Examples with ColumnEncoding io.prestosql.orc.metadata.ColumnEncoding used on opensource projects

Example 1 with ColumnEncoding

use of io.prestosql.orc.metadata.ColumnEncoding in project hetu-core by openlookeng.

the class AbstractOrcRecordReader method advanceToNextStripe.

private void advanceToNextStripe() throws IOException {
    currentStripeSystemMemoryContext.close();
    currentStripeSystemMemoryContext = systemMemoryUsage.newAggregatedMemoryContext();
    rowGroups = ImmutableList.<RowGroup>of().iterator();
    if (currentStripe >= 0) {
        if (stripeStatisticsValidation.isPresent()) {
            OrcWriteValidation.StatisticsValidation statisticsValidation = stripeStatisticsValidation.get();
            long offset = stripes.get(currentStripe).getOffset();
            writeValidation.get().validateStripeStatistics(orcDataSource.getId(), offset, statisticsValidation.build().get());
            statisticsValidation.reset();
        }
    }
    currentStripe++;
    if (currentStripe >= stripes.size()) {
        return;
    }
    if (currentStripe > 0) {
        currentStripePosition += stripes.get(currentStripe - 1).getNumberOfRows();
    }
    StripeInformation stripeInformation = stripes.get(currentStripe);
    validateWriteStripe(stripeInformation.getNumberOfRows());
    Stripe stripe = stripeReader.readStripe(stripeInformation, currentStripeSystemMemoryContext);
    if (stripe != null) {
        // Give readers access to dictionary streams
        InputStreamSources dictionaryStreamSources = stripe.getDictionaryStreamSources();
        ColumnMetadata<ColumnEncoding> columnEncodings = stripe.getColumnEncodings();
        for (AbstractColumnReader columnReader : columnReaders) {
            if (columnReader != null) {
                ZoneId fileTimeZone = stripe.getFileTimeZone();
                columnReader.startStripe(fileTimeZone, dictionaryStreamSources, columnEncodings);
            }
        }
        rowGroups = stripe.getRowGroups().iterator();
    }
}

Example 2 with ColumnEncoding

use of io.prestosql.orc.metadata.ColumnEncoding in project hetu-core by openlookeng.

the class SliceDictionaryColumnWriter method bufferOutputData.

private void bufferOutputData() {
    checkState(closed);
    checkState(!directEncoded);
    Block dictionaryElements = dictionary.getElementBlock();
    // write dictionary in sorted order
    int[] sortedDictionaryIndexes = getSortedDictionaryNullsLast(dictionaryElements);
    for (int sortedDictionaryIndex : sortedDictionaryIndexes) {
        if (!dictionaryElements.isNull(sortedDictionaryIndex)) {
            int length = dictionaryElements.getSliceLength(sortedDictionaryIndex);
            dictionaryLengthStream.writeLong(length);
            Slice value = dictionaryElements.getSlice(sortedDictionaryIndex, 0, length);
            dictionaryDataStream.writeSlice(value);
        }
    }
    columnEncoding = new ColumnEncoding(DICTIONARY_V2, dictionaryElements.getPositionCount() - 1);
    // build index from original dictionary index to new sorted position
    int[] originalDictionaryToSortedIndex = new int[sortedDictionaryIndexes.length];
    for (int sortOrdinal = 0; sortOrdinal < sortedDictionaryIndexes.length; sortOrdinal++) {
        int dictionaryIndex = sortedDictionaryIndexes[sortOrdinal];
        originalDictionaryToSortedIndex[dictionaryIndex] = sortOrdinal;
    }
    if (!rowGroups.isEmpty()) {
        presentStream.recordCheckpoint();
        dataStream.recordCheckpoint();
    }
    for (DictionaryRowGroup rowGroup : rowGroups) {
        IntBigArray dictionaryIndexes = rowGroup.getDictionaryIndexes();
        for (int position = 0; position < rowGroup.getValueCount(); position++) {
            presentStream.writeBoolean(dictionaryIndexes.get(position) != 0);
        }
        for (int position = 0; position < rowGroup.getValueCount(); position++) {
            int originalDictionaryIndex = dictionaryIndexes.get(position);
            // index zero in original dictionary is reserved for null
            if (originalDictionaryIndex != 0) {
                int sortedIndex = originalDictionaryToSortedIndex[originalDictionaryIndex];
                if (sortedIndex < 0) {
                    throw new IllegalArgumentException();
                }
                dataStream.writeLong(sortedIndex);
            }
        }
        presentStream.recordCheckpoint();
        dataStream.recordCheckpoint();
    }
    // free the dictionary memory
    dictionary.clear();
    dictionaryDataStream.close();
    dictionaryLengthStream.close();
    dataStream.close();
    presentStream.close();
}

Example 3 with ColumnEncoding

use of io.prestosql.orc.metadata.ColumnEncoding in project hetu-core by openlookeng.

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<OrcDataOutput> bufferStripeData(long stripeStartOffset, FlushReason flushReason) throws IOException {
    if (stripeRowCount == 0) {
        verify(flushReason == CLOSED, "An empty stripe is not allowed");
        // column writers must be closed or the reset call will fail
        columnWriters.forEach(ColumnWriter::close);
        return ImmutableList.of();
    }
    if (rowGroupRowCount > 0) {
        finishRowGroup();
    }
    // convert any dictionary encoded column with a low compression ratio to direct
    dictionaryCompressionOptimizer.finalOptimize(bufferedBytes);
    columnWriters.forEach(ColumnWriter::close);
    List<OrcDataOutput> outputData = new ArrayList<>();
    List<Stream> allStreams = new ArrayList<>(columnWriters.size() * 3);
    // get index streams
    long indexLength = 0;
    for (ColumnWriter columnWriter : columnWriters) {
        for (StreamDataOutput indexStream : columnWriter.getIndexStreams(metadataWriter)) {
            // The ordering is critical because the stream only contain a length with no offset.
            outputData.add(indexStream);
            allStreams.add(indexStream.getStream());
            indexLength += indexStream.size();
        }
    }
    // 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();
    }
    Collections.sort(dataStreams);
    // add data streams
    for (StreamDataOutput dataStream : dataStreams) {
        // The ordering is critical because the stream only contain a length with no offset.
        outputData.add(dataStream);
        allStreams.add(dataStream.getStream());
    }
    Map<OrcColumnId, ColumnEncoding> columnEncodings = new HashMap<>();
    columnWriters.forEach(columnWriter -> columnEncodings.putAll(columnWriter.getColumnEncodings()));
    Map<OrcColumnId, ColumnStatistics> columnStatistics = new HashMap<>();
    columnWriters.forEach(columnWriter -> columnStatistics.putAll(columnWriter.getColumnStripeStatistics()));
    // the 0th column is a struct column for the whole row
    columnEncodings.put(ROOT_COLUMN, new ColumnEncoding(DIRECT, 0));
    columnStatistics.put(ROOT_COLUMN, new ColumnStatistics((long) stripeRowCount, 0, null, null, null, null, null, null, null, null));
    // add footer
    StripeFooter stripeFooter = new StripeFooter(allStreams, toColumnMetadata(columnEncodings, orcTypes.size()), ZoneId.of("UTC"));
    Slice footer = metadataWriter.writeStripeFooter(stripeFooter);
    outputData.add(createDataOutput(footer));
    // create final stripe statistics
    StripeStatistics statistics = new StripeStatistics(toColumnMetadata(columnStatistics, orcTypes.size()));
    recordValidation(validation -> validation.addStripeStatistics(stripeStartOffset, statistics));
    StripeInformation stripeInformation = new StripeInformation(stripeRowCount, stripeStartOffset, indexLength, dataLength, footer.length());
    ClosedStripe closedStripe = new ClosedStripe(stripeInformation, statistics);
    closedStripes.add(closedStripe);
    closedStripesRetainedBytes += closedStripe.getRetainedSizeInBytes();
    recordValidation(validation -> validation.addStripe(stripeInformation.getNumberOfRows()));
    stats.recordStripeWritten(flushReason, stripeInformation.getTotalLength(), stripeInformation.getNumberOfRows(), dictionaryCompressionOptimizer.getDictionaryMemoryBytes());
    return outputData;
}

Example 4 with ColumnEncoding

use of io.prestosql.orc.metadata.ColumnEncoding in project hetu-core by openlookeng.

the class StripeReader method readStripe.

public Stripe readStripe(StripeInformation stripe, AggregatedMemoryContext systemMemoryUsage) throws IOException {
    // read the stripe footer
    OrcStripeFooterCacheKey cacheKey = new OrcStripeFooterCacheKey();
    cacheKey.setOrcDataSourceId(new OrcDataSourceIdWithTimeStamp(orcDataSource.getId(), orcDataSource.getLastModifiedTime()));
    cacheKey.setStripeOffset(stripe.getOffset());
    StripeFooter stripeFooter;
    if (orcCacheProperties.isStripeFooterCacheEnabled()) {
        try {
            stripeFooter = orcCacheStore.getStripeFooterCache().get(cacheKey, () -> this.readStripeFooter(stripe, systemMemoryUsage));
        } catch (UncheckedExecutionException | ExecutionException executionException) {
            handleCacheLoadException(executionException);
            log.debug(executionException.getCause(), "Error while caching ORC stripe footer. Falling back to default flow");
            stripeFooter = readStripeFooter(stripe, systemMemoryUsage);
        }
    } else {
        stripeFooter = readStripeFooter(stripe, systemMemoryUsage);
    }
    ColumnMetadata<ColumnEncoding> columnEncodings = stripeFooter.getColumnEncodings();
    if (writeValidation.isPresent()) {
        writeValidation.get().validateTimeZone(orcDataSource.getId(), stripeFooter.getTimeZone());
    }
    ZoneId fileTimeZone = stripeFooter.getTimeZone();
    // get streams for selected columns
    Map<StreamId, Stream> streams = new HashMap<>();
    for (Stream stream : stripeFooter.getStreams()) {
        if (includedOrcColumnIds.contains(stream.getColumnId()) && isSupportedStreamType(stream, types.get(stream.getColumnId()).getOrcTypeKind())) {
            streams.put(new StreamId(stream), stream);
        }
    }
    // handle stripes with more than one row group
    boolean invalidCheckPoint = false;
    if (stripe.getNumberOfRows() > rowsInRowGroup) {
        // determine ranges of the stripe to read
        Map<StreamId, DiskRange> diskRanges = getDiskRanges(stripeFooter.getStreams());
        diskRanges = Maps.filterKeys(diskRanges, Predicates.in(streams.keySet()));
        // read the file regions
        Map<StreamId, OrcChunkLoader> streamsData = readDiskRanges(stripe.getOffset(), diskRanges, systemMemoryUsage);
        // read the bloom filter for each column
        Map<OrcColumnId, List<HashableBloomFilter>> bloomFilterIndexes = readBloomFilterIndexes(streams, streamsData, stripe);
        // read the row index for each column
        Map<StreamId, List<RowGroupIndex>> columnIndexes = readColumnIndexes(streams, streamsData, bloomFilterIndexes, stripe);
        if (writeValidation.isPresent()) {
            writeValidation.get().validateRowGroupStatistics(orcDataSource.getId(), stripe.getOffset(), columnIndexes);
        }
        // select the row groups matching the tuple domain
        Set<Integer> selectedRowGroups = selectRowGroups(stripe, columnIndexes);
        // if all row groups are skipped, return null
        if (selectedRowGroups.isEmpty()) {
            // set accounted memory usage to zero
            systemMemoryUsage.close();
            return null;
        }
        // value streams
        Map<StreamId, ValueInputStream<?>> valueStreams = createValueStreams(streams, streamsData, columnEncodings);
        // build the dictionary streams
        InputStreamSources dictionaryStreamSources = createDictionaryStreamSources(streams, valueStreams, columnEncodings);
        // build the row groups
        try {
            List<RowGroup> rowGroups = createRowGroups(stripe.getNumberOfRows(), streams, valueStreams, columnIndexes, selectedRowGroups, columnEncodings);
            return new Stripe(stripe.getNumberOfRows(), fileTimeZone, columnEncodings, rowGroups, dictionaryStreamSources);
        } catch (InvalidCheckpointException e) {
            // The ORC file contains a corrupt checkpoint stream treat the stripe as a single row group.
            invalidCheckPoint = true;
        }
    }
    // stripe only has one row group
    ImmutableMap.Builder<StreamId, DiskRange> diskRangesBuilder = ImmutableMap.builder();
    for (Entry<StreamId, DiskRange> entry : getDiskRanges(stripeFooter.getStreams()).entrySet()) {
        StreamId streamId = entry.getKey();
        if (streams.containsKey(streamId)) {
            diskRangesBuilder.put(entry);
        }
    }
    ImmutableMap<StreamId, DiskRange> diskRanges = diskRangesBuilder.build();
    // read the file regions
    Map<StreamId, OrcChunkLoader> streamsData = readDiskRanges(stripe.getOffset(), diskRanges, systemMemoryUsage);
    long minAverageRowBytes = 0;
    for (Entry<StreamId, Stream> entry : streams.entrySet()) {
        if (entry.getKey().getStreamKind() == ROW_INDEX) {
            List<RowGroupIndex> rowGroupIndexes;
            if (orcCacheProperties.isRowIndexCacheEnabled()) {
                OrcRowIndexCacheKey indexCacheKey = new OrcRowIndexCacheKey();
                indexCacheKey.setOrcDataSourceId(new OrcDataSourceIdWithTimeStamp(orcDataSource.getId(), orcDataSource.getLastModifiedTime()));
                indexCacheKey.setStripeOffset(stripe.getOffset());
                indexCacheKey.setStreamId(entry.getKey());
                try {
                    rowGroupIndexes = orcCacheStore.getRowIndexCache().get(indexCacheKey, () -> metadataReader.readRowIndexes(hiveWriterVersion, new OrcInputStream(streamsData.get(entry.getKey()))));
                } catch (UncheckedExecutionException | ExecutionException executionException) {
                    handleCacheLoadException(executionException);
                    log.debug(executionException.getCause(), "Error while caching row group indexes. Falling back to default flow");
                    rowGroupIndexes = metadataReader.readRowIndexes(hiveWriterVersion, new OrcInputStream(streamsData.get(entry.getKey())));
                }
            } else {
                rowGroupIndexes = metadataReader.readRowIndexes(hiveWriterVersion, new OrcInputStream(streamsData.get(entry.getKey())));
            }
            checkState(rowGroupIndexes.size() == 1 || invalidCheckPoint, "expect a single row group or an invalid check point");
            long totalBytes = 0;
            long totalRows = 0;
            for (RowGroupIndex rowGroupIndex : rowGroupIndexes) {
                ColumnStatistics columnStatistics = rowGroupIndex.getColumnStatistics();
                if (columnStatistics.hasMinAverageValueSizeInBytes()) {
                    totalBytes += columnStatistics.getMinAverageValueSizeInBytes() * columnStatistics.getNumberOfValues();
                    totalRows += columnStatistics.getNumberOfValues();
                }
            }
            if (totalRows > 0) {
                minAverageRowBytes += totalBytes / totalRows;
            }
        }
    }
    // value streams
    Map<StreamId, ValueInputStream<?>> valueStreams = createValueStreams(streams, streamsData, columnEncodings);
    // build the dictionary streams
    InputStreamSources dictionaryStreamSources = createDictionaryStreamSources(streams, valueStreams, columnEncodings);
    // build the row group
    ImmutableMap.Builder<StreamId, InputStreamSource<?>> builder = ImmutableMap.builder();
    for (Entry<StreamId, ValueInputStream<?>> entry : valueStreams.entrySet()) {
        builder.put(entry.getKey(), new ValueInputStreamSource<>(entry.getValue()));
    }
    RowGroup rowGroup = new RowGroup(0, 0, stripe.getNumberOfRows(), minAverageRowBytes, new InputStreamSources(builder.build()));
    return new Stripe(stripe.getNumberOfRows(), fileTimeZone, columnEncodings, ImmutableList.of(rowGroup), dictionaryStreamSources);
}

Example 5 with ColumnEncoding

use of io.prestosql.orc.metadata.ColumnEncoding in project hetu-core by openlookeng.

the class StripeReader method createRowGroups.

private List<RowGroup> createRowGroups(int rowsInStripe, Map<StreamId, Stream> streams, Map<StreamId, ValueInputStream<?>> valueStreams, Map<StreamId, List<RowGroupIndex>> columnIndexes, Set<Integer> selectedRowGroups, ColumnMetadata<ColumnEncoding> encodings) throws InvalidCheckpointException {
    ImmutableList.Builder<RowGroup> rowGroupBuilder = ImmutableList.builder();
    for (int rowGroupId : selectedRowGroups) {
        Map<StreamId, StreamCheckpoint> checkpoints = getStreamCheckpoints(includedOrcColumnIds, types, decompressor.isPresent(), rowGroupId, encodings, streams, columnIndexes);
        int rowOffset = rowGroupId * rowsInRowGroup;
        int rowsInGroup = Math.min(rowsInStripe - rowOffset, rowsInRowGroup);
        long minAverageRowBytes = columnIndexes.entrySet().stream().mapToLong(e -> e.getValue().get(rowGroupId).getColumnStatistics().getMinAverageValueSizeInBytes()).sum();
        rowGroupBuilder.add(createRowGroup(rowGroupId, rowOffset, rowsInGroup, minAverageRowBytes, valueStreams, checkpoints));
    }
    return rowGroupBuilder.build();
}