Uploaded image for project: 'IMPALA'
  1. IMPALA
  2. IMPALA-11810

The mem estimate is incorrect in HdfsScanNode

    XMLWordPrintableJSON

Details

    • Improvement
    • Status: Open
    • Major
    • Resolution: Unresolved
    • Impala 4.2.0, Impala 4.1.1
    • None
    • fe
    • None
    • ghx-label-12

    Description

      About perInstanceMemEstimate calculate in the below method of the HdfsScanNode class:

      @Override  
      public void computeNodeResourceProfile(TQueryOptions queryOptions) {
      ...
          long avgScanRangeBytes = (long) Math.ceil(sumValues(totalBytesPerFs_) / (double) scanRangeSize);
      ...
      }

      All table data file sizes are used in the calculation of scan hdfs memory consumption, it is unreasonable. Because not all data in the table needs to be read into the memory, but only the materialized fields are read, so only the data file size occupied by the materialized field needs to be calculated.

      Attachments

        1. after_used_materialized.png
          114 kB
          jhkcool
        2. current_totalBytes.png
          238 kB
          jhkcool

        Activity

          People

            Unassigned Unassigned
            jhkcool jhkcool
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

            Dates

              Created:
              Updated: