Uploaded image for project: 'Hive'
  1. Hive
  2. HIVE-4679

WebHCat can deadlock Hadoop if the number of concurrently running tasks if higher or equal than the number of mappers

    XMLWordPrintableJSON

    Details

    • Type: Bug
    • Status: Closed
    • Priority: Major
    • Resolution: Fixed
    • Affects Version/s: 0.10.0
    • Fix Version/s: 0.12.0
    • Component/s: HCatalog
    • Labels:
      None

      Description

      o In the current Templeton design, each time a Job is submitted thru the REST API (it can be Pig/Hive or MR job), it will consume one Hadoop map slot. Given that the number of map slots is finite in the cluster (16 node cluster will have 32 map slots), in some circumstances, a user can deadlock the cluster if Templeton job submission pipeline takes over all map slots (Templeton map tasks will wait for the actual underlying jobs to complete, what will never happen, given that Hadoop has no free map slots to schedule new tasks).

      o HCat queries use a different mechanism and do not contribute to the deadlock.

        Attachments

        1. HIVE-4679.2.patch
          3 kB
          Eugene Koifman
        2. HIVE-4679.patch
          3 kB
          Eugene Koifman

          Issue Links

            Activity

              People

              • Assignee:
                ekoifman Eugene Koifman
                Reporter:
                ekoifman Eugene Koifman
              • Votes:
                0 Vote for this issue
                Watchers:
                5 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: