我有一个存储过程,其中在Begin和Commit tran下有以下两个事务。
UPDATE mytable
SET UserID = @ToUserID
WHERE UserID = @UserID
DELETE FROM mytable
WHERE UserID = @UserID
在多次执行该存储过程时,我陷入了死锁。这是死锁图:
<deadlock-list>
<deadlock victim="process16409057468">
<process-list>
<process id="process16409057468" taskpriority="0" logused="912" waitresource="RID: 6:1:2392:152" waittime="3022" ownerId="6283339" transactionname="user_transaction" lasttranstarted="2019-02-08T21:08:24.663" XDES="0x16401b98490" lockMode="U" schedulerid="8" kpid="23924" status="suspended" spid="92" sbid="0" ecid="0" priority="0" trancount="2" lastbatchstarted="2019-02-08T21:08:24.667" lastbatchcompleted="2019-02-08T21:08:24.667" lastattention="1900-01-01T00:00:00.667" clientapp=".Net SqlClient Data Provider" hostname="GYAAN" hostpid="5624" loginname="sa" isolationlevel="read uncommitted (1)" xactid="6283339" currentdb="6" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
<executionStack>
<frame procname="mytable" line="377" stmtstart="33320" stmtend="33540" sqlhandle="0x030006004f6bf63211085201eaa9000001000000000000000000000000000000000000000000000000000000">
UPDATE mytable
SET UserID = @ToUserID
WHERE UserID = @UserID
</frame>
</executionStack>
<inputbuf>
Proc [Database Id = 6 Object Id = 855010127]
</inputbuf>
</process>
<process id="process163feab3088" taskpriority="0" logused="912" waitresource="RID: 6:1:2392:149" waittime="99" ownerId="6282851" transactionname="user_transaction" lasttranstarted="2019-02-08T21:08:22.107" XDES="0x16401b20490" lockMode="U" schedulerid="3" kpid="33220" status="suspended" spid="81" sbid="0" ecid="0" priority="0" trancount="2" lastbatchstarted="2019-02-08T21:08:22.103" lastbatchcompleted="2019-02-08T21:08:22.107" lastattention="1900-01-01T00:00:00.107" clientapp=".Net SqlClient Data Provider" hostname="GYAAN" hostpid="5624" loginname="sa" isolationlevel="read uncommitted (1)" xactid="6282851" currentdb="6" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
<executionStack>
<frame procname="mytable" line="382" stmtstart="33650" stmtend="33848" sqlhandle="0x030006004f6bf63211085201eaa9000001000000000000000000000000000000000000000000000000000000">
DELETE FROM mytable
WHERE UserID = @UserID
</frame>
</executionStack>
<inputbuf>
Proc [Database Id = 6 Object Id = 855010127]
</inputbuf>
</process>
</process-list>
<resource-list>
<ridlock fileid="1" pageid="2392" dbid="6" objectname="mytable" id="lock164096b7800" mode="X" associatedObjectId="72057594051493888">
<owner-list>
<owner id="process163feab3088" mode="X"/>
</owner-list>
<waiter-list>
<waiter id="process16409057468" mode="U" requestType="wait"/>
</waiter-list>
</ridlock>
<ridlock fileid="1" pageid="2392" dbid="6" objectname="mytable" id="lock163f7fb2c80" mode="X" associatedObjectId="72057594051493888">
<owner-list>
<owner id="process16409057468" mode="X"/>
</owner-list>
<waiter-list>
<waiter id="process163feab3088" mode="U" requestType="wait"/>
</waiter-list>
</ridlock>
</resource-list>
</deadlock>
</deadlock-list>
有人可以解释一下死锁的原因和原因吗?
此刻我很难理解这个X和U锁定流程。
您能解释一下流程X和U卡在死锁之间会发生什么吗?
答案 0 :(得分:8)
您没有提供足够的查询详细信息,但是您共享的死锁图清楚地表明,由于并行性,它是“ writer-writer” 死锁,因为所有授予或请求的锁都是 X 或 U 。
<resource-list>
<ridlock fileid="1" pageid="2392" dbid="6" objectname="xx" id="lock164096b7800" mode="X" associatedObjectId="72057594051493888">
<owner-list>
<owner id="process163feab3088" mode="X"/>
</owner-list>
<waiter-list>
<waiter id="process16409057468" mode="U" requestType="wait"/>
</waiter-list>
</ridlock>
<ridlock fileid="1" pageid="2392" dbid="6" objectname="mytable" id="lock163f7fb2c80" mode="X" associatedObjectId="72057594051493888">
<owner-list>
<owner id="process16409057468" mode="X"/>
</owner-list>
<waiter-list>
<waiter id="process163feab3088" mode="U" requestType="wait"/>
</waiter-list>
</ridlock>
</resource-list>
关于 writer-writer 死锁的重要一件事是 SQL Server 持有排他锁直到事务提交,这与共享锁不同,共享锁在默认的read committed
中隔离级别。
没有查询详细信息,就不会在论文中找出错误的确切原因。通常,您需要对查询进行重构,以避免类似
的死锁。SELECT
查询移到事务外部,以便仅返回
提交的数据,而不是返回包含修改的数据
可能会回滚。MAXDOP
提示以强制其顺序运行将删除
查询内并行性死锁的任何更改。死锁的另一个常见原因是,当您读取数据并打算稍后通过放置共享锁来更新或删除数据时,以下UPDATE
语句无法获取必要的更新锁,因为资源是已经被导致死锁的另一个进程阻止了。
要解决此问题,您可以使用WITH (SERIALIZABLE)
选择记录,如下所示
UPDATE mytable WITH (SERIALIZABLE)
SET UserID = @ToUserID
WHERE UserID = @UserID
这将对记录进行必要的更新锁,并将停止其他过程以获取记录上的任何锁(共享/独占),并防止出现任何死锁。
您还需要查找查询的顺序,错误的顺序可能导致循环死锁。在这种情况下,查询会等待另一笔交易完成。
答案 1 :(得分:2)
除了您的示例是先更新然后删除同一记录(看起来是错误的)外,您可以在执行DDL之前获取所有必需的xlock。
select UserID
FROM mytable with(xlock, holdlock, rowlock)
WHERE UserID in (@ToUserID, @UserID)
UPDATE mytable
SET UserID = @ToUserID
WHERE UserID = @UserID
DELETE FROM mytable
WHERE UserID = @UserID