死锁原因的排查

 

 

1，打开跟踪，写入日志
DBCC TRACEON (3605,1222,-1) –3605写入errorlog,1222死锁
DBCC TRACEON(1222,-1) /若在启动时，加-T 1222
同时，可开profile中的Deadlock graph跟踪，以图形化
2，分析安装目录下生成的日志
   1）确定死锁的资源, 据对象不同查看方式不同。
DBCC TRACEON(3604)

DBCC PAGE(8,1,96,1)
DBCC TRACEOFF(3604)

死锁产生的前提：双方互占有了对方所需求的资源，若资源并不必要，可过滤掉。
资源不必要：1）扫描了对方的资源，扫描过的就会加锁，避免被扫描到，如加索引等。
3,死锁处理方法：
 1）从性能出发，优化sql
 2）从业务逻辑出发，看是否可去掉对死锁资料的关联。
 3）若还是不能解决，死锁是正常情况，避免不了，但可避免输出1205错误信息给客户端，方法就是加上try catch,可以等一会儿再重新执行。
如：

/*避免死锁显示给客户端 */
DECLARE @retries INT ;
SET @retries = 4 ;
WHILE ( @retries > 0 )
    BEGIN
        BEGIN TRY
            BEGIN TRANSACTION ;
            -- place sql code here
            SET @retries = 0 ;
            COMMIT TRANSACTION ;
        END TRY
        BEGIN CATCH
            -- Error is a deadlock
            IF ( ERROR_NUMBER() = 1205 )
                SET @retries = @retries - 1 ;
                -- Error is not a deadlock
            ELSE
                BEGIN
                    DECLARE @ErrorMessage NVARCHAR(4000) ;
                    DECLARE @ErrorSeverity INT ;
                    DECLARE @ErrorState INT ;
                    SELECT  @ErrorMessage = ERROR_MESSAGE() ,
                            @ErrorSeverity = ERROR_SEVERITY() ,
                            @ErrorState = ERROR_STATE() ;
                    -- Re-Raise the Error that caused the problem
                    RAISERROR (@ErrorMessage, -- Message text.
                    @ErrorSeverity, -- Severity.
                    @ErrorState -- State.
                    ) ;
                    SET @retries = 0 ;
                END
            IF XACT_STATE() <> 0
                ROLLBACK TRANSACTION ;
        END CATCH ;
    END ;
GO

/*死锁模拟
1,建立数据
2，连续两个事务
*/
drop table Employee_Demo_Heap
go
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
CREATE TABLE [Employee_Demo_Heap](
	[EmployeeID] [int] NOT NULL,
	[NationalIDNumber] [nvarchar](15) NOT NULL,
	[ContactID] [int] NOT NULL,
	[LoginID] [nvarchar](256) NOT NULL,
	[ManagerID] [int] NULL,
	[Title] [nvarchar](50) NOT NULL,
	[BirthDate] [datetime] NOT NULL,
	[MaritalStatus] [nchar](1) NOT NULL,
	[Gender] [nchar](1) NOT NULL,
	[HireDate] [datetime] NOT NULL,
	[ModifiedDate] [datetime] NOT NULL   DEFAULT (getdate()),
 CONSTRAINT [PK_Employee_EmployeeID_Demo_Heap] PRIMARY KEY nonCLUSTERED
(
	[EmployeeID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF,
IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]
GO

CREATE NONCLUSTERED INDEX [IX_Employee_ManagerID_Demo_Heap] ON [Employee_Demo_Heap]
(
	[ManagerID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO
CREATE NONCLUSTERED INDEX [IX_Employee_ModifiedDate_Demo_Heap] ON [Employee_Demo_Heap]
(
	[ModifiedDate] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO
insert into Employee_Demo_Heap
select [EmployeeID] ,
	[NationalIDNumber] ,
	[ContactID] ,
	[LoginID] ,
	[ManagerID],
	[Title] ,
	[BirthDate] ,
	[MaritalStatus] ,
	[Gender] ,
	[HireDate] ,
	[ModifiedDate] from HumanResources.Employee
go

现在就用下面这组脚本模拟出一个死锁来。在一个连接里，运行下面的语句。反复开启事务。在这个事务里，先修改一条NationalIDNumber=‘480951955’的记录，然后再把它查询出来。做完以后，提交事务。
set nocount on
go
while 1=1
begin
begin tran
update dbo.Employee_Demo_Heap
set BirthDate = getdate()
where NationalIDNumber = '480951955'
select * from dbo.Employee_Demo_Heap
where NationalIDNumber = '480951955'
commit tran
end
	在另外一个连接里，也运行这些语句。唯一的差别是这次修改和查询的是另一条NationalIDNumber = ‘407505660’的记录。
set nocount on
go
while 1=1
begin
begin tran
update dbo.Employee_Demo_Heap
set BirthDate = getdate()
where NationalIDNumber = '407505660'
select * from dbo.Employee_Demo_Heap
where NationalIDNumber = '407505660'
commit tran
end