Uploaded image for project: 'Hive'
  1. Hive
  2. HIVE-24577

Task resubmission bug

    XMLWordPrintableJSON

    Details

    • Type: Bug
    • Status: Open
    • Priority: Major
    • Resolution: Unresolved
    • Affects Version/s: 2.3.4
    • Fix Version/s: 2.3.8
    • Component/s: Hive
    • Labels:
      None
    • Environment:

      hive-2.3.4

      Description

      When hive execute jobs in parallel(control by “hive.exec.parallel” parameter), tasks submit to yarn with parallel. If the jobs completed simultaneously, then Their children task may submit more than ones.

      In our production cluster, we have a query with the stage dependencies is below:

      STAGE DEPENDENCIES:
        Stage-1 is a root stage
        Stage-2 depends on stages: Stage-1, Stage-10, Stage-14
        Stage-7 depends on stages: Stage-2 , consists of Stage-4, Stage-3, Stage-5
        Stage-4
        Stage-0 depends on stages: Stage-4, Stage-3, Stage-6
        Stage-3
        Stage-5
        Stage-6 depends on stages: Stage-5
        Stage-18 is a root stage
        Stage-9 depends on stages: Stage-18
        Stage-10 depends on stages: Stage-9
        Stage-19 is a root stage
        Stage-13 depends on stages: Stage-19
        Stage-14 depends on stages: Stage-13
      

      There is a certain probability that Stage-10 and Stage-14 will complete at the same time, then their children Stage-2 was submitted twice. As bellow log:

      2021-01-03T13:35:32,079  INFO [d207a1c7-287d-4f03-83c8-f2c42ed878a9 main] ql.Driver: Launching Job 6 out of 6
      2021-01-03T13:35:32,080  INFO [d207a1c7-287d-4f03-83c8-f2c42ed878a9 main] ql.Driver: Starting task [Stage-2:MAPRED] in parallel
      2021-01-03T13:35:32,082  INFO [d207a1c7-287d-4f03-83c8-f2c42ed878a9 main] ql.Driver: Launching Job 7 out of 6
      2021-01-03T13:35:32,083  INFO [d207a1c7-287d-4f03-83c8-f2c42ed878a9 main] ql.Driver: Starting task [Stage-2:MAPRED] in parallel
      

       

        Attachments

        1. HIVE-24577.patch
          2 kB
          hezhang

          Issue Links

            Activity

              People

              • Assignee:
                heiheizhang hezhang
                Reporter:
                gjhkael guojh
              • Votes:
                0 Vote for this issue
                Watchers:
                3 Start watching this issue

                Dates

                • Created:
                  Updated: