Uploaded image for project: 'Avro'
  1. Avro
  2. AVRO-985

Remove byte-by-byte copying in RecordBuilderBase.defaultValue for non-complex types

    Details

    • Type: Improvement
    • Status: Open
    • Priority: Major
    • Resolution: Unresolved
    • Affects Version/s: 1.6.1
    • Fix Version/s: None
    • Component/s: java
    • Labels:
      None

      Description

      In one section of RecordBuilderBase.defaultValue(Field) (quoted below) a bytewise copy of the default object is created based on the JSON value provided. However, this is an extremely inefficient operation and causes large slowdowns when building large object sets, including latency spikes when the binary encoder flushes.

      A simple workaround for a majority of cases would be to have a separate code path for "primitives" (fixed, string, boolean, int, double, enum, float, bytes) that allows direct creation rather than a full bytewise copy (and subsequent deep copy).

      RecordBuilderBase.java:

          // If not cached, get the default Java value by encoding the default JSON
          // value and then decoding it:
          if (defaultValue == null) {
            ByteArrayOutputStream baos = new ByteArrayOutputStream();
            encoder = EncoderFactory.get().binaryEncoder(baos, encoder);
            ResolvingGrammarGenerator.encode(
                encoder, field.schema(), defaultJsonValue);
            encoder.flush();
            decoder = DecoderFactory.get().binaryDecoder(
                baos.toByteArray(), decoder);
            defaultValue = new GenericDatumReader(
                field.schema()).read(null, decoder);
            defaultSchemaValues.putIfAbsent(field.pos(), defaultValue);
          }
      

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                Unassigned
                Reporter:
                dkaminsky Douglas Kaminsky
              • Votes:
                0 Vote for this issue
                Watchers:
                1 Start watching this issue

                Dates

                • Created:
                  Updated: