Question

Can someone tell me why the following SQL Server queries is deadlocking and what is the solution to fixing it?

<deadlock-list>
  <deadlock victim="process88b5b8">
    <process-list>
      <process id="process88b5b8" taskpriority="0" logused="76132" waitresource="RID: 32:1:151867:174" waittime="5093" ownerId="65554098" transactionguid="0xedf3314c05f1124cbe8d480cd092e03e" transactionname="DTCXact" lasttranstarted="2011-09-02T19:00:29.690" XDES="0x1029e040" lockMode="S" schedulerid="1" kpid="5108" status="suspended" spid="118" sbid="0" ecid="0" priority="0" transcount="2" lastbatchstarted="2011-09-02T19:00:31.317" lastbatchcompleted="2011-09-02T19:00:31.300" hostname="MELWFPL382S" hostpid="0" loginname="MM4" isolationlevel="repeatable read (3)" xactid="65554098" currentdb="32" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
        <executionStack>
          <frame procname="adhoc" line="1" stmtstart="78" sqlhandle="0x020000004b4b0a0d63e1040095143cbaa0174ffc3e076067"> delete from PARTIES where PARTYEXTERNALREF=@P0 and ISCOUNTERPARTY='N' and PARTYID in (select PARTYID from NAB_PARTY_EXTEND (nolock) where PARTYTYPE=@P1)     </frame>
          <frame procname="unknown" line="1" sqlhandle="0x000000000000000000000000000000000000000000000000">unknown     </frame>
        </executionStack>
        <inputbuf>(@P0 nvarchar(4000),@P1 nvarchar(4000))delete from PARTIES where PARTYEXTERNALREF=@P0 and ISCOUNTERPARTY='N' and PARTYID in (select PARTYID from NAB_PARTY_EXTEND (nolock) where PARTYTYPE=@P1)                    </inputbuf>
      </process>
      <process id="process9196a8" taskpriority="0" logused="132612" waitresource="RID: 32:1:140302:31" waittime="5046" ownerId="65554657" transactionguid="0x7313c78fecc8914dac3ed821cd7c21fe" transactionname="DTCXact" lasttranstarted="2011-09-02T19:00:34.100" XDES="0x12835778" lockMode="S" schedulerid="2" kpid="3692" status="suspended" spid="94" sbid="0" ecid="0" priority="0" transcount="2" lastbatchstarted="2011-09-02T19:00:35.690" lastbatchcompleted="2011-09-02T19:00:35.687" hostname="MELWFPL382S" hostpid="0" loginname="MM4" isolationlevel="repeatable read (3)" xactid="65554657" currentdb="32" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
        <executionStack>
          <frame procname="adhoc" line="1" stmtstart="78" sqlhandle="0x020000004b4b0a0d63e1040095143cbaa0174ffc3e076067">delete from PARTIES where PARTYEXTERNALREF=@P0 and ISCOUNTERPARTY='N' and PARTYID in (select PARTYID from NAB_PARTY_EXTEND (nolock) where PARTYTYPE=@P1)     </frame>
          <frame procname="unknown" line="1" sqlhandle="0x000000000000000000000000000000000000000000000000">unknown     </frame>
        </executionStack>
        <inputbuf>(@P0 nvarchar(4000),@P1 nvarchar(4000))delete from PARTIES where PARTYEXTERNALREF=@P0 and ISCOUNTERPARTY='N' and PARTYID in (select PARTYID from NAB_PARTY_EXTEND (nolock) where PARTYTYPE=@P1)                    </inputbuf>
      </process>
    </process-list>
    <resource-list>
      <ridlock fileid="1" pageid="140302" dbid="32" objectname="mm4_melwfpl382s.dbo.COUNTERPARTYSSI" id="lock170fa500" mode="X" associatedObjectId="72057595803336704">
        <owner-list>
          <owner id="process88b5b8" mode="X" />
        </owner-list>
        <waiter-list>
          <waiter id="process9196a8" mode="S" requestType="wait" />
        </waiter-list>
      </ridlock>
      <ridlock fileid="1" pageid="151867" dbid="32" objectname="mm4_melwfpl382s.dbo.COUNTERPARTYSSI" id="lock20e65d80" mode="X" associatedObjectId="72057595803336704">
        <owner-list>
          <owner id="process9196a8" mode="X" />
        </owner-list>
        <waiter-list>
          <waiter id="process88b5b8" mode="S" requestType="wait" />
        </waiter-list>
      </ridlock>
    </resource-list>
  </deadlock>
</deadlock-list>

What I don’t understand is how two process can have an exclusive lock on the same object.

There is an index (IDX_NC_PARTIES_PARTYEXTERNALREF_ISCOUNTERPARTY_PARTYID) on the PARTIES table and the database is set to read committed snapshot.

Thanks,

Wayne.

Answer 1

• process 9196a8 has page 151867 slot 174 in X mode and wants page 140302 slot 31 in S mode
• process 88b5b8 has page 140302 slot 31 in X mode and wants page 151867 slot 174 in S mode
• the two deletes run under isolationlevel="repeatable read (3)"

So the deadlock occur on the base heap of the table (RID locks instead of key locks implies a heap not a Btree). The high isolation level (likely caused by DTC, judging from the xact name) makes the RCSI setting irrelevant.

What type are the columns PARTYEXTERNALREF and PARTYTYPE? The parameters passed in are NVARCHAR (ie. Unicode) and if the columns are VARCHAR (ie. Ascii) then due to the rules of data type precedence the NC index would not be used. Because of the table scan involved, together with of the high isolation level in use, a deadlock is almost unavoidable.

The solution would be to use VARCHAR type parameters for @P0 and @P1 so the NC index would be leveraged to avoid the table scan.

If the parameters are already of type VARCHAR and you can confirm from the execution plan that a seek on the NC is used, the my first question would be what else is the transaction doing, other than the delete statements?

BTW, you only give the name of the NC index but I assume is on (PARTYEXTERNALREF, ISCOUNTERPARTY, PARTYID).

Update

Since your comment say that the columns are NVARCHAR then the tables scan hypothesis is probably wrong. There are three more possibilities to cause a deadlock that need investigation:

• any other statement run by the transaction before the DELETE (this is the most likely)
• any overlap in the rows selected by two DELETE statements involved in the deadlock
• hash collision

For the first two hypotheses only you can do anything right now (investigate if they are correct). For the last one I can tell you how to verify it, but is not trivial. It is unlikely to happen and a bit difficult to prove, but it is possible. Since you know of as deadlock case (the attached XML), use it as investigation base:

• restore a point-in-time copy of the database with stop at 2011-09-02T19:00:29.690
• run DBCC TRACEON(3604,-1)
• using DBCC PAGE (<restored db id>, 1, 151867, 3) inspect the values in slot 174
• using DBCC PAGE(, 1, 140302, 3)` inspect the values at slot 31
• run SELECT %%lockres%% FROM PARTIES WHERE PARTYEXTERNALREF = ... AND ISCOUNTERPARTY='N' and PARTYID=... and pass in the values read above
• compars the resulted lock hash values, if they match then you have a hash collision and this caused the deadlock.