Uploaded image for project: 'Lucene - Core'
  1. Lucene - Core
  2. LUCENE-9378

Configurable compression for BinaryDocValues

    XMLWordPrintableJSON

    Details

    • Type: Improvement
    • Status: Resolved
    • Priority: Major
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: 8.8
    • Component/s: None
    • Labels:
      None
    • Lucene Fields:
      New

      Description

      Lucene 8.5.1 includes a change to always compress BinaryDocValues. This caused (~30%) reduction in our red-line QPS (throughput). 

      We think users should be given some way to opt-in for this compression feature instead of always being enabled which can have a substantial query time cost as we saw during our upgrade. Michael McCandless suggested one possible approach by introducing a mode in Lucene80DocValuesFormat (COMPRESSED and UNCOMPRESSED) and allowing users to create a custom Codec subclassing the default Codec and pick the format they want.

      Idea is similar to Lucene50StoredFieldsFormat which has two modes, Mode.BEST_SPEED and Mode.BEST_COMPRESSION.

      Here's related issues for adding benchmark covering BINARY doc values query-time performance - https://github.com/mikemccand/luceneutil/issues/61

        Attachments

        1. image-2020-06-12-22-17-30-339.png
          14 kB
          Alex Klibisz
        2. image-2020-06-12-22-17-53-961.png
          14 kB
          Alex Klibisz
        3. image-2020-06-12-22-18-24-527.png
          14 kB
          Alex Klibisz
        4. image-2020-06-12-22-18-48-919.png
          14 kB
          Alex Klibisz
        5. snapshot-v77x.nps
          32 kB
          Alex Klibisz
        6. snapshots-v76x.nps
          28 kB
          Alex Klibisz
        7. hotspots-v77x.png
          136 kB
          Alex Klibisz
        8. hotspots-v76x.png
          129 kB
          Alex Klibisz
        9. snapshot-v77x.nps
          32 kB
          Alex Klibisz
        10. snapshots-v76x.nps
          28 kB
          Alex Klibisz
        11. hotspots-v77x.png
          136 kB
          Alex Klibisz
        12. hotspots-v76x.png
          129 kB
          Alex Klibisz
        13. snapshots-v76x.nps
          28 kB
          Alex Klibisz
        14. snapshot-v77x.nps
          32 kB
          Alex Klibisz
        15. hotspots-v76x.png
          129 kB
          Alex Klibisz
        16. hotspots-v77x.png
          136 kB
          Alex Klibisz
        17. hotspots-v76x.png
          129 kB
          Alex Klibisz
        18. hotspots-v76x.png
          129 kB
          Alex Klibisz
        19. hotspots-v77x.png
          136 kB
          Alex Klibisz

          Issue Links

            Activity

              People

              • Assignee:
                Unassigned
                Reporter:
                viral.gandhi Viral Gandhi
              • Votes:
                4 Vote for this issue
                Watchers:
                12 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved:

                  Time Tracking

                  Estimated:
                  Original Estimate - Not Specified
                  Not Specified
                  Remaining:
                  Remaining Estimate - 0h
                  0h
                  Logged:
                  Time Spent - 5h 40m
                  5h 40m