Examples with OrcWriter com.facebook.presto.orc.OrcWriter used on opensource projects

Example 1 with OrcWriter

use of com.facebook.presto.orc.OrcWriter in project presto by prestodb.

the class TestWriterBlockRawSize method testFileMetadataRawSize.

@Test
public void testFileMetadataRawSize() throws IOException {
    Type type = INTEGER;
    List<Type> types = ImmutableList.of(type);
    int numBlocksPerRowGroup = 3;
    int numBlocksPerStripe = numBlocksPerRowGroup * 5;
    int numStripes = 4;
    int numBlocksPerFile = numBlocksPerStripe * numStripes + 1;
    BlockBuilder blockBuilder = type.createBlockBuilder(null, NUM_ELEMENTS * 2);
    for (int i = 0; i < NUM_ELEMENTS; i++) {
        blockBuilder.appendNull();
        type.writeLong(blockBuilder, i);
    }
    long blockRawSize = ((FixedWidthType) type).getFixedSize() * NUM_ELEMENTS + NUM_ELEMENTS;
    Block block = blockBuilder.build();
    Block[] blocks = new Block[] { block };
    OrcWriterOptions writerOptions = OrcWriterOptions.builder().withRowGroupMaxRowCount(block.getPositionCount() * numBlocksPerRowGroup).withFlushPolicy(DefaultOrcWriterFlushPolicy.builder().withStripeMaxRowCount(block.getPositionCount() * numBlocksPerStripe).build()).build();
    for (OrcEncoding encoding : OrcEncoding.values()) {
        try (TempFile tempFile = new TempFile()) {
            OrcWriter writer = createOrcWriter(tempFile.getFile(), encoding, ZSTD, Optional.empty(), types, writerOptions, new OrcWriterStats());
            for (int i = 0; i < numBlocksPerFile; i++) {
                writer.write(new Page(blocks));
            }
            writer.close();
            writer.validate(new FileOrcDataSource(tempFile.getFile(), new DataSize(1, MEGABYTE), new DataSize(1, MEGABYTE), new DataSize(1, MEGABYTE), true));
            Footer footer = OrcTester.getFileMetadata(tempFile.getFile(), encoding).getFooter();
            verifyValue(encoding, footer.getRawSize(), blockRawSize * numBlocksPerFile);
            assertEquals(footer.getStripes().size(), numStripes + 1);
            int numBlocksRemaining = numBlocksPerFile;
            for (StripeInformation stripeInfo : footer.getStripes()) {
                int numBlocksInStripe = Math.min(numBlocksRemaining, numBlocksPerStripe);
                verifyValue(encoding, stripeInfo.getRawDataSize(), blockRawSize * numBlocksInStripe);
                numBlocksRemaining -= numBlocksInStripe;
            }
        }
    }
}

Example 2 with OrcWriter

use of com.facebook.presto.orc.OrcWriter in project presto by prestodb.

the class OrcFileRewriter method rewrite.

public OrcFileInfo rewrite(FileSystem fileSystem, Map<String, Type> allColumnTypes, Path input, Path output, BitSet rowsToDelete) throws IOException {
    try (ThreadContextClassLoader ignored = new ThreadContextClassLoader(FileSystem.class.getClassLoader());
        OrcDataSource dataSource = orcDataEnvironment.createOrcDataSource(fileSystem, input, readerAttributes)) {
        OrcReader reader = new OrcReader(dataSource, ORC, orcFileTailSource, stripeMetadataSourceFactory, new RaptorOrcAggregatedMemoryContext(), new OrcReaderOptions(readerAttributes.getMaxMergeDistance(), readerAttributes.getTinyStripeThreshold(), HUGE_MAX_READ_BLOCK_SIZE, readerAttributes.isZstdJniDecompressionEnabled()), false, NO_ENCRYPTION, DwrfKeyProvider.EMPTY, new RuntimeStats());
        if (reader.getFooter().getNumberOfRows() < rowsToDelete.length()) {
            throw new IOException("File has fewer rows than deletion vector");
        }
        int deleteRowCount = rowsToDelete.cardinality();
        if (reader.getFooter().getNumberOfRows() == deleteRowCount) {
            return new OrcFileInfo(0, 0);
        }
        if (reader.getFooter().getNumberOfRows() >= Integer.MAX_VALUE) {
            throw new IOException("File has too many rows");
        }
        int inputRowCount = toIntExact(reader.getFooter().getNumberOfRows());
        Map<String, Integer> currentColumnIds = IntStream.range(0, reader.getColumnNames().size()).boxed().collect(toMap(reader.getColumnNames()::get, i -> i));
        ImmutableList.Builder<Type> writerColumnTypesBuilder = ImmutableList.builder();
        ImmutableList.Builder<String> writerColumnIdsBuilder = ImmutableList.builder();
        ImmutableList.Builder<Integer> readerColumnIndexBuilder = ImmutableList.builder();
        // Build columns for writer; keep the right ordinal
        Map<String, Type> orderedAllColumnTypes = new TreeMap<>(Comparator.comparingLong(Long::parseLong));
        orderedAllColumnTypes.putAll(allColumnTypes);
        for (Map.Entry<String, Type> columnType : orderedAllColumnTypes.entrySet()) {
            // Get the intersection of the provide columns and the actual columns
            Integer currentColumnIndex = currentColumnIds.get(columnType.getKey());
            if (currentColumnIndex != null) {
                readerColumnIndexBuilder.add(currentColumnIndex);
                writerColumnTypesBuilder.add(columnType.getValue());
                writerColumnIdsBuilder.add(columnType.getKey());
            }
        }
        List<Type> writerColumnTypes = writerColumnTypesBuilder.build();
        List<String> writerColumnIds = writerColumnIdsBuilder.build();
        List<Integer> readerColumnIndex = readerColumnIndexBuilder.build();
        Map<Integer, Type> readerColumns = IntStream.range(0, readerColumnIndex.size()).boxed().collect(toMap(readerColumnIndex::get, writerColumnTypes::get));
        if (writerColumnTypes.isEmpty()) {
            // no intersection; directly return
            return new OrcFileInfo(0, 0);
        }
        StorageTypeConverter converter = new StorageTypeConverter(typeManager);
        List<Type> writerStorageTypes = writerColumnTypes.stream().map(converter::toStorageType).collect(toImmutableList());
        long start = System.nanoTime();
        Map<String, String> userMetadata = ImmutableMap.of();
        if (reader.getFooter().getUserMetadata().containsKey(OrcFileMetadata.KEY)) {
            // build metadata if the original file has it
            ImmutableMap.Builder<Long, TypeSignature> metadataBuilder = ImmutableMap.builder();
            for (int i = 0; i < writerColumnIds.size(); i++) {
                metadataBuilder.put(Long.parseLong(writerColumnIds.get(i)), writerColumnTypes.get(i).getTypeSignature());
            }
            userMetadata = ImmutableMap.of(OrcFileMetadata.KEY, METADATA_CODEC.toJson(new OrcFileMetadata(metadataBuilder.build())));
        }
        StorageTypeConverter storageTypeConverter = new StorageTypeConverter(typeManager);
        try (Closer<OrcBatchRecordReader, IOException> recordReader = closer(reader.createBatchRecordReader(storageTypeConverter.toStorageTypes(readerColumns), TRUE, DEFAULT_STORAGE_TIMEZONE, new RaptorOrcAggregatedMemoryContext(), INITIAL_BATCH_SIZE), OrcBatchRecordReader::close);
            Closer<OrcWriter, IOException> writer = closer(new OrcWriter(orcDataEnvironment.createOrcDataSink(fileSystem, output), writerColumnIds, writerStorageTypes, ORC, compression, Optional.empty(), NO_ENCRYPTION, getDefaultOrcWriterOptions(), userMetadata, DEFAULT_STORAGE_TIMEZONE, validate, HASHED, stats), OrcWriter::close)) {
            OrcFileInfo fileInfo = rewrite(recordReader.get(), writer.get(), rowsToDelete, writerColumnTypes, readerColumnIndexBuilder.build());
            log.debug("Rewrote file %s in %s (input rows: %s, output rows: %s)", input.getName(), nanosSince(start), inputRowCount, inputRowCount - deleteRowCount);
            return fileInfo;
        }
    } catch (NotSupportedException e) {
        throw new PrestoException(NOT_SUPPORTED, e.getMessage(), e);
    }
}