Uploaded image for project: 'Core Server'
  1. Core Server
  2. SERVER-14342

Shard crashes on split operation if collection doesn't exist on that shard

    XMLWordPrintable

    Details

    • Type: Bug
    • Status: Closed
    • Priority: Major - P3
    • Resolution: Fixed
    • Affects Version/s: 2.4.10
    • Fix Version/s: 2.4.11
    • Component/s: Sharding
    • Labels:
      None
    • Operating System:
      ALL

      Description

      Issue Status as of Aug 21, 2014

      ISSUE SUMMARY
      On a sharded cluster, a shard may crash when a splitChunkCommand operation is issued on a non-existing namespace. One way to reproduce this is to connect to a shard, drop a sharded collection and then issue the split command through mongos on that collection.

      Non-sharded systems are not affected by this issue.

      USER IMPACT
      The shard where the namespace doesn't exist logs a stack trace in the log and crashes.

      WORKAROUNDS
      N/A

      AFFECTED VERSIONS
      MongoDB 2.4 production releases up to 2.4.10 are affected by this issue.

      FIX VERSION
      The fix is included in the 2.4.11 production release.

      RESOLUTION DETAILS
      Do not perform any split operations if the namespace doesn't exist on a given shard.

      Original description

      Mon Jun 16 15:17:05.222 Invalid access at address: 0xbc from thread: conn48
      Mon Jun 16 15:17:05.222 Got signal: 11 (Segmentation fault).
      Mon Jun 16 15:17:05.227 Backtrace:
      0xde8c31 0x6d0b19 0x6d10a2 0x7f1613e04030 0xce06dc 0x8e1f9a 0x8e2a1d 0x8e3f72 0xa89910 0xa8e1dc 0xa0182e 0xa02b73 0x6eb838 0xdd51fe 0x7f1613dfbb50 0x7f161319f0ed 
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo15printStackTraceERSo+0x21) [0xde8c31]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo10abruptQuitEi+0x399) [0x6d0b19]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo24abruptQuitWithAddrSignalEiP7siginfoPv+0x262) [0x6d10a2]
       /lib/x86_64-linux-gnu/libpthread.so.0(+0xf030) [0x7f1613e04030]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo17SplitChunkCommand3runERKSsRNS_7BSONObjEiRSsRNS_14BSONObjBuilderEb+0x8e9c) [0xce06dc]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo12_execCommandEPNS_7CommandERKSsRNS_7BSONObjEiRSsRNS_14BSONObjBuilderEb+0x3a) [0x8e1f9a]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo7Command11execCommandEPS0_RNS_6ClientEiPKcRNS_7BSONObjERNS_14BSONObjBuilderEb+0x71d) [0x8e2a1d]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo12_runCommandsEPKcRNS_7BSONObjERNS_11_BufBuilderINS_16TrivialAllocatorEEERNS_14BSONObjBuilderEbi+0x5f2) [0x8e3f72]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo11runCommandsEPKcRNS_7BSONObjERNS_5CurOpERNS_11_BufBuilderINS_16TrivialAllocatorEEERNS_14BSONObjBuilderEbi+0x40) [0xa89910]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo8runQueryERNS_7MessageERNS_12QueryMessageERNS_5CurOpES1_+0xd7c) [0xa8e1dc]
       /intucell/packages/mongodb-2.4.10/bin/mongod() [0xa0182e]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo16assembleResponseERNS_7MessageERNS_10DbResponseERKNS_11HostAndPortE+0x393) [0xa02b73]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo16MyMessageHandler7processERNS_7MessageEPNS_21AbstractMessagingPortEPNS_9LastErrorE+0x98) [0x6eb838]
       /intucell/packages/mongodb-2.4.10/bin/mongod(_ZN5mongo17PortMessageServer17handleIncomingMsgEPv+0x42e) [0xdd51fe]
       /lib/x86_64-linux-gnu/libpthread.so.0(+0x6b50) [0x7f1613dfbb50]
       /lib/x86_64-linux-gnu/libc.so.6(clone+0x6d) [0x7f161319f0ed]
      *** End of input ***

      $ addr2line -i -e mongod 0xce06dc
      /mntfast/data/slave/Linux_64bit_V2.4/mongo/src/mongo/db/namespace_details-inl.h:101
      /mntfast/data/slave/Linux_64bit_V2.4/mongo/src/mongo/db/namespace_details.h:214
      /mntfast/data/slave/Linux_64bit_V2.4/mongo/src/mongo/db/namespace_details-inl.h:72
      /mntfast/data/slave/Linux_64bit_V2.4/mongo/src/mongo/s/d_split.cpp:807

      d_split.cpp v 2.4.10

      ...
                          Client::ReadContext ctx( ns );
                          NamespaceDetails *d = nsdetails( ns );
       
      807:                const IndexDetails *idx = d->findIndexByPrefix( keyPattern ,
                                                                          true ); /* exclude multikeys */
                          if ( idx == NULL ) {
                              break;
                          }

      This can be triggered when the namespace does not exists when the split command was issued. One way to reproduce this is to connect to the shard, drop the collection and then issue the split command through mongos.

        Attachments

          Issue Links

            Activity

              People

              • Votes:
                0 Vote for this issue
                Watchers:
                9 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: