org.apache.arrow.vector.VectorSchemaRoot Java Examples

/**
 * Read next.
 */
public byte[] nextToBytes() throws IOException {
  ByteArrayOutputStream out = new ByteArrayOutputStream();
  VectorSchemaRoot schemaRoot = nextToSchemaRoot();
  ArrowFileWriter writer = new ArrowFileWriter( schemaRoot, null, Channels.newChannel( out ) );
  writer.start();
  writer.writeBatch();
  writer.end();
  writer.close();
  return out.toByteArray();
}
 

private Iterator<InternalRow> toArrowRows(VectorSchemaRoot root, List<String> namesInOrder) {
    ColumnVector[] columns = namesInOrder.stream()
            .map(name -> root.getVector(name))
            .map(vector -> new ArrowSchemaConverter(vector))
            .collect(Collectors.toList()).toArray(new ColumnVector[0]);

    ColumnarBatch batch = new ColumnarBatch(columns);
    batch.setNumRows(root.getRowCount());
    return batch.rowIterator();
}
 

/**
 * Creates the vector schema from incoming container and referenced fields.
 * @param input
 * @param referencedFields
 * @return the vector schema root.
 */
public static VectorSchemaRoot getSchemaRoot(VectorAccessible input, Set referencedFields) {
  List<FieldVector> fv = ImmutableList.copyOf(input)
    .stream()
    .map(vw -> ((FieldVector)vw.getValueVector()))
    .filter(fVec -> referencedFields.contains(fVec.getField()))
    .collect(Collectors.toList());

  List<Field> fields = fv.stream()
    .map(fieldVec -> fieldVec.getField())
    .collect(Collectors.toList());

  Schema schemaWithOnlyReferencedFields = new Schema(fields);
  VectorSchemaRoot root = new VectorSchemaRoot(
    schemaWithOnlyReferencedFields,
    fv,
    0
  );
  return root;
}
 

/**
 * Append from arrow byte array.
 */
public void append( final byte[] buffer ) throws IOException {
  ArrowFileReader arrowReader = new ArrowFileReader(
      new SeekableInMemoryByteChannel( buffer ) , new RootAllocator( Integer.MAX_VALUE ) );
  List<ArrowBlock> blockList = arrowReader.getRecordBlocks();
  for ( ArrowBlock block : blockList ) {
    VectorSchemaRoot root = arrowReader.getVectorSchemaRoot();
    arrowReader.loadRecordBatch(block);
    append( root );
  }
}
 

@Override
@SuppressWarnings("ConstantConditions")
public void emitResults() throws IOException {
	byte[] udfResult;
	while ((udfResult = userDefinedFunctionResultQueue.poll()) != null) {
		bais.setBuffer(udfResult, 0, udfResult.length);
		reader.loadNextBatch();
		VectorSchemaRoot root = reader.getVectorSchemaRoot();
		if (arrowReader == null) {
			arrowReader = ArrowUtils.createRowDataArrowReader(root, outputType);
		}
		for (int i = 0; i < root.getRowCount(); i++) {
			RowData input = forwardedInputQueue.poll();
			reuseJoinedRow.setRowKind(input.getRowKind());
			rowDataWrapper.collect(reuseJoinedRow.replace(input, arrowReader.read(i)));
		}
	}
}
 

@Override
@SuppressWarnings("ConstantConditions")
public void emitResults() throws IOException {
	byte[] udfResult;
	while ((udfResult = userDefinedFunctionResultQueue.poll()) != null) {
		bais.setBuffer(udfResult, 0, udfResult.length);
		reader.loadNextBatch();
		VectorSchemaRoot root = reader.getVectorSchemaRoot();
		if (arrowReader == null) {
			arrowReader = ArrowUtils.createRowArrowReader(root, outputType);
		}
		for (int i = 0; i < root.getRowCount(); i++) {
			CRow input = forwardedInputQueue.poll();
			cRowWrapper.setChange(input.change());
			cRowWrapper.collect(Row.join(input.row(), arrowReader.read(i)));
		}
	}
}
 

@Test
public void testCreateRowArrowReader() {
	VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
	RowArrowReader reader = ArrowUtils.createRowArrowReader(root, rowType);
	ArrowFieldReader[] fieldReaders = reader.getFieldReaders();
	for (int i = 0; i < fieldReaders.length; i++) {
		assertEquals(testFields.get(i).f5, fieldReaders[i].getClass());
	}
}
 

private File createArrowFile(String fileName, Schema schema, Object[][] data,
                             int rowsPerRecordBatch)
throws IOException
{
  File file = resultFolder.newFile(fileName);
  VectorSchemaRoot root = VectorSchemaRoot.create(schema, allocator);

  try (ArrowWriter writer = new ArrowStreamWriter(
      root, new DictionaryProvider.MapDictionaryProvider(),
      new FileOutputStream(file)))
  {
    writer.start();

    for (int i = 0; i < data[0].length; )
    {
      int rowsToAppend = Math.min(rowsPerRecordBatch, data[0].length - i);
      root.setRowCount(rowsToAppend);

      for (int j = 0; j < data.length; j++)
      {
        FieldVector vector = root.getFieldVectors().get(j);

        switch (vector.getMinorType())
        {
          case INT:
            writeIntToField(vector, data[j], i, rowsToAppend);
            break;
        }
      }

      writer.writeBatch();
      i += rowsToAppend;
    }
  }

  return file;
}
 

/**
 * Creates an {@link ArrowWriter} for the specified {@link VectorSchemaRoot}.
 */
public static ArrowWriter<Row> createRowArrowWriter(VectorSchemaRoot root, RowType rowType) {
	ArrowFieldWriter<Row>[] fieldWriters = new ArrowFieldWriter[root.getFieldVectors().size()];
	List<FieldVector> vectors = root.getFieldVectors();
	for (int i = 0; i < vectors.size(); i++) {
		FieldVector vector = vectors.get(i);
		vector.allocateNew();
		fieldWriters[i] = createRowArrowFieldWriter(vector, rowType.getTypeAt(i));
	}

	return new ArrowWriter<>(root, fieldWriters);
}
 

/**
 * Creates an {@link ArrowWriter} for blink planner for the specified {@link VectorSchemaRoot}.
 */
public static ArrowWriter<RowData> createRowDataArrowWriter(VectorSchemaRoot root, RowType rowType) {
	ArrowFieldWriter<RowData>[] fieldWriters = new ArrowFieldWriter[root.getFieldVectors().size()];
	List<FieldVector> vectors = root.getFieldVectors();
	for (int i = 0; i < vectors.size(); i++) {
		FieldVector vector = vectors.get(i);
		vector.allocateNew();
		fieldWriters[i] = createArrowFieldWriterForRow(vector, rowType.getTypeAt(i));
	}

	return new ArrowWriter<>(root, fieldWriters);
}
 

/**
 * Creates an {@link ArrowReader} for the specified {@link VectorSchemaRoot}.
 */
public static RowArrowReader createRowArrowReader(VectorSchemaRoot root, RowType rowType) {
	List<ArrowFieldReader> fieldReaders = new ArrayList<>();
	List<FieldVector> fieldVectors = root.getFieldVectors();
	for (int i = 0; i < fieldVectors.size(); i++) {
		fieldReaders.add(createRowArrowFieldReader(fieldVectors.get(i), rowType.getTypeAt(i)));
	}

	return new RowArrowReader(fieldReaders.toArray(new ArrowFieldReader[0]));
}
 

/**
 * Creates an {@link ArrowReader} for blink planner for the specified {@link VectorSchemaRoot}.
 */
public static RowDataArrowReader createRowDataArrowReader(VectorSchemaRoot root, RowType rowType) {
	List<ColumnVector> columnVectors = new ArrayList<>();
	List<FieldVector> fieldVectors = root.getFieldVectors();
	for (int i = 0; i < fieldVectors.size(); i++) {
		columnVectors.add(createColumnVector(fieldVectors.get(i), rowType.getTypeAt(i)));
	}

	return new RowDataArrowReader(columnVectors.toArray(new ColumnVector[0]));
}
 

@Override
public void emitResults() throws IOException {
	byte[] udfResult;
	while ((udfResult = userDefinedFunctionResultQueue.poll()) != null) {
		bais.setBuffer(udfResult, 0, udfResult.length);
		reader.loadNextBatch();
		VectorSchemaRoot root = reader.getVectorSchemaRoot();
		if (arrowReader == null) {
			arrowReader = ArrowUtils.createRowArrowReader(root, userDefinedFunctionOutputType);
		}
		for (int i = 0; i < root.getRowCount(); i++) {
			resultCollector.collect(Row.join(forwardedInputQueue.poll(), arrowReader.read(i)));
		}
	}
}
 

@Override
public void open() throws Exception {
	super.open();
	allocator = ArrowUtils.getRootAllocator().newChildAllocator("writer", 0, Long.MAX_VALUE);
	root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(getInputType()), allocator);
	arrowWriter = createArrowWriter();
	arrowStreamWriter = new ArrowStreamWriter(root, null, baos);
	arrowStreamWriter.start();
	currentBatchCount = 0;
}
 

public MockLlapArrowBatchRecordReader(long arrowAllocatorMax) {
  BufferAllocator allocator = RootAllocatorFactory.INSTANCE.getOrCreateRootAllocator(arrowAllocatorMax);
  IntVector vector = new IntVector("a", allocator);
  vector.allocateNewSafe();
  for(int i = 0; i < testVector.length; i++) {
    vector.set(i, testVector[i]);
  }
  vector.setValueCount(testVector.length);
  List<Field> fields = Lists.newArrayList(vector.getField());
  List<FieldVector> vectors = new ArrayList<>();
  vectors.add(vector);
  vectorSchemaRoot = new VectorSchemaRoot(fields, vectors, testVector.length);
}
 

@Test
public void testCreateRowDataArrowReader() {
	VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
	RowDataArrowReader reader = ArrowUtils.createRowDataArrowReader(root, rowType);
	ColumnVector[] columnVectors = reader.getColumnVectors();
	for (int i = 0; i < columnVectors.length; i++) {
		assertEquals(testFields.get(i).f6, columnVectors[i].getClass());
	}
}
 

@Test
public void testCreateRowArrowWriter() {
	VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
	ArrowWriter<Row> writer = ArrowUtils.createRowArrowWriter(root, rowType);
	ArrowFieldWriter<Row>[] fieldWriters = writer.getFieldWriters();
	for (int i = 0; i < fieldWriters.length; i++) {
		assertEquals(testFields.get(i).f3, fieldWriters[i].getClass());
	}
}
 

@Test
public void testCreateRowDataArrowWriter() {
	VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
	ArrowWriter<RowData> writer = ArrowUtils.createRowDataArrowWriter(root, rowType);
	ArrowFieldWriter<RowData>[] fieldWriters = writer.getFieldWriters();
	for (int i = 0; i < fieldWriters.length; i++) {
		assertEquals(testFields.get(i).f4, fieldWriters[i].getClass());
	}
}
 

@Test
public void testReadArrowBatches() throws IOException {
	VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
	ArrowWriter<RowData> arrowWriter = ArrowUtils.createRowDataArrowWriter(root, rowType);
	ByteArrayOutputStream baos = new ByteArrayOutputStream();
	ArrowStreamWriter arrowStreamWriter = new ArrowStreamWriter(root, null, baos);
	arrowStreamWriter.start();

	List<RowData> testData = Arrays.asList(
		new GenericRowData(rowType.getFieldCount()),
		new GenericRowData(rowType.getFieldCount()),
		new GenericRowData(rowType.getFieldCount()),
		new GenericRowData(rowType.getFieldCount()),
		new GenericRowData(rowType.getFieldCount()));
	int batches = 3;
	List<List<RowData>> subLists = Lists.partition(testData, testData.size() / batches + 1);
	for (List<RowData> subList : subLists) {
		for (RowData value : subList) {
			arrowWriter.write(value);
		}
		arrowWriter.finish();
		arrowStreamWriter.writeBatch();
		arrowWriter.reset();
	}

	assertEquals(batches,
		ArrowUtils.readArrowBatches(Channels.newChannel(new ByteArrayInputStream(baos.toByteArray()))).length);
}
 

@Override
public Tuple2<ArrowWriter<Row>, ArrowStreamWriter> createArrowWriter(OutputStream outputStream) throws IOException {
	VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
	ArrowWriter<Row> arrowWriter = ArrowUtils.createRowArrowWriter(root, rowType);
	ArrowStreamWriter arrowStreamWriter = new ArrowStreamWriter(root, null, outputStream);
	arrowStreamWriter.start();
	return Tuple2.of(arrowWriter, arrowStreamWriter);
}
 

@Override
public Tuple2<ArrowWriter<RowData>, ArrowStreamWriter> createArrowWriter(OutputStream outputStream) throws IOException {
	VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
	ArrowWriter<RowData> arrowWriter = ArrowUtils.createRowDataArrowWriter(root, rowType);
	ArrowStreamWriter arrowStreamWriter = new ArrowStreamWriter(root, null, outputStream);
	arrowStreamWriter.start();
	return Tuple2.of(arrowWriter, arrowStreamWriter);
}
 

public ArrowSourceFunctionTest() {
	super(VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator),
		serializer,
		Comparator.comparing(o -> o.getString(0)),
		new DeeplyEqualsChecker()
			.withCustomCheck(
				(o1, o2) -> o1 instanceof RowData && o2 instanceof RowData,
				(o1, o2, checker) -> deepEqualsBaseRow(
					(RowData) o1,
					(RowData) o2,
					(RowDataSerializer) serializer.duplicate(),
					(RowDataSerializer) serializer.duplicate())));
}
 

@Test
public void testReadSchemaAndRecordsFromByteArray() throws Exception {
    BufferAllocator allocator = new RootAllocator(Long.MAX_VALUE);

    int valueCount = 3;
    List<Field> fields = new ArrayList<>();
    fields.add(ArrowConverter.field("field1",new ArrowType.FloatingPoint(FloatingPointPrecision.SINGLE)));
    fields.add(ArrowConverter.intField("field2"));

    List<FieldVector> fieldVectors = new ArrayList<>();
    fieldVectors.add(ArrowConverter.vectorFor(allocator,"field1",new float[] {1,2,3}));
    fieldVectors.add(ArrowConverter.vectorFor(allocator,"field2",new int[] {1,2,3}));


    org.apache.arrow.vector.types.pojo.Schema schema = new org.apache.arrow.vector.types.pojo.Schema(fields);

    VectorSchemaRoot schemaRoot1 = new VectorSchemaRoot(schema, fieldVectors, valueCount);
    VectorUnloader vectorUnloader = new VectorUnloader(schemaRoot1);
    vectorUnloader.getRecordBatch();
    ByteArrayOutputStream byteArrayOutputStream = new ByteArrayOutputStream();
    try(ArrowFileWriter arrowFileWriter = new ArrowFileWriter(schemaRoot1,null,newChannel(byteArrayOutputStream))) {
        arrowFileWriter.writeBatch();
    } catch (IOException e) {
        log.error("",e);
    }

    byte[] arr = byteArrayOutputStream.toByteArray();
    val arr2 = ArrowConverter.readFromBytes(arr);
    assertEquals(2,arr2.getFirst().numColumns());
    assertEquals(3,arr2.getRight().size());

    val arrowCols = ArrowConverter.toArrowColumns(allocator,arr2.getFirst(),arr2.getRight());
    assertEquals(2,arrowCols.size());
    assertEquals(valueCount,arrowCols.get(0).getValueCount());
}
 

@Test
public void testReadSchemaAndRecordsFromByteArray() throws Exception {
    BufferAllocator allocator = new RootAllocator(Long.MAX_VALUE);

    int valueCount = 3;
    List<Field> fields = new ArrayList<>();
    fields.add(ArrowConverter.field("field1",new ArrowType.FloatingPoint(FloatingPointPrecision.SINGLE)));
    fields.add(ArrowConverter.intField("field2"));

    List<FieldVector> fieldVectors = new ArrayList<>();
    fieldVectors.add(ArrowConverter.vectorFor(allocator,"field1",new float[] {1,2,3}));
    fieldVectors.add(ArrowConverter.vectorFor(allocator,"field2",new int[] {1,2,3}));


    org.apache.arrow.vector.types.pojo.Schema schema = new org.apache.arrow.vector.types.pojo.Schema(fields);

    VectorSchemaRoot schemaRoot1 = new VectorSchemaRoot(schema, fieldVectors, valueCount);
    VectorUnloader vectorUnloader = new VectorUnloader(schemaRoot1);
    vectorUnloader.getRecordBatch();
    ByteArrayOutputStream byteArrayOutputStream = new ByteArrayOutputStream();
    try(ArrowFileWriter arrowFileWriter = new ArrowFileWriter(schemaRoot1,null,newChannel(byteArrayOutputStream))) {
        arrowFileWriter.writeBatch();
    } catch (IOException e) {
        e.printStackTrace();
    }

    byte[] arr = byteArrayOutputStream.toByteArray();
    val arr2 = ArrowConverter.readFromBytes(arr);
    assertEquals(2,arr2.getFirst().numColumns());
    assertEquals(3,arr2.getRight().size());

    val arrowCols = ArrowConverter.toArrowColumns(allocator,arr2.getFirst(),arr2.getRight());
    assertEquals(2,arrowCols.size());
    assertEquals(valueCount,arrowCols.get(0).getValueCount());
}
 

public static int run( final String[] args ) throws IOException{
  CommandLine cl;
  try{
    CommandLineParser clParser = new GnuParser();
    cl = clParser.parse( createOptions( args ) , args );
  }catch( ParseException e ){
    printHelp( args );
    throw new IOException( e );
  }

  if( cl.hasOption( "help" ) ){
    printHelp( args );
    return 0;
  }

  String input = cl.getOptionValue( "input" , null );
  String output = cl.getOptionValue( "output" , null );

  Configuration config = new Configuration();

  ArrowFileReader arrowReader = new ArrowFileReader( new FileInputStream( input ).getChannel() , new RootAllocator( Integer.MAX_VALUE ) );
  OutputStream out = FileUtil.create( output );
  MDSWriter writer = new MDSWriter( out , config );
  List<ArrowBlock> blockList = arrowReader.getRecordBlocks();
  for( ArrowBlock block : blockList ){
    VectorSchemaRoot root = arrowReader.getVectorSchemaRoot();
    arrowReader.loadRecordBatch(block);
    List<FieldVector> fieldVectorList = root.getFieldVectors();
    Spread spread = ArrowSpreadUtil.toSpread( root.getRowCount() , fieldVectorList );
    writer.append( spread );
  }
  arrowReader.close();
  writer.close();

  return 0;
}
 

public boolean setUpTwister2ArrowWrite(int workerId) throws Exception {
  LOG.fine("%%%%%%%%% worker id details:" + workerId + "\t" + arrowFile);
  this.root = VectorSchemaRoot.create(Schema.fromJSON(arrowSchema), this.rootAllocator);
  Path path = new Path(arrowFile);
  this.fileSystem = FileSystemUtils.get(path);
  this.fsDataOutputStream = fileSystem.create(path);
  this.twister2ArrowOutputStream = new Twister2ArrowOutputStream(this.fsDataOutputStream);
  DictionaryProvider.MapDictionaryProvider provider
      = new DictionaryProvider.MapDictionaryProvider();
  if (!flag) {
    this.arrowFileWriter = new ArrowFileWriter(root, provider,
        this.fsDataOutputStream.getChannel());
  } else {
    this.arrowFileWriter = new ArrowFileWriter(root, provider, this.twister2ArrowOutputStream);
  }

  LOG.info("root schema fields:" + root.getSchema().getFields());
  for (Field field : root.getSchema().getFields()) {
    FieldVector vector = root.getVector(field.getName());
    if (vector.getMinorType().equals(Types.MinorType.INT)) {
      this.generatorMap.put(vector, new IntVectorGenerator());
    } else if (vector.getMinorType().equals(Types.MinorType.BIGINT)) {
      this.generatorMap.put(vector, new BigIntVectorGenerator());
    } else if (vector.getMinorType().equals(Types.MinorType.FLOAT4)) {
      this.generatorMap.put(vector, new FloatVectorGenerator());
    } else {
      throw new RuntimeException("unsupported arrow write type");
    }
  }
  return true;
}
 

public static void transferFromRoot(VectorSchemaRoot root, VectorContainer container, BufferAllocator allocator) {
  container.clear();
  // iterate over and transfer columns
  for (FieldVector fv : root.getFieldVectors()) {
    final TransferPair tp = fv.getTransferPair(allocator);
    tp.transfer();
    container.add(tp.getTo());
  }

  container.setRecordCount(root.getRowCount());
  container.addSchema(root.getSchema());
  container.buildSchema();
}
 

public static VectorSchemaRoot getVectorSchemaRoot(final VectorAccessible batch) {
  List<FieldVector> fieldVectors = FluentIterable.from(batch)
      .transform(new Function<VectorWrapper<?>, FieldVector>() {
        @Override
        public FieldVector apply(VectorWrapper<?> vectorWrapper) {
          return (FieldVector)vectorWrapper.getValueVector();
        }
      }).toList();
  int rowCount = batch.getRecordCount();
  List<Field> fields = batch.getSchema().getFields();
  VectorSchemaRoot root = new VectorSchemaRoot(fields , fieldVectors, rowCount);
  return root;
}
 

@Override
public void setup(VectorAccessible incoming, OutputEntryListener listener, WriteStatsListener statsListener) throws IOException {
  root = new VectorSchemaRoot(ImmutableList.copyOf(incoming)
    .stream()
    .map(vw -> ((FieldVector) vw.getValueVector()))
    .collect(Collectors.toList()));
  unloader = new VectorUnloader(root);
  creator = store.putStream(
    descriptor, root.getSchema());
}
 

public void start(ServerStreamListener listener) throws InterruptedException {
  logger.debug("trying to start, waiting for schema for {}", descriptor);
  countDownLatch.await();
  if (root == null) {
    logger.warn("root was not set for {}, not starting listener properly", descriptor);
    root = VectorSchemaRoot.create(new Schema(ImmutableList.of()), allocator);
  }
  listener.start(root);
  this.listener = listener;
}