哈希冲突导致死锁？

Posted 2023-03-22

技术标签:

【中文标题】哈希冲突导致死锁？【英文标题】：Deadlock due to hash collision? 【发布时间】：2012-10-14 04:23:11 【问题描述】：

我有一个死锁，这是由于两个不同的进程 ID 在同一个聚集索引上的排他锁而发生的。数据库处于 RCSI 模式。我有两个问题。

1) 除了两个 UPDATE 语句重叠的行之外，这可能是由哈希冲突引起的吗？

2) 在下面的跟踪 XML 文件中，查询上有参数。我应该使用 SQL Profiler 跟踪哪些事件以获取参数值？

感谢您的帮助。谢谢

跟踪 XML 文件

<deadlock-list>
 <deadlock victim="process8ad948">
  <process-list>
   <process id="process8ad948" taskpriority="0" logused="164520" waitresource="KEY: 7:72057601661599744 (01c1926aca97)" waittime="2833" ownerId="7491036" transactionname="user_transaction" lasttranstarted="2012-10-08T13:16:03.067" XDES="0x6b561b950" lockMode="U" schedulerid="4" kpid="2476" status="suspended" spid="94" sbid="0" ecid="0" priority="0" trancount="2" lastbatchstarted="2012-10-08T13:16:57.753" lastbatchcompleted="2012-10-08T13:16:57.750" clientapp="PeopleSoft" hostname="STGAAMY-PPSAPP" hostpid="31772" loginname="HRTRNADM" isolationlevel="read committed (2)" xactid="7491036" currentdb="7" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
    <executionStack>
     <frame procname="adhoc" line="1" stmtstart="440" sqlhandle="0x020000007b8741293b8e4d6542633962bc48489d3946a5d1">
UPDATE PS_GP_NEW_RTO_WRK SET ORIG_CAL_RUN_ID = (SELECT DISTINCT C.ORIG_CAL_RUN_ID FROM PS_GP_PYE_PRC_STAT B ,PS_GP_OLD_RTO_WRK C WHERE C.EMPLID BETWEEN @P1 AND @P2 AND C.CAL_RUN_ID=@P3 AND C.EMPLID = PS_GP_NEW_RTO_WRK.EMPLID AND C.CAL_RUN_ID = PS_GP_NEW_RTO_WRK.CAL_RUN_ID AND C.EMPL_RCD = PS_GP_NEW_RTO_WRK.EMPL_RCD AND C.CAL_ID = PS_GP_NEW_RTO_WRK.CAL_ID AND C.GP_PAYGROUP = PS_GP_NEW_RTO_WRK.GP_PAYGROUP AND B.CAL_ID= C.CAL_ID AND B.GP_PAYGROUP= C.GP_PAYGROUP AND B.EMPLID=C.EMPLID AND B.EMPL_RCD=C.EMPL_RCD AND B.ORIG_CAL_RUN_ID=C.ORIG_CAL_RUN_ID AND C.SEL_ACTION IN (N&apos;R&apos;,N&apos;Z&apos;) AND B.SEL_ACTION=N&apos;A&apos;) WHERE EMPLID BETWEEN @P4 AND @P5 AND CAL_RUN_ID=@P6 AND (N&apos;N&apos;=@P7 OR EMPLID IN (SELECT EMPLID FROM PS_GP_GRP_LIST_RUN WHERE RUN_CNTL_ID=@P8 AND OPRID=@P9) ) AND EXISTS (SELECT B.ORIG_CAL_RUN_ID FROM PS_GP_PYE_PRC_STAT B ,PS_GP_OLD_RTO_WRK C WHERE C.EMPLID BETWEEN @P10 AND @P11 AND C.CAL_RUN_ID=@P12 AND C.EMPLID = PS_GP_NEW_RTO_WRK.EMPLID AND C.CAL_RUN_ID = PS_GP_NEW_RTO_WRK.CAL_RUN_ID AND C.EMPL_RCD = PS_GP_NEW_RTO     </frame>
     <frame procname="unknown" line="1" sqlhandle="0x000000000000000000000000000000000000000000000000">
unknown     </frame>
    </executionStack>
    <inputbuf>
(@P1 nvarchar(508),@P2 nvarchar(508),@P3 nvarchar(508),@P4 nvarchar(508),@P5 nvarchar(508),@P6 nvarchar(508),@P7 nvarchar(508),@P8 nvarchar(508),@P9 nvarchar(508),@P10 nvarchar(508),@P11 nvarchar(508),@P12 nvarchar(508))UPDATE PS_GP_NEW_RTO_WRK SET ORIG_CAL_RUN_ID = (SELECT DISTINCT C.ORIG_CAL_RUN_ID FROM PS_GP_PYE_PRC_STAT B ,PS_GP_OLD_RTO_WRK C WHERE C.EMPLID BETWEEN @P1 AND @P2 AND C.CAL_RUN_ID=@P3 AND C.EMPLID = PS_GP_NEW_RTO_WRK.EMPLID AND C.CAL_RUN_ID = PS_GP_NEW_RTO_WRK.CAL_RUN_ID AND C.EMPL_RCD = PS_GP_NEW_RTO_WRK.EMPL_RCD AND C.CAL_ID = PS_GP_NEW_RTO_WRK.CAL_ID AND C.GP_PAYGROUP = PS_GP_NEW_RTO_WRK.GP_PAYGROUP AND B.CAL_ID= C.CAL_ID AND B.GP_PAYGROUP= C.GP_PAYGROUP AND B.EMPLID=C.EMPLID AND B.EMPL_RCD=C.EMPL_RCD AND B.ORIG_CAL_RUN_ID=C.ORIG_CAL_RUN_ID AND C.SEL_ACTION IN (N&apos;R&apos;,N&apos;Z&apos;) AND B.SEL_ACTION=N&apos;A&apos;) WHERE EMPLID BETWEEN @P4 AND @P5 AND CAL_RUN_ID=@P6 AND (N&apos;N&apos;=@P7 OR EMPLID IN (SELECT EMPLID FROM PS_GP_GRP_LIST_RUN WHERE RUN_CNTL_ID=@P8 AND OPRID=@P9) ) AND EXISTS (SELECT B.ORIG_CAL_RUN_ID FRO    </inputbuf>
   </process>
   <process id="process5ff948" taskpriority="0" logused="2793096" waitresource="KEY: 7:72057601661599744 (9e1bb6c02ace)" waittime="1659" ownerId="7509883" transactionname="user_transaction" lasttranstarted="2012-10-08T13:16:50.867" XDES="0x6b6beb950" lockMode="U" schedulerid="1" kpid="2472" status="suspended" spid="95" sbid="0" ecid="0" priority="0" trancount="2" lastbatchstarted="2012-10-08T13:16:59.400" lastbatchcompleted="2012-10-08T13:16:59.400" clientapp="PeopleSoft" hostname="STGAAMY-PPSAPP" hostpid="32912" loginname="HRTRNADM" isolationlevel="read committed (2)" xactid="7509883" currentdb="7" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
    <executionStack>
     <frame procname="adhoc" line="1" stmtstart="440" sqlhandle="0x020000007b8741293b8e4d6542633962bc48489d3946a5d1">
UPDATE PS_GP_NEW_RTO_WRK SET ORIG_CAL_RUN_ID = (SELECT DISTINCT C.ORIG_CAL_RUN_ID FROM PS_GP_PYE_PRC_STAT B ,PS_GP_OLD_RTO_WRK C WHERE C.EMPLID BETWEEN @P1 AND @P2 AND C.CAL_RUN_ID=@P3 AND C.EMPLID = PS_GP_NEW_RTO_WRK.EMPLID AND C.CAL_RUN_ID = PS_GP_NEW_RTO_WRK.CAL_RUN_ID AND C.EMPL_RCD = PS_GP_NEW_RTO_WRK.EMPL_RCD AND C.CAL_ID = PS_GP_NEW_RTO_WRK.CAL_ID AND C.GP_PAYGROUP = PS_GP_NEW_RTO_WRK.GP_PAYGROUP AND B.CAL_ID= C.CAL_ID AND B.GP_PAYGROUP= C.GP_PAYGROUP AND B.EMPLID=C.EMPLID AND B.EMPL_RCD=C.EMPL_RCD AND B.ORIG_CAL_RUN_ID=C.ORIG_CAL_RUN_ID AND C.SEL_ACTION IN (N&apos;R&apos;,N&apos;Z&apos;) AND B.SEL_ACTION=N&apos;A&apos;) WHERE EMPLID BETWEEN @P4 AND @P5 AND CAL_RUN_ID=@P6 AND (N&apos;N&apos;=@P7 OR EMPLID IN (SELECT EMPLID FROM PS_GP_GRP_LIST_RUN WHERE RUN_CNTL_ID=@P8 AND OPRID=@P9) ) AND EXISTS (SELECT B.ORIG_CAL_RUN_ID FROM PS_GP_PYE_PRC_STAT B ,PS_GP_OLD_RTO_WRK C WHERE C.EMPLID BETWEEN @P10 AND @P11 AND C.CAL_RUN_ID=@P12 AND C.EMPLID = PS_GP_NEW_RTO_WRK.EMPLID AND C.CAL_RUN_ID = PS_GP_NEW_RTO_WRK.CAL_RUN_ID AND C.EMPL_RCD = PS_GP_NEW_RTO     </frame>
     <frame procname="unknown" line="1" sqlhandle="0x000000000000000000000000000000000000000000000000">
unknown     </frame>
    </executionStack>
    <inputbuf>
(@P1 nvarchar(508),@P2 nvarchar(508),@P3 nvarchar(508),@P4 nvarchar(508),@P5 nvarchar(508),@P6 nvarchar(508),@P7 nvarchar(508),@P8 nvarchar(508),@P9 nvarchar(508),@P10 nvarchar(508),@P11 nvarchar(508),@P12 nvarchar(508))UPDATE PS_GP_NEW_RTO_WRK SET ORIG_CAL_RUN_ID = (SELECT DISTINCT C.ORIG_CAL_RUN_ID FROM PS_GP_PYE_PRC_STAT B ,PS_GP_OLD_RTO_WRK C WHERE C.EMPLID BETWEEN @P1 AND @P2 AND C.CAL_RUN_ID=@P3 AND C.EMPLID = PS_GP_NEW_RTO_WRK.EMPLID AND C.CAL_RUN_ID = PS_GP_NEW_RTO_WRK.CAL_RUN_ID AND C.EMPL_RCD = PS_GP_NEW_RTO_WRK.EMPL_RCD AND C.CAL_ID = PS_GP_NEW_RTO_WRK.CAL_ID AND C.GP_PAYGROUP = PS_GP_NEW_RTO_WRK.GP_PAYGROUP AND B.CAL_ID= C.CAL_ID AND B.GP_PAYGROUP= C.GP_PAYGROUP AND B.EMPLID=C.EMPLID AND B.EMPL_RCD=C.EMPL_RCD AND B.ORIG_CAL_RUN_ID=C.ORIG_CAL_RUN_ID AND C.SEL_ACTION IN (N&apos;R&apos;,N&apos;Z&apos;) AND B.SEL_ACTION=N&apos;A&apos;) WHERE EMPLID BETWEEN @P4 AND @P5 AND CAL_RUN_ID=@P6 AND (N&apos;N&apos;=@P7 OR EMPLID IN (SELECT EMPLID FROM PS_GP_GRP_LIST_RUN WHERE RUN_CNTL_ID=@P8 AND OPRID=@P9) ) AND EXISTS (SELECT B.ORIG_CAL_RUN_ID FRO    </inputbuf>
   </process>
  </process-list>
  <resource-list>
   <keylock hobtid="72057601661599744" dbid="7" objectname="HRTRN91.dbo.PS_GP_NEW_RTO_WRK" indexname="PS_GP_NEW_RTO_WRK" id="lock4991e6c00" mode="X" associatedObjectId="72057601661599744">
    <owner-list>
     <owner id="process5ff948" mode="X"/>
    </owner-list>
    <waiter-list>
     <waiter id="process8ad948" mode="U" requestType="wait"/>
    </waiter-list>
   </keylock>
   <keylock hobtid="72057601661599744" dbid="7" objectname="HRTRN91.dbo.PS_GP_NEW_RTO_WRK" indexname="PS_GP_NEW_RTO_WRK" id="lockdf24e800" mode="X" associatedObjectId="72057601661599744">
    <owner-list>
     <owner id="process8ad948" mode="X"/>
    </owner-list>
    <waiter-list>
     <waiter id="process5ff948" mode="U" requestType="wait"/>
    </waiter-list>
   </keylock>
  </resource-list>
 </deadlock>
</deadlock-list>

表结构

CREATE TABLE [dbo].[PS_GP_NEW_RTO_WRK](
    [EMPLID] [nvarchar](11) NOT NULL,
    [CAL_RUN_ID] [nvarchar](18) NOT NULL,
    [EMPL_RCD] [smallint] NOT NULL,
    [GP_PAYGROUP] [nvarchar](10) NOT NULL,
    [CAL_ID] [nvarchar](18) NOT NULL,
    [ORIG_CAL_RUN_ID] [nvarchar](18) NOT NULL,
    [RSLT_VER_NUM] [smallint] NOT NULL,
    [RSLT_REV_NUM] [smallint] NOT NULL,
    [PRC_ORD_TS] [dbo].[PSDATETIME] NULL,
    [SEL_STAT] [nvarchar](1) NOT NULL,
    [SEL_RSN] [nvarchar](2) NOT NULL,
    [RTO_PRC_ID] [nvarchar](10) NOT NULL,
    [PYMT_DT] [dbo].[PSDATE] NULL,
    [PRC_BGN_DT] [dbo].[PSDATE] NULL,
    [PRC_END_DT] [dbo].[PSDATE] NULL
) ON [PRIMARY]

GO

ALTER TABLE [dbo].[PS_GP_NEW_RTO_WRK] SET (LOCK_ESCALATION = ENABLE)
GO

索引结构

CREATE UNIQUE CLUSTERED INDEX [PS_GP_NEW_RTO_WRK] ON [dbo].[PS_GP_NEW_RTO_WRK] 
(
    [EMPLID] ASC,
    [CAL_RUN_ID] ASC,
    [EMPL_RCD] ASC,
    [GP_PAYGROUP] ASC,
    [CAL_ID] ASC,
    [ORIG_CAL_RUN_ID] 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]

【问题讨论】：

【参考方案1】：

1) 在开发/测试服务器上恢复备份并运行此查询：

SELECT *
FROM (
    SELECT [EMPLID],[CAL_RUN_ID],[EMPL_RCD],[GP_PAYGROUP],[CAL_ID],[ORIG_CAL_RUN_ID], %%lockres%% AS KeyLockID
    FROM [dbo].[PS_GP_NEW_RTO_WRK]
) x
WHERE x.KeyLockID IN ('(01c1926aca97)','(9e1bb6c02ace)')

(01c1926aca97) 和 (9e1bb6c02ace) 值是“锁定 ID”（请参阅 /deadlock-list/deadlock/process-list/process/@waitresource 值：KEY: 7:72057601661599744 (01c1926aca97)、KEY: 7:72057601661599744 (9e1bb6c02ace)）。如果你得到 3 行或更多行，那么答案是肯定的。

2) 此外，您可以检查差异。计划：

SELECT qp.*, txt.*
FROM sys.dm_exec_cached_plans cp
CROSS APPLY sys.dm_exec_query_plan(cp.plan_handle) qp
CROSS APPLY sys.dm_exec_sql_text(cp.plan_handle) txt
WHERE txt.text LIKE N'UPDATE PS\_GP\_NEW\_RTO\_WRK SET ORIG\_CAL\_RUN\_ID = (SELECT DISTINCT C.ORIG\_CAL\_RUN\_ID FROM PS\_GP\_PYE\_PRC\_STAT B ,PS\_GP\_OLD\_RTO\_WRK C WHERE C.EMPLID BETWEEN @P1 AND @P2 AND C.CAL\_RUN\_ID=@P3 AND C.EMPLID = PS\_GP\_NEW\_RTO\_WRK.EMPLID AND C.CAL\_RUN\_ID = PS\_GP\_NEW\_RTO\_WRK.CAL\_RUN\_ID AND C.EMPL\_RCD = PS\_GP\_NEW\_RTO\_WRK.EMPL\_RCD AND C.CAL\_ID = PS\_GP\_NEW\_RTO\_WRK.CAL\_ID AND C.GP\_PAYGROUP = PS\_GP\_NEW\_RTO\_WRK.GP\_PAYGROUP AND B.CAL\_ID= C.CAL\_ID AND B.GP\_PAYGROUP= C.GP\_PAYGROUP AND B.EMPLID=C.EMPLID AND B.EMPL\_RCD=C.EMPL\_RCD AND B.ORIG\_CAL\_RUN\_ID=C.ORIG\_CAL\_RUN\_ID AND C.SEL\_ACTION IN (N&apos;R&apos;,N&apos;Z&apos;) AND B.SEL\_ACTION=N&apos;A&apos;) WHERE EMPLID BETWEEN @P4 AND @P5 AND CAL\_RUN\_ID=@P6 AND (N&apos;N&apos;=@P7 OR EMPLID IN (SELECT EMPLID FROM PS\_GP\_GRP\_LIST\_RUN WHERE RUN\_CNTL\_ID=@P8 AND OPRID=@P9) ) AND EXISTS (SELECT B.ORIG\_CAL\_RUN\_ID FROM PS\_GP\_PYE\_PRC\_STAT B ,PS\_GP\_OLD\_RTO\_WRK C WHERE C.EMPLID BETWEEN @P10 AND @P11 AND C.CAL\_RUN\_ID=@P12 AND C.EMPLID = PS\_GP\_NEW\_RTO\_WRK.EMPLID AND C.CAL\_RUN\_ID = PS\_GP\_NEW\_RTO\_WRK.CAL\_RUN\_ID AND C.EMPL\_RCD = PS\_GP\_NEW\_RTO%' ESCAPE N'\'
ORDER BY txt.text

SELECT qp.*, txt.*
FROM sys.dm_exec_cached_plans cp
CROSS APPLY sys.dm_exec_query_plan(cp.plan_handle) qp
CROSS APPLY sys.dm_exec_sql_text(cp.plan_handle) txt
WHERE txt.text LIKE N'(@P1 nvarchar(508),@P2 nvarchar(508),@P3 nvarchar(508),@P4 nvarchar(508),@P5 nvarchar(508),@P6 nvarchar(508),@P7 nvarchar(508),@P8 nvarchar(508),@P9 nvarchar(508),@P10 nvarchar(508),@P11 nvarchar(508),@P12 nvarchar(508))UPDATE PS\_GP\_NEW\_RTO\_WRK SET ORIG\_CAL\_RUN\_ID = (SELECT DISTINCT C.ORIG\_CAL\_RUN\_ID FROM PS\_GP\_PYE\_PRC\_STAT B ,PS\_GP\_OLD\_RTO\_WRK C WHERE C.EMPLID BETWEEN @P1 AND @P2 AND C.CAL\_RUN\_ID=@P3 AND C.EMPLID = PS\_GP\_NEW\_RTO\_WRK.EMPLID AND C.CAL\_RUN\_ID = PS\_GP\_NEW\_RTO\_WRK.CAL\_RUN\_ID AND C.EMPL\_RCD = PS\_GP\_NEW\_RTO\_WRK.EMPL\_RCD AND C.CAL\_ID = PS\_GP\_NEW\_RTO\_WRK.CAL\_ID AND C.GP\_PAYGROUP = PS\_GP\_NEW\_RTO\_WRK.GP\_PAYGROUP AND B.CAL\_ID= C.CAL\_ID AND B.GP\_PAYGROUP= C.GP\_PAYGROUP AND B.EMPLID=C.EMPLID AND B.EMPL\_RCD=C.EMPL\_RCD AND B.ORIG\_CAL\_RUN\_ID=C.ORIG\_CAL\_RUN\_ID AND C.SEL\_ACTION IN (N&apos;R&apos;,N&apos;Z&apos;) AND B.SEL\_ACTION=N&apos;A&apos;) WHERE EMPLID BETWEEN @P4 AND @P5 AND CAL\_RUN\_ID=@P6 AND (N&apos;N&apos;=@P7 OR EMPLID IN (SELECT EMPLID FROM PS\_GP\_GRP\_LIST\_RUN WHERE RUN\_CNTL\_ID=@P8 AND OPRID=@P9) ) AND EXISTS (SELECT B.ORIG\_CAL\_RUN\_ID FRO%' ESCAPE N'\'
ORDER BY txt.text

如果您为每个 UPDATE 或 (@...)UPDATE 语句获得不同的计划（例如串行计划和并行计划），那么这个事实会排除初始假设（哈希冲突）。

扩展评论：

T = SPID94 开始执行UPDATE语句（串行计划）

T+1 = SPID95开始执行UPDATE语句（并行计划）

T+2 = SPID94 在行/键 4 上获得 X 锁

T+3 = SPID95-thread1 尝试对行/键 4 进行 U 锁定（已被 SPID94 锁定）。此外，SPID95-thread2 在行/键 9 上采用 X 锁。

T+4 = SPID94 尝试对行/键 9 进行 U 锁定（已被 SPID95 锁定）。

结论：SPID94 在行/键 4 上有一个 X 锁，并在行/键 9 上请求一个 U 锁。SPID95 在行/键 9 上有一个 X 锁，并在行/键 4 上请求一个 U 锁。所以，我们陷入僵局。

【讨论】：

*** 这个事实可能不包括 *** 感谢 Bogdan，我得到了第 1 点，但在第 2 点，为什么每个更新语句都有不同的串行和并行计划排除哈希冲突的原因？两个差异。计划可以解释死锁而不是哈希冲突。通常，UPDATE 语句在行/键上获取 U 锁，验证过滤器 (WHERE)，如果过滤器为真，则 U 锁转换为 X 锁。看看我的评论（最后一个答案）。另外，尝试发布这两个 UPDATE 的执行计划。

以上是关于哈希冲突导致死锁？的主要内容，如果未能解决你的问题，请参考以下文章