Examples with MessageType org.apache.parquet.schema.MessageType used on opensource projects

Example 36 with MessageType

use of org.apache.parquet.schema.MessageType in project hive by apache.

the class TestParquetFilterPredicate method testFilterBetween.

@Test
public void testFilterBetween() {
    MessageType schema = MessageTypeParser.parseMessageType("message test {  required int32 bCol; }");
    SearchArgument sarg = SearchArgumentFactory.newBuilder().between("bCol", PredicateLeaf.Type.LONG, 1L, 5L).build();
    FilterPredicate p = ParquetFilterPredicateConverter.toFilterPredicate(sarg, schema);
    String expected = "and(lteq(bCol, 5), not(lt(bCol, 1)))";
    assertEquals(expected, p.toString());
    sarg = SearchArgumentFactory.newBuilder().between("bCol", PredicateLeaf.Type.LONG, 5L, 1L).build();
    p = ParquetFilterPredicateConverter.toFilterPredicate(sarg, schema);
    expected = "and(lteq(bCol, 1), not(lt(bCol, 5)))";
    assertEquals(expected, p.toString());
    sarg = SearchArgumentFactory.newBuilder().between("bCol", PredicateLeaf.Type.LONG, 1L, 1L).build();
    p = ParquetFilterPredicateConverter.toFilterPredicate(sarg, schema);
    expected = "and(lteq(bCol, 1), not(lt(bCol, 1)))";
    assertEquals(expected, p.toString());
}

Example 37 with MessageType

use of org.apache.parquet.schema.MessageType in project druid by druid-io.

the class DruidParquetReadSupport method prepareForRead.

@Override
public RecordMaterializer<GenericRecord> prepareForRead(Configuration configuration, Map<String, String> keyValueMetaData, MessageType fileSchema, ReadContext readContext) {
    MessageType parquetSchema = readContext.getRequestedSchema();
    Schema avroSchema = new AvroSchemaConverter(configuration).convert(parquetSchema);
    Class<? extends AvroDataSupplier> suppClass = configuration.getClass(AVRO_DATA_SUPPLIER, SpecificDataSupplier.class, AvroDataSupplier.class);
    AvroDataSupplier supplier = ReflectionUtils.newInstance(suppClass, configuration);
    return new AvroRecordMaterializer<GenericRecord>(parquetSchema, avroSchema, supplier.get());
}

Example 38 with MessageType

use of org.apache.parquet.schema.MessageType in project h2o-3 by h2oai.

the class VecParquetReader method initReader.

private void initReader() throws IOException {
    assert reader == null;
    List<BlockMetaData> blocks = metadata.getBlocks();
    MessageType fileSchema = metadata.getFileMetaData().getSchema();
    reader = new InternalParquetRecordReader<>(new ChunkReadSupport(writer, chunkSchema));
    Configuration conf = VecFileSystem.makeConfiguration(vec);
    reader.initialize(fileSchema, metadata.getFileMetaData().getKeyValueMetaData(), VecFileSystem.VEC_PATH, blocks, conf);
}

Example 39 with MessageType

use of org.apache.parquet.schema.MessageType in project drill by apache.

the class ParquetRecordWriter method newSchema.

private void newSchema() throws IOException {
    List<Type> types = Lists.newArrayList();
    for (MaterializedField field : batchSchema) {
        if (field.getPath().equalsIgnoreCase(WriterPrel.PARTITION_COMPARATOR_FIELD)) {
            continue;
        }
        types.add(getType(field));
    }
    schema = new MessageType("root", types);
    int initialBlockBufferSize = max(MINIMUM_BUFFER_SIZE, blockSize / this.schema.getColumns().size() / 5);
    pageStore = ColumnChunkPageWriteStoreExposer.newColumnChunkPageWriteStore(this.oContext, codecFactory.getCompressor(codec), schema);
    int initialPageBufferSize = max(MINIMUM_BUFFER_SIZE, min(pageSize + pageSize / 10, initialBlockBufferSize));
    store = new ColumnWriteStoreV1(pageStore, pageSize, initialPageBufferSize, enableDictionary, writerVersion, new ParquetDirectByteBufferAllocator(oContext));
    MessageColumnIO columnIO = new ColumnIOFactory(false).getColumnIO(this.schema);
    consumer = columnIO.getRecordWriter(store);
    setUp(schema, consumer);
}

Example 40 with MessageType

use of org.apache.parquet.schema.MessageType in project drill by apache.

the class TestFileGenerator method generateParquetFile.

public static void generateParquetFile(String filename, ParquetTestProperties props) throws Exception {
    int currentBooleanByte = 0;
    WrapAroundCounter booleanBitCounter = new WrapAroundCounter(7);
    Configuration configuration = new Configuration();
    configuration.set(FileSystem.FS_DEFAULT_NAME_KEY, "file:///");
    //"message m { required int32 integer; required int64 integer64; required boolean b; required float f; required double d;}"
    FileSystem fs = FileSystem.get(configuration);
    Path path = new Path(filename);
    if (fs.exists(path)) {
        fs.delete(path, false);
    }
    String messageSchema = "message m {";
    for (FieldInfo fieldInfo : props.fields.values()) {
        messageSchema += " required " + fieldInfo.parquetType + " " + fieldInfo.name + ";";
    }
    // remove the last semicolon, java really needs a join method for strings...
    // TODO - nvm apparently it requires a semicolon after every field decl, might want to file a bug
    //messageSchema = messageSchema.substring(schemaType, messageSchema.length() - 1);
    messageSchema += "}";
    MessageType schema = MessageTypeParser.parseMessageType(messageSchema);
    CompressionCodecName codec = CompressionCodecName.UNCOMPRESSED;
    ParquetFileWriter w = new ParquetFileWriter(configuration, schema, path);
    w.start();
    HashMap<String, Integer> columnValuesWritten = new HashMap<>();
    int valsWritten;
    for (int k = 0; k < props.numberRowGroups; k++) {
        w.startBlock(props.recordsPerRowGroup);
        currentBooleanByte = 0;
        booleanBitCounter.reset();
        for (FieldInfo fieldInfo : props.fields.values()) {
            if (!columnValuesWritten.containsKey(fieldInfo.name)) {
                columnValuesWritten.put(fieldInfo.name, 0);
                valsWritten = 0;
            } else {
                valsWritten = columnValuesWritten.get(fieldInfo.name);
            }
            String[] path1 = { fieldInfo.name };
            ColumnDescriptor c1 = schema.getColumnDescription(path1);
            w.startColumn(c1, props.recordsPerRowGroup, codec);
            final int valsPerPage = (int) Math.ceil(props.recordsPerRowGroup / (float) fieldInfo.numberOfPages);
            // 1 MB
            final int PAGE_SIZE = 1024 * 1024;
            byte[] bytes;
            RunLengthBitPackingHybridValuesWriter defLevels = new RunLengthBitPackingHybridValuesWriter(MAX_EXPECTED_BIT_WIDTH_FOR_DEFINITION_LEVELS, valsPerPage, PAGE_SIZE, new DirectByteBufferAllocator());
            RunLengthBitPackingHybridValuesWriter repLevels = new RunLengthBitPackingHybridValuesWriter(MAX_EXPECTED_BIT_WIDTH_FOR_DEFINITION_LEVELS, valsPerPage, PAGE_SIZE, new DirectByteBufferAllocator());
            // for variable length binary fields
            int bytesNeededToEncodeLength = 4;
            if (fieldInfo.bitLength > 0) {
                bytes = new byte[(int) Math.ceil(valsPerPage * fieldInfo.bitLength / 8.0)];
            } else {
                // the twelve at the end is to account for storing a 4 byte length with each value
                int totalValLength = ((byte[]) fieldInfo.values[0]).length + ((byte[]) fieldInfo.values[1]).length + ((byte[]) fieldInfo.values[2]).length + 3 * bytesNeededToEncodeLength;
                // used for the case where there is a number of values in this row group that is not divisible by 3
                int leftOverBytes = 0;
                if (valsPerPage % 3 > 0) {
                    leftOverBytes += ((byte[]) fieldInfo.values[1]).length + bytesNeededToEncodeLength;
                }
                if (valsPerPage % 3 > 1) {
                    leftOverBytes += ((byte[]) fieldInfo.values[2]).length + bytesNeededToEncodeLength;
                }
                bytes = new byte[valsPerPage / 3 * totalValLength + leftOverBytes];
            }
            int bytesPerPage = (int) (valsPerPage * (fieldInfo.bitLength / 8.0));
            int bytesWritten = 0;
            for (int z = 0; z < fieldInfo.numberOfPages; z++, bytesWritten = 0) {
                for (int i = 0; i < valsPerPage; i++) {
                    repLevels.writeInteger(0);
                    defLevels.writeInteger(1);
                    //System.out.print(i + ", " + (i % 25 == 0 ? "\n gen " + fieldInfo.name + ": " : ""));
                    if (fieldInfo.values[0] instanceof Boolean) {
                        bytes[currentBooleanByte] |= bitFields[booleanBitCounter.val] & ((boolean) fieldInfo.values[valsWritten % 3] ? allBitsTrue : allBitsFalse);
                        booleanBitCounter.increment();
                        if (booleanBitCounter.val == 0) {
                            currentBooleanByte++;
                        }
                        valsWritten++;
                        if (currentBooleanByte > bytesPerPage) {
                            break;
                        }
                    } else {
                        if (fieldInfo.values[valsWritten % 3] instanceof byte[]) {
                            System.arraycopy(ByteArrayUtil.toByta(((byte[]) fieldInfo.values[valsWritten % 3]).length), 0, bytes, bytesWritten, bytesNeededToEncodeLength);
                            System.arraycopy(fieldInfo.values[valsWritten % 3], 0, bytes, bytesWritten + bytesNeededToEncodeLength, ((byte[]) fieldInfo.values[valsWritten % 3]).length);
                            bytesWritten += ((byte[]) fieldInfo.values[valsWritten % 3]).length + bytesNeededToEncodeLength;
                        } else {
                            System.arraycopy(ByteArrayUtil.toByta(fieldInfo.values[valsWritten % 3]), 0, bytes, i * (fieldInfo.bitLength / 8), fieldInfo.bitLength / 8);
                        }
                        valsWritten++;
                    }
                }
                byte[] fullPage = new byte[2 * 4 * valsPerPage + bytes.length];
                byte[] repLevelBytes = repLevels.getBytes().toByteArray();
                byte[] defLevelBytes = defLevels.getBytes().toByteArray();
                System.arraycopy(bytes, 0, fullPage, 0, bytes.length);
                System.arraycopy(repLevelBytes, 0, fullPage, bytes.length, repLevelBytes.length);
                System.arraycopy(defLevelBytes, 0, fullPage, bytes.length + repLevelBytes.length, defLevelBytes.length);
                w.writeDataPage((props.recordsPerRowGroup / fieldInfo.numberOfPages), fullPage.length, BytesInput.from(fullPage), RLE, RLE, PLAIN);
                currentBooleanByte = 0;
            }
            w.endColumn();
            columnValuesWritten.remove(fieldInfo.name);
            columnValuesWritten.put(fieldInfo.name, valsWritten);
        }
        w.endBlock();
    }
    w.end(new HashMap<String, String>());
    logger.debug("Finished generating parquet file {}", path.getName());
}