Spark 引擎层面 OrcBatch 代码分析

OrcColumnarBatchReader.java

 /**
   * Return true if there exists more data in the next batch. If exists, prepare the next batch
   * by copying from ORC VectorizedRowBatch columns to Spark ColumnarBatch columns.
   */
  private boolean nextBatch() throws IOException {
    recordReader.nextBatch(batch);
    int batchSize = batch.size;
    if (batchSize == 0) {
      return false;
    }
    columnarBatch.setNumRows(batchSize);

    if (!copyToSpark) {
      for (int i = 0; i < requiredFields.length; i++) {
        if (requestedColIds[i] != -1) {
        //是存放的引用，所以设置一下大小即可
          ((OrcColumnVector) orcVectorWrappers[i]).setBatchSize(batchSize);
        }
      }
      return true;
    }

    for (WritableColumnVector vector : columnVectors) {
      vector.reset();
    }

    for (int i = 0; i < requiredFields.length; i++) {
      StructField field = requiredFields[i];
      WritableColumnVector toColumn = columnVectors[i];

      if (requestedColIds[i] >= 0) {
      // 是获取的所有行的数据，然后按需
        ColumnVector fromColumn = batch.cols[requestedColIds[i]];

        if (fromColumn.isRepeating) {
          putRepeatingValues(batchSize, field, fromColumn, toColumn);
        } else if (fromColumn.noNulls) {
          putNonNullValues(batchSize, field, fromColumn, toColumn);
        } else {
          putValues(batchSize, field, fromColumn, toColumn);
        }
      }
    }
    return true;
  }

orc.enableVectorizedReader

// SQLConf.scala

  val ORC_VECTORIZED_READER_ENABLED = buildConf("spark.sql.orc.enableVectorizedReader")
    .doc("Enables vectorized orc decoding.")
    .booleanConf
    .createWithDefault(true)