Uploaded image for project: 'IMPALA'
  1. IMPALA
  2. IMPALA-8522

Consider being smarter about deleting insert staging directories upon errors

    XMLWordPrintableJSON

    Details

    • Type: Improvement
    • Status: Open
    • Priority: Minor
    • Resolution: Unresolved
    • Affects Version/s: Impala 2.11.0, Impala 3.0, Impala 2.12.0, Impala 3.1.0, Impala 3.2.0
    • Fix Version/s: None
    • Component/s: Distributed Exec
    • Labels:
      None
    • Epic Color:
      ghx-label-5

      Description

      My investigation into IMPALA-7176 showed that someones hdfsClose() can be very slow if the file was deleted out from under the client.

      Impala actually does this on the error cleanup path for inserts - Coordinator::FinalizeHdfsInsert() will delete the whole staging directory while fragments may still be running. This isn't a bad fallback in case a fragment goes missing (e.g. a node crash) or gets stuck permanently, but leads to unnecessary errors otherwise. It seems like we could wait a few seconds if the fragments are still running to give the fragments a chance to clean up after themselves.

      Maybe this isn't worth the hassle or it could be fixed on the HDFS side - I filed HDFS-14479

        Attachments

          Activity

            People

            • Assignee:
              Unassigned
              Reporter:
              tarmstrong Tim Armstrong
            • Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

              Dates

              • Created:
                Updated: