Uploaded image for project: 'Core Server'
  1. Core Server
  2. SERVER-4649

mongod primary locks up and never kills cursors stalling connections for ever

    • Type: Icon: Bug Bug
    • Resolution: Incomplete
    • Priority: Icon: Critical - P2 Critical - P2
    • None
    • Affects Version/s: 2.0.1
    • Component/s: None
    • Labels:
    • Environment:
      Debian Squeeze with official 10gen deb packages.
      Two servers running on a replicaset.

      Each server is also running numactl --interleave=all wrapper.

      Mixed clients but mainly used by the PHP driver.
    • ALL

      At what seems to be random times, the mongod primary never seems to finish some queries.
      Some cursors had been running for 6+ hours. It has only happened on the primary.

      The log always has these entries when this occurs:

      [conn123812] warning: virtual size (129449MB) - mapped size (123342MB) is large. could indicate a memory leak

      [initandlisten] connection accepted from 172.16.49.98:40540 #123818
      [conn123816] end connection 172.16.49.98:40539
      [conn123817] Assertion failure cc->_pinValue < 100 db/clientcursor.h 309
      [conn123817] killcursors exception: assertion db/clientcursor.h:309 1ms

      And as we can see in the log, the queries that work seem to take forever:
      [conn123170] query x.scores nscanned:1015 scanAndOrder:1 nreturned:306 reslen:12260 302422ms

      The server resources doesn't seem to be exhausted and restarting the server gets everything back to normal again.

        1. pinvalue.txt
          30 kB
        2. oplog - start to crash.log
          1.89 MB
        3. oplog.log
          2.44 MB
        4. mms2.gif
          mms2.gif
          131 kB
        5. mms.gif
          mms.gif
          65 kB
        6. currentop.log
          998 kB

            Assignee:
            kristina Kristina Chodorow (Inactive)
            Reporter:
            balboah Johnny Boy
            Votes:
            0 Vote for this issue
            Watchers:
            7 Start watching this issue

              Created:
              Updated:
              Resolved: