Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-5899

Viewing specific stage information which contains thousands of tasks will freak out the driver and CPU cores from where it runs

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Resolved
    • Critical
    • Resolution: Duplicate
    • 1.2.1, 1.3.0
    • None
    • Web UI
    • None
    • CentOS 7, Spark Standalone

    Description

      If a user ever tries to view specific stage stats, for example, a repartition stage which involves 30000 partitions, the Web UI attempts to load every single task result onto a single webpage, which is completely destroying CPU usage on the driver, which subsequently causes the remaining tasks / jobs to be nearly impossible to complete.

      Ideally the task results should be paged (if not too much trouble) to prevent this from happening.

      Attachments

        Issue Links

          Activity

            People

              Unassigned Unassigned
              mkman84 Mark Khaitman
              Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: