INSERT 和 SELECT SP 之间的死锁
Posted
技术标签:
【中文标题】INSERT 和 SELECT SP 之间的死锁【英文标题】:Deadlock between INSERT and SELECT SPs 【发布时间】:2014-03-19 13:27:57 【问题描述】:我遇到了一个应用程序问题,我遇到了我认为永远不会发生的死锁。我有两个存储过程。其中一个只是读取表的一部分,其中一个在表中添加一行,或者如果存在主键冲突,则更新现有行。
我已经阅读了有关存储过程可以获得共享读锁的信息,如果两个存储过程都升级为排他锁,则会出现死锁。我在这里看不到这种情况,因为 GetList 存储过程只是读取而从不写入。这些存储过程死锁还有其他可能的原因吗?
以下关于死锁的报告:
<?xml version="1.0" encoding="UTF-8"?>
<deadlock-list>
<deadlock victim="process3e37f62c8">
<process-list>
<process id="process3e37f62c8" taskpriority="0" logused="0" waitresource="KEY: 7:72057594152681472 (de2613e7782e)" waittime="2535" ownerId="7732055781" transactionname="SELECT" lasttranstarted="2014-03-17T17:48:31.437" XDES="0x1298bc1c0" lockMode="S" schedulerid="11" kpid="21372" status="suspended" spid="121" sbid="0" ecid="0" priority="0" trancount="0" lastbatchstarted="2014-03-17T17:48:31.437" lastbatchcompleted="2014-03-17T17:48:31.437" clientapp=".Net SqlClient Data Provider" hostname="STO2AP07" hostpid="4908" loginname="(hidden)" isolationlevel="read committed (2)" xactid="7732055781" currentdb="7" lockTimeout="4294967295" clientoption1="673185824" clientoption2="128056">
<executionStack>
<frame procname="(hidden).dbo.GetList" line="18" stmtstart="1060" stmtend="1942" sqlhandle="0x03000700072cfd1b7b8be800cfa200000100000000000000">
SELECT
TOP (@NumRows)
[T1].[Id] ,
[T1].[A] ,
[T1].[B] ,
[T1].[C] ,
[T1].[D] ,
[T1].[E] ,
[T1].[F] ,
[T1].[G] ,
[T1].[H] ,
[T1].[I] ,
[T1].[J]
FROM [Item] AS [T1] where [J] > @LastUpdatedDateTime ORDER BY [T1].[LastUpdatedField] ASC
</frame>
</executionStack>
<inputbuf>Proc [Database Id = 7 Object Id = 469576711]</inputbuf>
</process>
<process id="process5a514c8" taskpriority="0" logused="244" waitresource="KEY: 7:72057594152615936 (16f70bd264f5)" waittime="2535" ownerId="7732055725" transactionname="user_transaction" lasttranstarted="2014-03-17T17:48:31.437" XDES="0x6602dfa00" lockMode="X" schedulerid="13" kpid="21196" status="suspended" spid="267" sbid="0" ecid="0" priority="0" trancount="2" lastbatchstarted="2014-03-17T17:48:31.437" lastbatchcompleted="2014-03-17T17:48:31.437" clientapp=".Net SqlClient Data Provider" hostname="STO2AP07" hostpid="4908" loginname="(hidden)" isolationlevel="read committed (2)" xactid="7732055725" currentdb="7" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
<executionStack>
<frame procname="(hidden).dbo.AddToList" line="52" stmtstart="2484" stmtend="3294" sqlhandle="0x030007005cbf2019598be800cfa200000100000000000000">
UPDATE [Item]
SET
[A] = @A ,
[B] = @B ,
[C] = @C ,
[D] = @D ,
[E] = @E ,
[F] = @F ,
[G] = @G ,
[H] = @H ,
[I] = @I ,
[J] = @J
WHERE [Id] = @Id
</frame>
</executionStack>
<inputbuf>Proc [Database Id = 7 Object Id = 421576540]</inputbuf>
</process>
</process-list>
<resource-list>
<keylock hobtid="72057594152681472" dbid="7" objectname="(hidden).dbo.Item" indexname="PK_Item_1" id="lock47c381b80" mode="X" associatedObjectId="72057594152681472">
<owner-list>
<owner id="process5a514c8" mode="X" />
</owner-list>
<waiter-list>
<waiter id="process3e37f62c8" mode="S" requestType="wait" />
</waiter-list>
</keylock>
<keylock hobtid="72057594152615936" dbid="7" objectname="(hidden).dbo.Item" indexname="IX_LastUpdatedField" id="lock402a95800" mode="S" associatedObjectId="72057594152615936">
<owner-list>
<owner id="process3e37f62c8" mode="S" />
</owner-list>
<waiter-list>
<waiter id="process5a514c8" mode="X" requestType="wait" />
</waiter-list>
</keylock>
</resource-list>
</deadlock>
</deadlock-list>
以下是名为GetList的存储过程:
ALTER PROCEDURE [dbo].[GetList]
@LastUpdatedDateTime datetime,
@NumRows int
WITH RECOMPILE
AS
BEGIN
SET NOCOUNT ON;
SELECT TOP (@NumRows)
[T1].[Id]
, [T1].[A]
, [T1].[B]
, [T1].[C]
, [T1].[D]
, [T1].[E]
, [T1].[F]
, [T1].[G]
, [T1].[H]
, [T1].[I]
, [T1].[J]
FROM [Item] AS [T1] where J >= @LastUpdatedDateTime
ORDER BY [T1].[J] ASC
END
以下是名为 AddToList 的存储过程:
ALTER PROCEDURE [dbo].[AddToList]
@Id int,
@A varchar(30),
@B decimal(18,2),
@V varchar(30),
@D varchar(512),
@E datetime,
@F datetime,
@G bit,
@H int,
@I int
@J datetime,
AS
IF NOT EXISTS(SELECT Id FROM Item WHERE [Id] = @Id)
BEGIN
SET NOCOUNT ON;
INSERT INTO [Item]
( [Id]
, [A]
, [B]
, [C]
, [D]
, [E]
, [F]
, [G]
, [H]
, [I]
, [J]
)
VALUES
( @Id
, @A
, @B
, @C
, @D
, @E
, @F
, @G
, @H
, @I
, @J
)
END
ELSE
BEGIN
UPDATE [Item] SET
[A] = @A
, [B] = @B
, [C] = @C
, [D] = @D
, [E] = @E
, [F] = @F
, [G] = @G
, [H] = @H
, [I] = @I
, [J] = @J
WHERE [Id] = @Id
END
【问题讨论】:
【参考方案1】:选择这样做:
-
从 LastUpdatedDateTime 索引读取(S 锁)
查找对应的 CI 行(S 锁)
更新是这样的:
-
写入 CI(X 锁)
写入 LastUpdatedDateTime 索引(X 锁定)
他们以相反的顺序访问两个不兼容的资源。死锁。
阅读器是否在高于READ COMMITTED
的隔离级别下运行? READ COMMITTED
应该在读取锁定的行后立即释放锁。
如果您可以使读取事务使用快照隔离模型,死锁将以简单、可靠和永久的方式消失。
如果不可能,请使用READ COMMITTED
隔离级别。
如果这不可行,您将不得不弄乱锁定提示(难以维护)或实现死锁重试逻辑。
【讨论】:
感谢您的快速回复!现在我想我明白他们是如何陷入僵局的。您不知道有一种方法可以跟踪获取锁的顺序吗? READ COMMITTED 已经是当前的隔离级别了,还能死锁吗? 如果 iso 级别真的是 READ COMMITTED 他们不应该死锁。也许这里正在发生其他事情。 SQL Profiler 可以跟踪锁。 这肯定是 READ COMMITTED,我想我得进一步调查一下以上是关于INSERT 和 SELECT SP 之间的死锁的主要内容,如果未能解决你的问题,请参考以下文章
mysql insert into select 语句为啥会造成死锁
InnoDB RR隔离级别下INSERT SELECT两种死锁案例剖析