Examples with StripeStatistics com.facebook.presto.orc.metadata.statistics.StripeStatistics used on opensource projects

Example 1 with StripeStatistics

use of com.facebook.presto.orc.metadata.statistics.StripeStatistics in project presto by prestodb.

the class OrcWriter method bufferFileFooter.

/**
 * Collect the data for for the file footer.  This is not the actual data, but
 * instead are functions that know how to write the data.
 */
private List<DataOutput> bufferFileFooter() throws IOException {
    List<DataOutput> outputData = new ArrayList<>();
    Metadata metadata = new Metadata(closedStripes.stream().map(ClosedStripe::getStatistics).collect(toList()));
    Slice metadataSlice = metadataWriter.writeMetadata(metadata);
    outputData.add(createDataOutput(metadataSlice));
    numberOfRows = closedStripes.stream().mapToLong(stripe -> stripe.getStripeInformation().getNumberOfRows()).sum();
    List<ColumnStatistics> fileStats = toFileStats(closedStripes.stream().map(ClosedStripe::getStatistics).map(StripeStatistics::getColumnStatistics).collect(toList()));
    recordValidation(validation -> validation.setFileStatistics(fileStats));
    Map<String, Slice> userMetadata = this.userMetadata.entrySet().stream().collect(Collectors.toMap(Entry::getKey, entry -> utf8Slice(entry.getValue())));
    unencryptedStats = new ArrayList<>();
    Map<Integer, Map<Integer, Slice>> encryptedStats = new HashMap<>();
    addStatsRecursive(fileStats, 0, new HashMap<>(), unencryptedStats, encryptedStats);
    Optional<DwrfEncryption> dwrfEncryption;
    if (dwrfWriterEncryption.isPresent()) {
        ImmutableList.Builder<EncryptionGroup> encryptionGroupBuilder = ImmutableList.builder();
        List<WriterEncryptionGroup> writerEncryptionGroups = dwrfWriterEncryption.get().getWriterEncryptionGroups();
        for (int i = 0; i < writerEncryptionGroups.size(); i++) {
            WriterEncryptionGroup group = writerEncryptionGroups.get(i);
            Map<Integer, Slice> groupStats = encryptedStats.get(i);
            encryptionGroupBuilder.add(new EncryptionGroup(group.getNodes(), // reader will just use key metadata from the stripe
            Optional.empty(), group.getNodes().stream().map(groupStats::get).collect(toList())));
        }
        dwrfEncryption = Optional.of(new DwrfEncryption(dwrfWriterEncryption.get().getKeyProvider(), encryptionGroupBuilder.build()));
    } else {
        dwrfEncryption = Optional.empty();
    }
    Optional<DwrfStripeCacheData> dwrfStripeCacheData = dwrfStripeCacheWriter.map(DwrfStripeCacheWriter::getDwrfStripeCacheData);
    Slice dwrfStripeCacheSlice = metadataWriter.writeDwrfStripeCache(dwrfStripeCacheData);
    outputData.add(createDataOutput(dwrfStripeCacheSlice));
    Optional<List<Integer>> dwrfStripeCacheOffsets = dwrfStripeCacheWriter.map(DwrfStripeCacheWriter::getOffsets);
    Footer footer = new Footer(numberOfRows, rowGroupMaxRowCount, OptionalLong.of(rawSize), closedStripes.stream().map(ClosedStripe::getStripeInformation).collect(toList()), orcTypes, ImmutableList.copyOf(unencryptedStats), userMetadata, dwrfEncryption, dwrfStripeCacheOffsets);
    closedStripes.clear();
    closedStripesRetainedBytes = 0;
    Slice footerSlice = metadataWriter.writeFooter(footer);
    outputData.add(createDataOutput(footerSlice));
    recordValidation(validation -> validation.setVersion(metadataWriter.getOrcMetadataVersion()));
    Slice postscriptSlice = metadataWriter.writePostscript(footerSlice.length(), metadataSlice.length(), columnWriterOptions.getCompressionKind(), columnWriterOptions.getCompressionMaxBufferSize(), dwrfStripeCacheData);
    outputData.add(createDataOutput(postscriptSlice));
    outputData.add(createDataOutput(Slices.wrappedBuffer((byte) postscriptSlice.length())));
    return outputData;
}

Example 2 with StripeStatistics

use of com.facebook.presto.orc.metadata.statistics.StripeStatistics in project presto by prestodb.

the class OrcWriteValidation method validateStripeStatistics.

public void validateStripeStatistics(OrcDataSourceId orcDataSourceId, List<StripeInformation> actualStripes, List<StripeStatistics> actualStripeStatistics) throws OrcCorruptionException {
    requireNonNull(actualStripes, "actualStripes is null");
    requireNonNull(actualStripeStatistics, "actualStripeStatistics is null");
    if (actualStripeStatistics.isEmpty()) {
        // DWRF does not have stripe statistics
        return;
    }
    if (actualStripeStatistics.size() != stripeStatistics.size()) {
        throw new OrcCorruptionException(orcDataSourceId, "Write validation failed: unexpected number of columns in stripe statistics");
    }
    for (int stripeIndex = 0; stripeIndex < actualStripes.size(); stripeIndex++) {
        long stripeOffset = actualStripes.get(stripeIndex).getOffset();
        StripeStatistics actual = actualStripeStatistics.get(stripeIndex);
        validateStripeStatistics(orcDataSourceId, stripeOffset, actual.getColumnStatistics());
    }
}

Example 3 with StripeStatistics

use of com.facebook.presto.orc.metadata.statistics.StripeStatistics in project presto by prestodb.

the class OrcWriteValidation method validateStripeStatistics.

public void validateStripeStatistics(OrcDataSourceId orcDataSourceId, long stripeOffset, List<ColumnStatistics> actual) throws OrcCorruptionException {
    StripeStatistics expected = stripeStatistics.get(stripeOffset);
    if (expected == null) {
        throw new OrcCorruptionException(orcDataSourceId, "Unexpected stripe at offset %s", stripeOffset);
    }
    validateColumnStatisticsEquivalent(orcDataSourceId, "Stripe at " + stripeOffset, actual, expected.getColumnStatistics());
}

Example 4 with StripeStatistics

use of com.facebook.presto.orc.metadata.statistics.StripeStatistics in project presto by prestodb.

the class OrcWriter method bufferStripeData.

/**
 * Collect the data for for the stripe.  This is not the actual data, but
 * instead are functions that know how to write the data.
 */
private List<DataOutput> bufferStripeData(long stripeStartOffset, FlushReason flushReason) throws IOException {
    if (stripeRowCount == 0) {
        return ImmutableList.of();
    }
    List<DataOutput> outputData = new ArrayList<>();
    List<Stream> unencryptedStreams = new ArrayList<>(columnWriters.size() * 3);
    Multimap<Integer, Stream> encryptedStreams = ArrayListMultimap.create();
    // get index streams
    long indexLength = 0;
    long offset = 0;
    int previousEncryptionGroup = -1;
    for (ColumnWriter columnWriter : columnWriters) {
        for (StreamDataOutput indexStream : columnWriter.getIndexStreams()) {
            // The ordering is critical because the stream only contain a length with no offset.
            // if the previous stream was part of a different encryption group, need to specify an offset so we know the column order
            outputData.add(indexStream);
            Optional<Integer> encryptionGroup = dwrfEncryptionInfo.getGroupByNodeId(indexStream.getStream().getColumn());
            if (encryptionGroup.isPresent()) {
                Stream stream = previousEncryptionGroup == encryptionGroup.get() ? indexStream.getStream() : indexStream.getStream().withOffset(offset);
                encryptedStreams.put(encryptionGroup.get(), stream);
                previousEncryptionGroup = encryptionGroup.get();
            } else {
                Stream stream = previousEncryptionGroup == -1 ? indexStream.getStream() : indexStream.getStream().withOffset(offset);
                unencryptedStreams.add(stream);
                previousEncryptionGroup = -1;
            }
            offset += indexStream.size();
            indexLength += indexStream.size();
        }
    }
    if (dwrfStripeCacheWriter.isPresent()) {
        dwrfStripeCacheWriter.get().addIndexStreams(ImmutableList.copyOf(outputData), indexLength);
    }
    // data streams (sorted by size)
    long dataLength = 0;
    List<StreamDataOutput> dataStreams = new ArrayList<>(columnWriters.size() * 2);
    for (ColumnWriter columnWriter : columnWriters) {
        List<StreamDataOutput> streams = columnWriter.getDataStreams();
        dataStreams.addAll(streams);
        dataLength += streams.stream().mapToLong(StreamDataOutput::size).sum();
    }
    streamLayout.reorder(dataStreams);
    // add data streams
    for (StreamDataOutput dataStream : dataStreams) {
        // The ordering is critical because the stream only contains a length with no offset.
        // if the previous stream was part of a different encryption group, need to specify an offset so we know the column order
        outputData.add(dataStream);
        Optional<Integer> encryptionGroup = dwrfEncryptionInfo.getGroupByNodeId(dataStream.getStream().getColumn());
        if (encryptionGroup.isPresent()) {
            Stream stream = previousEncryptionGroup == encryptionGroup.get() ? dataStream.getStream() : dataStream.getStream().withOffset(offset);
            encryptedStreams.put(encryptionGroup.get(), stream);
            previousEncryptionGroup = encryptionGroup.get();
        } else {
            Stream stream = previousEncryptionGroup == -1 ? dataStream.getStream() : dataStream.getStream().withOffset(offset);
            unencryptedStreams.add(stream);
            previousEncryptionGroup = -1;
        }
        offset += dataStream.size();
    }
    Map<Integer, ColumnEncoding> columnEncodings = new HashMap<>();
    columnWriters.forEach(columnWriter -> columnEncodings.putAll(columnWriter.getColumnEncodings()));
    Map<Integer, ColumnStatistics> columnStatistics = new HashMap<>();
    columnWriters.forEach(columnWriter -> columnStatistics.putAll(columnWriter.getColumnStripeStatistics()));
    // the 0th column is a struct column for the whole row
    columnEncodings.put(0, new ColumnEncoding(DIRECT, 0));
    columnStatistics.put(0, new ColumnStatistics((long) stripeRowCount, null));
    Map<Integer, ColumnEncoding> unencryptedColumnEncodings = columnEncodings.entrySet().stream().filter(entry -> !dwrfEncryptionInfo.getGroupByNodeId(entry.getKey()).isPresent()).collect(toImmutableMap(Entry::getKey, Entry::getValue));
    Map<Integer, ColumnEncoding> encryptedColumnEncodings = columnEncodings.entrySet().stream().filter(entry -> dwrfEncryptionInfo.getGroupByNodeId(entry.getKey()).isPresent()).collect(toImmutableMap(Entry::getKey, Entry::getValue));
    List<Slice> encryptedGroups = createEncryptedGroups(encryptedStreams, encryptedColumnEncodings);
    StripeFooter stripeFooter = new StripeFooter(unencryptedStreams, unencryptedColumnEncodings, encryptedGroups);
    Slice footer = metadataWriter.writeStripeFooter(stripeFooter);
    outputData.add(createDataOutput(footer));
    dwrfStripeCacheWriter.ifPresent(stripeCacheWriter -> stripeCacheWriter.addStripeFooter(createDataOutput(footer)));
    // create final stripe statistics
    StripeStatistics statistics = new StripeStatistics(toDenseList(columnStatistics, orcTypes.size()));
    recordValidation(validation -> validation.addStripeStatistics(stripeStartOffset, statistics));
    StripeInformation stripeInformation = new StripeInformation(stripeRowCount, stripeStartOffset, indexLength, dataLength, footer.length(), OptionalLong.of(stripeRawSize), dwrfEncryptionInfo.getEncryptedKeyMetadatas());
    ClosedStripe closedStripe = new ClosedStripe(stripeInformation, statistics);
    closedStripes.add(closedStripe);
    closedStripesRetainedBytes += closedStripe.getRetainedSizeInBytes();
    recordValidation(validation -> validation.addStripe(stripeInformation.getNumberOfRows()));
    stats.recordStripeWritten(flushPolicy.getStripeMinBytes(), flushPolicy.getStripeMaxBytes(), dictionaryMaxMemoryBytes, flushReason, dictionaryCompressionOptimizer.getDictionaryMemoryBytes(), stripeInformation);
    return outputData;
}