Examples with BufferAllocator org.apache.arrow.memory.BufferAllocator used on opensource projects

Example 1 with BufferAllocator

use of org.apache.arrow.memory.BufferAllocator in project flink by apache.

the class ArrowUtils method collectAsPandasDataFrame.

/**
 * Convert Flink table to Pandas DataFrame.
 */
public static CustomIterator<byte[]> collectAsPandasDataFrame(Table table, int maxArrowBatchSize) throws Exception {
    checkArrowUsable();
    BufferAllocator allocator = getRootAllocator().newChildAllocator("collectAsPandasDataFrame", 0, Long.MAX_VALUE);
    RowType rowType = (RowType) table.getResolvedSchema().toSourceRowDataType().getLogicalType();
    DataType defaultRowDataType = TypeConversions.fromLogicalToDataType(rowType);
    VectorSchemaRoot root = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(rowType), allocator);
    ByteArrayOutputStream baos = new ByteArrayOutputStream();
    ArrowStreamWriter arrowStreamWriter = new ArrowStreamWriter(root, null, baos);
    arrowStreamWriter.start();
    Iterator<Row> results = table.execute().collect();
    Iterator<Row> appendOnlyResults;
    if (isAppendOnlyTable(table)) {
        appendOnlyResults = results;
    } else {
        appendOnlyResults = filterOutRetractRows(results);
    }
    ArrowWriter arrowWriter = createRowDataArrowWriter(root, rowType);
    Iterator convertedResults = new Iterator<RowData>() {

        @Override
        public boolean hasNext() {
            return appendOnlyResults.hasNext();
        }

        @Override
        public RowData next() {
            DataFormatConverters.DataFormatConverter converter = DataFormatConverters.getConverterForDataType(defaultRowDataType);
            return (RowData) converter.toInternal(appendOnlyResults.next());
        }
    };
    return new CustomIterator<byte[]>() {

        @Override
        public boolean hasNext() {
            return convertedResults.hasNext();
        }

        @Override
        public byte[] next() {
            try {
                int i = 0;
                while (convertedResults.hasNext() && i < maxArrowBatchSize) {
                    i++;
                    arrowWriter.write(convertedResults.next());
                }
                arrowWriter.finish();
                arrowStreamWriter.writeBatch();
                return baos.toByteArray();
            } catch (Throwable t) {
                String msg = "Failed to serialize the data of the table";
                LOG.error(msg, t);
                throw new RuntimeException(msg, t);
            } finally {
                arrowWriter.reset();
                baos.reset();
                if (!hasNext()) {
                    root.close();
                    allocator.close();
                }
            }
        }
    };
}

Example 2 with BufferAllocator

use of org.apache.arrow.memory.BufferAllocator in project hive by apache.

the class TestJdbcWithMiniLlapVectorArrowBatch method runQueryUsingLlapArrowBatchReader.

private MultiSet<List<Object>> runQueryUsingLlapArrowBatchReader(String query, Map<String, String> extraHiveConfs) throws Exception {
    String url = miniHS2.getJdbcURL();
    if (extraHiveConfs != null) {
        url = url + "?" + extraHiveConfs.entrySet().stream().map(e -> e.getKey() + "=" + e.getValue()).collect(Collectors.joining(";"));
    }
    String user = System.getProperty("user.name");
    String pwd = user;
    String handleId = UUID.randomUUID().toString();
    // Get splits
    JobConf job = new JobConf(conf);
    job.set(LlapBaseInputFormat.URL_KEY, url);
    job.set(LlapBaseInputFormat.USER_KEY, user);
    job.set(LlapBaseInputFormat.PWD_KEY, pwd);
    job.set(LlapBaseInputFormat.QUERY_KEY, query);
    job.set(LlapBaseInputFormat.HANDLE_ID, handleId);
    job.set(LlapBaseInputFormat.USE_NEW_SPLIT_FORMAT, "false");
    BufferAllocator allocator = RootAllocatorFactory.INSTANCE.getOrCreateRootAllocator(Long.MAX_VALUE).newChildAllocator(UUID.randomUUID().toString(), 0, Long.MAX_VALUE);
    LlapBaseInputFormat llapBaseInputFormat = new LlapBaseInputFormat(true, allocator);
    InputSplit[] splits = llapBaseInputFormat.getSplits(job, 1);
    assertTrue(splits.length > 0);
    MultiSet<List<Object>> queryResult = new HashMultiSet<>();
    for (InputSplit split : splits) {
        System.out.println("Processing split " + Arrays.toString(split.getLocations()));
        RecordReader<NullWritable, ArrowWrapperWritable> reader = llapBaseInputFormat.getRecordReader(split, job, null);
        ArrowWrapperWritable wrapperWritable = new ArrowWrapperWritable();
        while (reader.next(NullWritable.get(), wrapperWritable)) {
            queryResult.addAll(collectResultFromArrowVector(wrapperWritable));
        }
        reader.close();
    }
    LlapBaseInputFormat.close(handleId);
    return queryResult;
}

Example 3 with BufferAllocator

use of org.apache.arrow.memory.BufferAllocator in project carbondata by apache.

the class ArrowCarbonReaderTest method testArrowReader.

@Test
public void testArrowReader() {
    String path = "./carbondata";
    try {
        FileUtils.deleteDirectory(new File(path));
        Field[] fields = new Field[13];
        fields[0] = new Field("stringField", DataTypes.STRING);
        fields[1] = new Field("shortField", DataTypes.SHORT);
        fields[2] = new Field("intField", DataTypes.INT);
        fields[3] = new Field("longField", DataTypes.LONG);
        fields[4] = new Field("doubleField", DataTypes.DOUBLE);
        fields[5] = new Field("boolField", DataTypes.BOOLEAN);
        fields[6] = new Field("dateField", DataTypes.DATE);
        fields[7] = new Field("timeField", DataTypes.TIMESTAMP);
        fields[8] = new Field("decimalField", DataTypes.createDecimalType(8, 2));
        fields[9] = new Field("varcharField", DataTypes.VARCHAR);
        fields[10] = new Field("arrayField", DataTypes.createArrayType(DataTypes.STRING));
        fields[11] = new Field("floatField", DataTypes.FLOAT);
        fields[12] = new Field("binaryField", DataTypes.BINARY);
        Map<String, String> map = new HashMap<>();
        map.put("complex_delimiter_level_1", "#");
        CarbonWriter writer = CarbonWriter.builder().outputPath(path).withLoadOptions(map).withCsvInput(new Schema(fields)).writtenBy("CarbonReaderTest").build();
        byte[] value = "Binary".getBytes();
        for (int i = 0; i < 10; i++) {
            Object[] row2 = new Object[] { "robot" + (i % 10), i % 10000, i, (Long.MAX_VALUE - i), ((double) i / 2), (true), "2019-03-02", "2019-02-12 03:03:34", 12.345, "varchar", "Hello#World#From#Carbon", 1.23, value };
            writer.write(row2);
        }
        writer.close();
        // Read data
        ArrowCarbonReader reader = CarbonReader.builder(path, "_temp").withRowRecordReader().buildArrowReader();
        Schema carbonSchema = CarbonSchemaReader.readSchema(path);
        byte[] data = reader.readArrowBatch(carbonSchema);
        BufferAllocator bufferAllocator = ArrowUtils.rootAllocator.newChildAllocator("toArrowBuffer", 0, Long.MAX_VALUE);
        ArrowRecordBatch arrowRecordBatch = ArrowConverter.byteArrayToArrowBatch(data, bufferAllocator);
        VectorSchemaRoot vectorSchemaRoot = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(carbonSchema, TimeZone.getDefault().getID()), bufferAllocator);
        VectorLoader vectorLoader = new VectorLoader(vectorSchemaRoot);
        vectorLoader.load(arrowRecordBatch);
        // check for 10 rows
        assertEquals(vectorSchemaRoot.getRowCount(), 10);
        List<FieldVector> fieldVectors = vectorSchemaRoot.getFieldVectors();
        // validate short column
        for (int i = 0; i < vectorSchemaRoot.getRowCount(); i++) {
            assertEquals(((SmallIntVector) fieldVectors.get(6)).get(i), i);
        }
        // validate float column
        for (int i = 0; i < vectorSchemaRoot.getRowCount(); i++) {
            assertEquals(((Float4Vector) fieldVectors.get(12)).get(i), (float) 1.23);
        }
        // validate date column
        for (int i = 0; i < vectorSchemaRoot.getRowCount(); i++) {
            assertArrayEquals(((VarCharVector) fieldVectors.get(1)).get(i), "2019-03-02".getBytes((StandardCharsets.UTF_8)));
        }
        // validate timestamp column
        for (int i = 0; i < vectorSchemaRoot.getRowCount(); i++) {
            assertArrayEquals(((VarCharVector) fieldVectors.get(2)).get(i), "2019-02-12 03:03:34".getBytes((StandardCharsets.UTF_8)));
        }
        arrowRecordBatch.close();
        vectorSchemaRoot.close();
        bufferAllocator.close();
        reader.close();
        // Read data with address (unsafe memory)
        ArrowCarbonReader reader1 = CarbonReader.builder(path, "_temp").withRowRecordReader().buildArrowReader();
        long address = reader1.readArrowBatchAddress(carbonSchema);
        int length = CarbonUnsafe.getUnsafe().getInt(address);
        byte[] data1 = new byte[length];
        CarbonUnsafe.getUnsafe().copyMemory(null, address + 4, data1, CarbonUnsafe.BYTE_ARRAY_OFFSET, length);
        bufferAllocator = ArrowUtils.rootAllocator.newChildAllocator("toArrowBuffer", 0, Long.MAX_VALUE);
        arrowRecordBatch = ArrowConverter.byteArrayToArrowBatch(data1, bufferAllocator);
        vectorSchemaRoot = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(carbonSchema, TimeZone.getDefault().getID()), bufferAllocator);
        vectorLoader = new VectorLoader(vectorSchemaRoot);
        vectorLoader.load(arrowRecordBatch);
        // check for 10 rows
        assertEquals(vectorSchemaRoot.getRowCount(), 10);
        List<FieldVector> fieldVectors1 = vectorSchemaRoot.getFieldVectors();
        // validate short column
        for (int i = 0; i < vectorSchemaRoot.getRowCount(); i++) {
            assertEquals(((SmallIntVector) fieldVectors1.get(6)).get(i), i);
        }
        // validate float column
        for (int i = 0; i < vectorSchemaRoot.getRowCount(); i++) {
            assertEquals(((Float4Vector) fieldVectors1.get(12)).get(i), (float) 1.23);
        }
        arrowRecordBatch.close();
        vectorSchemaRoot.close();
        bufferAllocator.close();
        // free the unsafe memory
        reader1.freeArrowBatchMemory(address);
        reader1.close();
        // Read as arrow vector
        ArrowCarbonReader reader2 = CarbonReader.builder(path, "_temp").withRowRecordReader().buildArrowReader();
        VectorSchemaRoot vectorSchemaRoot2 = reader2.readArrowVectors(carbonSchema);
        // check for 10 rows
        assertEquals(vectorSchemaRoot2.getRowCount(), 10);
        List<FieldVector> fieldVectors2 = vectorSchemaRoot2.getFieldVectors();
        // validate short column
        for (int i = 0; i < vectorSchemaRoot2.getRowCount(); i++) {
            assertEquals(((SmallIntVector) fieldVectors2.get(6)).get(i), i);
        }
        // validate float column
        for (int i = 0; i < vectorSchemaRoot2.getRowCount(); i++) {
            assertEquals(((Float4Vector) fieldVectors2.get(12)).get(i), (float) 1.23);
        }
        vectorSchemaRoot.close();
        reader2.close();
        // Read arrowSchema
        byte[] schema = CarbonSchemaReader.getArrowSchemaAsBytes(path);
        bufferAllocator = ArrowUtils.rootAllocator.newChildAllocator("toArrowBuffer", 0, Long.MAX_VALUE);
        arrowRecordBatch = ArrowConverter.byteArrayToArrowBatch(schema, bufferAllocator);
        vectorSchemaRoot = VectorSchemaRoot.create(ArrowUtils.toArrowSchema(carbonSchema, TimeZone.getDefault().getID()), bufferAllocator);
        vectorLoader = new VectorLoader(vectorSchemaRoot);
        vectorLoader.load(arrowRecordBatch);
        assertEquals(vectorSchemaRoot.getSchema().getFields().size(), 13);
        arrowRecordBatch.close();
        vectorSchemaRoot.close();
        bufferAllocator.close();
    } catch (Throwable e) {
        e.printStackTrace();
        Assert.fail(e.getMessage());
    } finally {
        try {
            FileUtils.deleteDirectory(new File(path));
        } catch (IOException e) {
            e.printStackTrace();
            Assert.fail(e.getMessage());
        }
    }
}