01 16:26:50.97 Server * Deadlocked Schedulers
2011-06-01 16:26:50.97 Server *
2011-06-01 16:26:50.97 Server * *******************************************************************************
2011-06-01 16:26:50.97 Server * -------------------------------------------------------------------------------
2011-06-01 16:26:50.97 Server * Short Stack Dump
2011-06-01 16:26:51.01 Server Stack Signature for the dump is 0x0000000000000258
BLOCKING的信息抓取有很多种方法。这里罗列了几种。并且对每种分析它的优缺点。以便我们选择。在枚举方法之前,我们先简单演示一下BLOCKING.
我们首先创建一个测试表:
DROP TABLE [TESTTABLE]
GO
CREATE TABLE [dbo].[TESTTABLE](
[ID] [int] NULL,
[NAME] [nvarchar](50) NULL
)
GO
INSERT INTO TESTTABLE VALUES (1, 'aaaa')
GO
然后打开一个查询窗口,执行下面的语句, 该语句修改一行数据,并等待3分钟,然后在结束transaction
BEGIN TRANSACTION
UPDATE TESTTABLE SET [NAME] = 'bbbb' WHERE [ID] = 1
WAITFOR DELAY '00:03:00'
COMMIT TRANSACTION
这时候,如果打开另外一个查询窗口,执行下面的语句,下面的语句就会被BLOCK住。
UPDATE TESTTABLE SET [NAME] = 'cccc' WHERE [ID] = 1
方法一, 抓取SQL Profiler
======================
SQL Profiler里面包含大量的信息。其中有一个事件在Errors and Warnings->Blocked Process Report专门用来获得blocking的情况。但是因为信息量比较大,而且我们并不能很好的估算在什么时候会产生blocking,另外在生产环境使用Profiler, 对性能可能会有影响,所以SQL Profiler并不是最合适的工具。我们在这里并不对它赘述。
方法二, 执行查询
================
如果我们检查问题的时候,blocking还存在,那么,我们可以直接可以运行几个查询,得知BLOCKING HEADER的信息
SELECT * FROM sys.sysprocesses where spid>50
上述查询只是告诉我们,BLOCKING HEADER的头是SPID=53, 但是并没有告诉我们SPID=53在做什么事情。我们可以用下面的查询,得到SPID=53的信息
DBCC INPUTBUFFER(53)
我们可以把上述的两个查询合并起来,用下面的查询:
SELECT SPID=p.spid,
DBName = convert(CHAR(20),d.name),
ProgramName = program_name,
LoginName = convert(CHAR(20),l.name),
HostName = convert(CHAR(20),hostname),
Status = p.status,
BlockedBy = p.blocked,
LoginTime = login_time,
QUERY = CAST(TEXT AS VARCHAR(MAX))
FROM MASTER.dbo.sysprocesses p
INNER JOIN MASTER.dbo.sysdatabases d
ON p.dbid = d.dbid
INNER JOIN MASTER.dbo.syslogins l
ON p.sid = l.sid
CROSS APPLY sys.dm_exec_sql_text(sql_handle)
WHERE p.blocked = 0
AND EXISTS (SELECT 1
FROM MASTER..sysprocesses p1
WHERE p1.blocked = p.spid)
这样,一次执行,就能告诉我们BLOCKING header的SPID信息,以及该SPID在做的语句。我们可以进一步研究该语句,以理解为什么该语句执行很慢。
用这个方法有一个缺点,就是使用的时候,要求BLOCKING是存在的。如果BLOCKING已经消失了,那么我们的方法就不管用了。
方法三,长期执行一个BLOCKING SCRIPT
==================================
因为我们通常无法知道BLOCKING什么时候会产生,所以通常的办法是我们长期运行一个BLOCKING SCRIPT, 这样,等下次发生的时候,我们就会有足够的信息。长期运行BLOCKING SCRIPT对性能基本上是没有影响的。因为我们每隔10秒钟抓取一次信息。缺点是,如果问题一个月才发生一次,那么,我们的BLOCKING日志信息会很大。所以这种方法适用于几天之内能重现问题。
运行方法如下:
如果要停止运行,我们按ctrl+c就可以了。
BLOCKING的信息存在log.out这个文件中
我们可以打开log.out这个文件, 会发现SPID 54被 SPID 53给Block住了。
而随后,我们可以看到SPID=53在做什么事情:
下面是BLOCKING SCRIPT的脚本, 我们可以把它存为blocking.sql
use master
go
while 1 =1
begin
print 'Start time: ' + convert(varchar(26), getdate(), 121)
Print 'Running processes'
select spid, blocked, waittype, waittime, lastwaittype, waitresource, dbid,uid, cpu, physical_io, memusage, login_time, last_batch,
open_tran, status, hostname, program_name, cmd, net_library, loginame
from sysprocesses
--where (kpid <> 0 ) or (spid < 51)
-- Change it if you only want to see the working processes
print '*********lockinfor***********'
select convert (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, 16) as Resource,
substring (u.name, 1, 8) As Mode,
substring (x.name, 1, 5) As Status
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'
order by spid
print 'inputbuffer for running processes'
declare @spid varchar(6)
declare ibuffer cursor fast_forward for
select cast (spid as varchar(6)) as spid from sysprocesses where spid >50
open ibuffer
fetch next from ibuffer into @spid
while (@@fetch_status != -1)
begin
print ''
print 'DBCC INPUTBUFFER FOR SPID ' + @spid
exec ('dbcc inputbuffer (' + @spid + ')')
fetch next from ibuffer into @spid
end
deallocate ibuffer
waitfor delay '0:0:10'
End
这种方法的缺陷就是,log.out会比较巨大,会占用很大的空间,如果blocking一个月甚至更长时间才发生一次,那我们的这个方法就不太适宜。
方法四,我们用Agent Job来检查BLOCKING
=====================================
长期运行一个BLOCKING SCRIPT的缺点是我们每隔一段时间,去查询信息,但是大多数收集的信息是无用的。所以会导致日志文件巨大,对于一个生产系统来讲,磁盘空间满可不是个好事情,另外,有一些客户对于用命令行来长期运行TSQL脚本有所顾忌,所以我们做了一个改进。这次,我们只收集有用的信息。对于无用的信息我们不关注。这样能极大减少日志大小。
我们首先创建一个观察数据库,然后建立两张表格 Blocking_sysprocesses和Blocking_SQLText, 建立一个存储过程和一个Job, 该Job每隔一段时间去调用存储过程。只有发现有blocking的,我们才记录到表格Blocking_sysprocesses和Blocking_SQLText这两个表格中。如果跟blocking无关,我们就不对它进行记录。下面是TSQL语句:
CREATE DATABASE [MonitorBlocking]
GO
USE [MonitorBlocking]
GO
CREATE TABLE Blocking_sysprocesses(
[spid] smallint,
[kpid] smallint,
[blocked] smallint,
[waitType] binary(2),
[waitTime] bigInt,
[lastWaitType] nchar(32),
[waitResource] nchar(256),
[dbID] smallint,
[uid] smallint,
[cpu] int,
[physical_IO] int,
[memusage] int,
[login_Time] datetime,
[last_Batch] datetime,
[open_Tran] smallint,
[status] nchar(30),
[sid] binary(86),
[hostName] nchar(128),
[program_Name] nchar(128),
[hostProcess] nchar(10),
[cmd] nchar(16),
[nt_Domain] nchar(128),
[nt_UserName] nchar(128),
[net_Library] nchar(12),
[loginName] nchar(128),
[context_Info] binary(128),
[sqlHandle] binary(20),
[CapturedTimeStamp] datetime
)
GO
CREATE TABLE [dbo].[Blocking_SqlText](
[spid] [smallint],
[sql_text] [nvarchar](2000),
[Capture_Timestamp] [datetime]
)
GO
CREATE PROCEDURE [dbo].[checkBlocking]
AS
BEGIN
SET NOCOUNT ON;
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
declare @Duration int -- in milliseconds, 1000 = 1 sec
declare @now datetime
declare @Processes int
select @Duration = 100 -- in milliseconds, 1000 = 1 sec
select @Processes = 0
select @now = getdate()
CREATE TABLE #Blocks_rg(
[spid] smallint,
[kpid] smallint,
[blocked] smallint,
[waitType] binary(2),
[waitTime] bigInt,
[lastWaitType] nchar(32),
[waitResource] nchar(256),
[dbID] smallint,
[uid] smallint,
[cpu] int,
[physical_IO] int,
[memusage] int,
[login_Time] datetime,
[last_Batch] datetime,
[open_Tran] smallint,
[status] nchar(30),
[sid] binary(86),
[hostName] nchar(128),
[program_Name] nchar(128),
[hostProcess] nchar(10),
[cmd] nchar(16),
[nt_Domain] nchar(128),
[nt_UserName] nchar(128),
[net_Library] nchar(12),
[loginName] nchar(128),
[context_Info] binary(128),
[sqlHandle] binary(20),
[CapturedTimeStamp] datetime
)
INSERT INTO #Blocks_rg
SELECT
[spid],
[kpid],
[blocked],
[waitType],
[waitTime],
[lastWaitType],
[waitResource],
[dbID],
[uid],
[cpu],
[physical_IO],
[memusage],
[login_Time],
[last_Batch],
[open_Tran],
[status],
[sid],
[hostName],
[program_name],
[hostProcess],
[cmd],
[nt_Domain],
[nt_UserName],
[net_Library],
[loginame],
[context_Info],
[sql_Handle],
@now as [Capture_Timestamp]
FROM master..sysprocesses where blocked <> 0
AND waitTime > @Duration
SET @Processes = @@rowcount
INSERT into #Blocks_rg
SELECT
src.[spid],
src.[kpid],
src.[blocked],
src.[waitType],
src.[waitTime],
src.[lastWaitType],
src.[waitResource],
src.[dbID],
src.[uid],
src.[cpu],
src.[physical_IO],
src.[memusage],
src.[login_Time],
src.[last_Batch],
src.[open_Tran],
src.[status],
src.[sid],
src.[hostName],
src.[program_name],
src.[hostProcess],
src.[cmd],
src.[nt_Domain],
src.[nt_UserName],
src.[net_Library],
src.[loginame],
src.[context_Info],
src.[sql_Handle]
,@now as [Capture_Timestamp]
FROM master..sysprocesses src inner join #Blocks_rg
trgt
on trgt.blocked = src.[spid]
if @Processes > 0
BEGIN
INSERT [dbo].[Blocking_sysprocesses]
SELECT * from #Blocks_rg
DECLARE @SQL_Handle binary(20), @SPID smallInt;
DECLARE cur_handle CURSOR FOR SELECT sqlHandle, spid FROM #Blocks_rg;
OPEN cur_Handle
FETCH NEXT FROM cur_handle INTO @SQL_Handle, @SPID
WHILE (@@FETCH_STATUS = 0)
BEGIN
INSERT [dbo].[Blocking_SqlText]
SELECT @SPID, CONVERT(nvarchar(4000), [text]) ,@now as [Capture_Timestamp]from ::fn_get_sql(@SQL_Handle)
FETCH NEXT FROM cur_handle INTO @SQL_Handle, @SPID
END
CLOSE cur_Handle
DEALLOCATE cur_Handle
END
DROP table #Blocks_rg
END
GO
USE msdb;
GO
EXEC dbo.sp_add_job
@job_name = N'MonitorBlocking';
GO
EXEC sp_add_jobstep
@job_name = N'MonitorBlocking',
@step_name = N'execute blocking script',
@subsystem = N'TSQL',
@command = N'exec checkBlocking',
@database_name=N'MonitorBlocking';
GO
EXEC sp_add_jobSchedule
@name = N'ScheduleBlockingCheck',
@job_name = N'MonitorBlocking',
@freq_type = 4, -- daily
@freq_interval = 1,
@freq_subday_type = 4,
@freq_subday_interval = 1
EXEC sp_add_jobserver @job_name = N'MonitorBlocking', @server_name = N'(local)'
当Blocking发生一段时间后,我们可以查询下面的两个表格,以得知当时问题发生时的blocking信息:
use MonitorBlocking
GO
SELECT * from Blocking_sqlText
SELECT * FROM Blocking_sysprocesses