Examples with CompressionCodecName org.apache.parquet.hadoop.metadata.CompressionCodecName used on opensource projects

Example 16 with CompressionCodecName

use of org.apache.parquet.hadoop.metadata.CompressionCodecName in project parquet-mr by apache.

the class ParquetMetadataCommand method printColumnChunk.

private void printColumnChunk(Logger console, int width, ColumnChunkMetaData column, MessageType schema) {
    String[] path = column.getPath().toArray();
    PrimitiveType type = primitive(schema, path);
    Preconditions.checkNotNull(type);
    ColumnDescriptor desc = schema.getColumnDescription(path);
    long size = column.getTotalSize();
    long count = column.getValueCount();
    float perValue = ((float) size) / count;
    CompressionCodecName codec = column.getCodec();
    Set<Encoding> encodings = column.getEncodings();
    EncodingStats encodingStats = column.getEncodingStats();
    String encodingSummary = encodingStats == null ? encodingsAsString(encodings, desc) : encodingStatsAsString(encodingStats);
    Statistics stats = column.getStatistics();
    String name = column.getPath().toDotString();
    PrimitiveType.PrimitiveTypeName typeName = type.getPrimitiveTypeName();
    if (typeName == PrimitiveType.PrimitiveTypeName.FIXED_LEN_BYTE_ARRAY) {
        console.info(String.format("%-" + width + "s  FIXED[%d] %s %-7s %-9d %-8s %-7s %s", name, type.getTypeLength(), shortCodec(codec), encodingSummary, count, humanReadable(perValue), stats == null || !stats.isNumNullsSet() ? "" : String.valueOf(stats.getNumNulls()), minMaxAsString(stats, type.getOriginalType())));
    } else {
        console.info(String.format("%-" + width + "s  %-9s %s %-7s %-9d %-10s %-7s %s", name, typeName, shortCodec(codec), encodingSummary, count, humanReadable(perValue), stats == null || !stats.isNumNullsSet() ? "" : String.valueOf(stats.getNumNulls()), minMaxAsString(stats, type.getOriginalType())));
    }
}

Example 17 with CompressionCodecName

use of org.apache.parquet.hadoop.metadata.CompressionCodecName in project parquet-mr by apache.

the class TestDirectCodecFactory method compressionCodecs.

@Test
public void compressionCodecs() throws Exception {
    final int[] sizes = { 4 * 1024, 1 * 1024 * 1024 };
    final boolean[] comp = { true, false };
    Set<CompressionCodecName> codecsToSkip = new HashSet<>();
    // not distributed because it is GPL
    codecsToSkip.add(LZO);
    // not distributed in the default version of Hadoop
    codecsToSkip.add(LZ4);
    // not distributed in the default version of Hadoop
    codecsToSkip.add(ZSTD);
    for (final int size : sizes) {
        for (final boolean useOnHeapComp : comp) {
            for (final Decompression decomp : Decompression.values()) {
                for (final CompressionCodecName codec : CompressionCodecName.values()) {
                    if (codecsToSkip.contains(codec)) {
                        continue;
                    }
                    test(size, codec, useOnHeapComp, decomp);
                }
            }
        }
    }
}

Example 18 with CompressionCodecName

use of org.apache.parquet.hadoop.metadata.CompressionCodecName in project parquet-mr by apache.

the class ParquetOutputFormat method getRecordWriter.

/**
 * {@inheritDoc}
 */
@Override
public RecordWriter<Void, T> getRecordWriter(TaskAttemptContext taskAttemptContext) throws IOException, InterruptedException {
    final Configuration conf = getConfiguration(taskAttemptContext);
    CompressionCodecName codec = getCodec(taskAttemptContext);
    String extension = codec.getExtension() + ".parquet";
    Path file = getDefaultWorkFile(taskAttemptContext, extension);
    return getRecordWriter(conf, file, codec);
}

Example 19 with CompressionCodecName

use of org.apache.parquet.hadoop.metadata.CompressionCodecName in project parquet-mr by apache.

the class CodecConfig method getHadoopCompressionCodec.

private CompressionCodecName getHadoopCompressionCodec() {
    CompressionCodecName codec;
    try {
        // find the right codec
        Class<?> codecClass = getHadoopOutputCompressorClass(CompressionCodecName.UNCOMPRESSED.getHadoopCompressionCodecClass());
        LOG.info("Compression set through hadoop codec: {}", codecClass.getName());
        codec = CompressionCodecName.fromCompressionCodec(codecClass);
    } catch (CompressionCodecNotSupportedException e) {
        LOG.warn("codec defined in hadoop config is not supported by parquet [{}] and will use UNCOMPRESSED", e.getCodecClass().getName(), e);
        codec = CompressionCodecName.UNCOMPRESSED;
    } catch (IllegalArgumentException e) {
        LOG.warn("codec class not found: {}", e.getMessage(), e);
        codec = CompressionCodecName.UNCOMPRESSED;
    }
    return codec;
}

Example 20 with CompressionCodecName

use of org.apache.parquet.hadoop.metadata.CompressionCodecName in project drill by axbaretto.

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());
}