Details

    • Type: Improvement Improvement
    • Status: Open
    • Priority: Major Major
    • Resolution: Unresolved
    • Affects Version/s: 0.94.14
    • Fix Version/s: None
    • Component/s: Region Assignment
    • Labels:
      None

      Description

      During master initializate, if there are regionservers shutdown improperly previous or start up failed, master will think this is a failover and assign regions one by one.
      We have a cluster with 200+ nodes and 32000+regions, it takes 50mins to start up all.
      The patch is based on 0.94 branch, and trunk is different now.

      1. HBASE-10209-94-2.patch
        6 kB
        binlijin
      2. HBASE-10209-94.patch
        5 kB
        binlijin

        Activity

        Hide
        chunhui shen added a comment -

        The improvement makes sense,

        but is there any risk on concurrent assignment,
        e.g. Server Shutdown event happen in the same time when assigning.

        Wait more body to take a see since I can't think the scenarios about assignment clearly.

        Show
        chunhui shen added a comment - The improvement makes sense, but is there any risk on concurrent assignment, e.g. Server Shutdown event happen in the same time when assigning. Wait more body to take a see since I can't think the scenarios about assignment clearly.
        binlijin made changes -
        Component/s Region Assignment [ 12319661 ]
        binlijin made changes -
        Affects Version/s 0.94.14 [ 12325047 ]
        binlijin made changes -
        Attachment HBASE-10209-94-2.patch [ 12619727 ]
        binlijin made changes -
        Description During master initializate, if there are regionservers shutdown improperly previous or start up failed, master will think this is a failover and assign regions one by one.
        We have a cluster with 200+ nodes and 32000+regions, it takes 50mins to start up all.
        During master initializate, if there are regionservers shutdown improperly previous or start up failed, master will think this is a failover and assign regions one by one.
        We have a cluster with 200+ nodes and 32000+regions, it takes 50mins to start up all.
        The patch is based on 0.94 branch, and trunk is different now.
        binlijin made changes -
        Field Original Value New Value
        Attachment HBASE-10209-94.patch [ 12619515 ]
        binlijin created issue -

          People

          • Assignee:
            Unassigned
            Reporter:
            binlijin
          • Votes:
            0 Vote for this issue
            Watchers:
            5 Start watching this issue

            Dates

            • Created:
              Updated:

              Development