Deadlock on bulk inserts

问题

I have several clients running an external executable using a C++ DLL to do bulk inserts though OLEDB into the same SQL Server table. The table has no indices, just a check constraint and a default. After I upgraded from SQL Server 2005 to 2008, all of a sudden I'm seeing deadlocks.

The deadlock XDL (anonymized):

<deadlock-list>
 <deadlock victim="process8057f048">
  <process-list>
   <process id="process8057f048" taskpriority="0" logused="0" waitresource="OBJECT: 21:1176443315:0 " waittime="3329" ownerId="33387042" transactionname="SetCnstNotTrusted" lasttranstarted="2011-10-05T09:15:12.227" XDES="0x1afc7c3b0" lockMode="Sch-M" schedulerid="2" kpid="2840" status="suspended" spid="66" sbid="0" ecid="0" priority="0" trancount="1" lastbatchstarted="2011-10-05T09:12:15.753" lastbatchcompleted="2011-10-05T09:12:15.750" clientapp="Ventyx Prosym" hostname="..." hostpid="5892" loginname="..." isolationlevel="read committed (2)" xactid="33371350" currentdb="21" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
    <executionStack>
     <frame procname="adhoc" line="1" sqlhandle="0x02000000d976c02496c28e4391ad91e2cf367700360cb812">
insert bulk table1(...)     </frame>
    </executionStack>
    <inputbuf>
insert bulk bulk table1(...)     </inputbuf>
   </process>
   <process id="processfffdc8" taskpriority="0" logused="0" waitresource="OBJECT: 21:1176443315:0 " waittime="44248" ownerId="33383515" transactionname="SetCnstNotTrusted" lasttranstarted="2011-10-05T09:14:31.310" XDES="0x1f7f503b0" lockMode="Sch-M" schedulerid="1" kpid="4940" status="suspended" spid="68" sbid="0" ecid="0" priority="0" trancount="1" lastbatchstarted="2011-10-05T09:12:15.720" lastbatchcompleted="2011-10-05T09:12:15.720" clientapp="Ventyx Prosym" hostname="EPMWIN7QA64-05" hostpid="8232" loginname="HESIAppUser" isolationlevel="read committed (2)" xactid="33371331" currentdb="21" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056">
    <executionStack>
     <frame procname="adhoc" line="1" sqlhandle="0x02000000d976c02496c28e4391ad91e2cf367700360cb812">
insert bulk bulk table1(...)     </frame>
    </executionStack>
    <inputbuf>
insert bulk bulk table1(...)     </inputbuf>
   </process>
  </process-list>
  <resource-list>
   <objectlock lockPartition="0" objid="1176443315" subresource="FULL" dbid="21" objectname="[...].dbo.table1" id="lock2dd099180" mode="IX" associatedObjectId="1176443315">
    <owner-list>
     <owner id="processfffdc8" mode="IX"/>
    </owner-list>
    <waiter-list>
     <waiter id="process8057f048" mode="Sch-M" requestType="wait"/>
    </waiter-list>
   </objectlock>
   <objectlock lockPartition="0" objid="1176443315" subresource="FULL" dbid="21" objectname="[...].dbo.table1" id="lock2dd099180" mode="IX" associatedObjectId="1176443315">
    <owner-list>
     <owner id="process8057f048" mode="IX"/>
    </owner-list>
    <waiter-list>
     <waiter id="processfffdc8" mode="Sch-M" requestType="wait"/>
    </waiter-list>
   </objectlock>
  </resource-list>
 </deadlock>
</deadlock-list>

Note that 1176443315 is the object ID for the table itself.

As pointed out by Remus, the problem was a CHECK CONSTRAINT on the table (note "SetCnstNotTrusted"). Luckily, I don't even need that constraint anymore (it was old, inherited and decrepit), and removing it did the trick.

回答1:

The SCH-M lock is because the bulk insert statement must mark the constraint as 'not trusted' (hence the transactionname="SetCnstNotTrusted"), which is a DDL operation. Disable the constraint before the bulk insert, then enable it back, or enforce the constraint during the bulk insert. See Controlling Constraint Checking by Bulk Import Operations.

来源：https://stackoverflow.com/questions/7691580/deadlock-on-bulk-inserts

标签

sql-server

oledb