Examples with InputStreamSources com.facebook.presto.orc.stream.InputStreamSources used on opensource projects

Example 1 with InputStreamSources

use of com.facebook.presto.orc.stream.InputStreamSources in project presto by prestodb.

the class TestLongDictionaryProvider method createLongDictionaryStreamSources.

private InputStreamSources createLongDictionaryStreamSources(Map<NodeId, long[]> streams, OrcAggregatedMemoryContext aggregatedMemoryContext) {
    SharedBuffer decompressionBuffer = new SharedBuffer(aggregatedMemoryContext.newOrcLocalMemoryContext("sharedDecompressionBuffer"));
    ImmutableMap.Builder<StreamId, InputStreamSource<?>> dictionaryStreamsBuilder = ImmutableMap.builder();
    for (Map.Entry<NodeId, long[]> entry : streams.entrySet()) {
        StreamId streamId = entry.getKey().toDictionaryDataStreamId();
        DynamicSliceOutput sliceOutput = createSliceOutput(streamId, entry.getValue());
        ValueInputStream<?> valueStream = createValueStream(sliceOutput.slice(), aggregatedMemoryContext, decompressionBuffer);
        StreamCheckpoint streamCheckpoint = getDictionaryStreamCheckpoint(streamId, LONG, ColumnEncoding.ColumnEncodingKind.DICTIONARY);
        InputStreamSource<?> streamSource = createCheckpointStreamSource(valueStream, streamCheckpoint);
        dictionaryStreamsBuilder.put(streamId, streamSource);
    }
    return new InputStreamSources(dictionaryStreamsBuilder.build());
}

Example 2 with InputStreamSources

use of com.facebook.presto.orc.stream.InputStreamSources in project presto by prestodb.

the class StripeReader method readStripe.

public Stripe readStripe(StripeInformation stripe, OrcAggregatedMemoryContext systemMemoryUsage, Optional<DwrfEncryptionInfo> decryptors, SharedBuffer sharedDecompressionBuffer) throws IOException {
    StripeId stripeId = new StripeId(orcDataSource.getId(), stripe.getOffset());
    // read the stripe footer
    StripeFooter stripeFooter = readStripeFooter(stripeId, stripe, systemMemoryUsage);
    // get streams for selected columns
    List<List<Stream>> allStreams = new ArrayList<>();
    allStreams.add(stripeFooter.getStreams());
    Map<StreamId, Stream> includedStreams = new HashMap<>();
    boolean hasRowGroupDictionary = addIncludedStreams(stripeFooter.getColumnEncodings(), stripeFooter.getStreams(), includedStreams);
    Map<Integer, ColumnEncoding> columnEncodings = new HashMap<>();
    Map<Integer, ColumnEncoding> stripeFooterEncodings = stripeFooter.getColumnEncodings();
    columnEncodings.putAll(stripeFooterEncodings);
    // included columns may be encrypted
    if (decryptors.isPresent()) {
        List<Slice> encryptedEncryptionGroups = stripeFooter.getStripeEncryptionGroups();
        for (Integer groupId : decryptors.get().getEncryptorGroupIds()) {
            StripeEncryptionGroup stripeEncryptionGroup = getStripeEncryptionGroup(decryptors.get().getEncryptorByGroupId(groupId), encryptedEncryptionGroups.get(groupId), dwrfEncryptionGroupColumns.get(groupId), systemMemoryUsage);
            allStreams.add(stripeEncryptionGroup.getStreams());
            columnEncodings.putAll(stripeEncryptionGroup.getColumnEncodings());
            boolean encryptedHasRowGroupDictionary = addIncludedStreams(stripeEncryptionGroup.getColumnEncodings(), stripeEncryptionGroup.getStreams(), includedStreams);
            hasRowGroupDictionary = encryptedHasRowGroupDictionary || hasRowGroupDictionary;
        }
    }
    // handle stripes with more than one row group or a dictionary
    boolean invalidCheckPoint = false;
    if ((stripe.getNumberOfRows() > rowsInRowGroup) || hasRowGroupDictionary) {
        // determine ranges of the stripe to read
        Map<StreamId, DiskRange> diskRanges = getDiskRanges(allStreams);
        diskRanges = Maps.filterKeys(diskRanges, Predicates.in(includedStreams.keySet()));
        // read the file regions
        Map<StreamId, OrcInputStream> streamsData = readDiskRanges(stripeId, diskRanges, systemMemoryUsage, decryptors, sharedDecompressionBuffer);
        // read the row index for each column
        Map<StreamId, List<RowGroupIndex>> columnIndexes = readColumnIndexes(includedStreams, streamsData, stripeId);
        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(includedStreams, streamsData, columnEncodings);
        // build the dictionary streams
        InputStreamSources dictionaryStreamSources = createDictionaryStreamSources(includedStreams, valueStreams, columnEncodings);
        // build the row groups
        try {
            List<RowGroup> rowGroups = createRowGroups(stripe.getNumberOfRows(), includedStreams, valueStreams, columnIndexes, selectedRowGroups, columnEncodings);
            return new Stripe(stripe.getNumberOfRows(), columnEncodings, rowGroups, dictionaryStreamSources);
        } catch (InvalidCheckpointException e) {
            // we must fail because the length of the row group dictionary is contained in the checkpoint stream.
            if (hasRowGroupDictionary) {
                throw new OrcCorruptionException(e, orcDataSource.getId(), "Checkpoints are corrupt");
            }
            invalidCheckPoint = true;
        }
    }
    // stripe only has one row group and no dictionary
    ImmutableMap.Builder<StreamId, DiskRange> diskRangesBuilder = ImmutableMap.builder();
    for (Entry<StreamId, DiskRange> entry : getDiskRanges(allStreams).entrySet()) {
        StreamId streamId = entry.getKey();
        if (includedStreams.keySet().contains(streamId)) {
            diskRangesBuilder.put(entry);
        }
    }
    ImmutableMap<StreamId, DiskRange> diskRanges = diskRangesBuilder.build();
    // read the file regions
    Map<StreamId, OrcInputStream> streamsData = readDiskRanges(stripeId, diskRanges, systemMemoryUsage, decryptors, sharedDecompressionBuffer);
    long totalBytes = 0;
    for (Entry<StreamId, Stream> entry : includedStreams.entrySet()) {
        if (entry.getKey().getStreamKind() == ROW_INDEX) {
            List<RowGroupIndex> rowGroupIndexes = metadataReader.readRowIndexes(hiveWriterVersion, streamsData.get(entry.getKey()), null);
            checkState(rowGroupIndexes.size() == 1 || invalidCheckPoint, "expect a single row group or an invalid check point");
            for (RowGroupIndex rowGroupIndex : rowGroupIndexes) {
                ColumnStatistics columnStatistics = rowGroupIndex.getColumnStatistics();
                if (columnStatistics.hasMinAverageValueSizeInBytes()) {
                    totalBytes += columnStatistics.getTotalValueSizeInBytes();
                }
            }
        }
    }
    // value streams
    Map<StreamId, ValueInputStream<?>> valueStreams = createValueStreams(includedStreams, streamsData, columnEncodings);
    // build the dictionary streams
    InputStreamSources dictionaryStreamSources = createDictionaryStreamSources(includedStreams, 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(), totalBytes, new InputStreamSources(builder.build()));
    return new Stripe(stripe.getNumberOfRows(), columnEncodings, ImmutableList.of(rowGroup), dictionaryStreamSources);
}

Example 3 with InputStreamSources

use of com.facebook.presto.orc.stream.InputStreamSources in project presto by prestodb.

the class AbstractOrcRecordReader method advanceToNextRowGroup.

private boolean advanceToNextRowGroup() throws IOException {
    nextRowInGroup = 0;
    if (currentRowGroup >= 0) {
        if (rowGroupStatisticsValidation.isPresent()) {
            OrcWriteValidation.StatisticsValidation statisticsValidation = rowGroupStatisticsValidation.get();
            long offset = stripes.get(currentStripe).getOffset();
            writeValidation.get().validateRowGroupStatistics(orcDataSource.getId(), offset, currentRowGroup, statisticsValidation.build());
            statisticsValidation.reset();
        }
    }
    while (!rowGroups.hasNext() && currentStripe < stripes.size()) {
        advanceToNextStripe();
        currentRowGroup = -1;
    }
    if (!rowGroups.hasNext()) {
        currentGroupRowCount = 0;
        return false;
    }
    currentRowGroup++;
    RowGroup currentRowGroup = rowGroups.next();
    currentGroupRowCount = toIntExact(currentRowGroup.getRowCount());
    if (currentRowGroup.getMinAverageRowBytes() > 0) {
        maxBatchSize = adjustMaxBatchSize(maxBatchSize, maxBlockBytes, currentRowGroup.getMinAverageRowBytes());
    }
    currentPosition = currentStripePosition + currentRowGroup.getRowOffset();
    filePosition = stripeFilePositions.get(currentStripe) + currentRowGroup.getRowOffset();
    // give reader data streams from row group
    InputStreamSources rowGroupStreamSources = currentRowGroup.getStreamSources();
    for (StreamReader column : streamReaders) {
        if (column != null) {
            column.startRowGroup(rowGroupStreamSources);
        }
    }
    return true;
}

Example 4 with InputStreamSources

use of com.facebook.presto.orc.stream.InputStreamSources in project presto by prestodb.

the class StripeReader method createDictionaryStreamSources.

public InputStreamSources createDictionaryStreamSources(Map<StreamId, Stream> streams, Map<StreamId, ValueInputStream<?>> valueStreams, Map<Integer, ColumnEncoding> columnEncodings) {
    ImmutableMap.Builder<StreamId, InputStreamSource<?>> dictionaryStreamBuilder = ImmutableMap.builder();
    for (Entry<StreamId, Stream> entry : streams.entrySet()) {
        StreamId streamId = entry.getKey();
        Stream stream = entry.getValue();
        int column = stream.getColumn();
        // only process dictionary streams
        ColumnEncodingKind columnEncoding = columnEncodings.get(column).getColumnEncoding(stream.getSequence()).getColumnEncodingKind();
        if (!isDictionary(stream, columnEncoding)) {
            continue;
        }
        // skip streams without data
        ValueInputStream<?> valueStream = valueStreams.get(streamId);
        if (valueStream == null) {
            continue;
        }
        OrcTypeKind columnType = types.get(stream.getColumn()).getOrcTypeKind();
        StreamCheckpoint streamCheckpoint = getDictionaryStreamCheckpoint(streamId, columnType, columnEncoding);
        InputStreamSource<?> streamSource = createCheckpointStreamSource(valueStream, streamCheckpoint);
        dictionaryStreamBuilder.put(streamId, streamSource);
    }
    return new InputStreamSources(dictionaryStreamBuilder.build());
}

Example 5 with InputStreamSources

use of com.facebook.presto.orc.stream.InputStreamSources in project presto by prestodb.

the class StripeReader method createRowGroup.

@VisibleForTesting
static RowGroup createRowGroup(int groupId, long rowsInStripe, long rowsInRowGroup, Map<StreamId, List<RowGroupIndex>> columnIndexes, Map<StreamId, ValueInputStream<?>> valueStreams, Map<StreamId, StreamCheckpoint> checkpoints) {
    long totalRowGroupBytes = columnIndexes.values().stream().mapToLong(e -> e.get(groupId).getColumnStatistics().getTotalValueSizeInBytes()).sum();
    long rowOffset = multiplyExact(groupId, rowsInRowGroup);
    int rowCount = toIntExact(Math.min(rowsInStripe - rowOffset, rowsInRowGroup));
    ImmutableMap.Builder<StreamId, InputStreamSource<?>> builder = ImmutableMap.builder();
    for (Entry<StreamId, StreamCheckpoint> entry : checkpoints.entrySet()) {
        StreamId streamId = entry.getKey();
        StreamCheckpoint checkpoint = entry.getValue();
        // skip streams without data
        ValueInputStream<?> valueStream = valueStreams.get(streamId);
        if (valueStream == null) {
            continue;
        }
        builder.put(streamId, createCheckpointStreamSource(valueStream, checkpoint));
    }
    InputStreamSources rowGroupStreams = new InputStreamSources(builder.build());
    return new RowGroup(groupId, rowOffset, rowCount, totalRowGroupBytes, rowGroupStreams);
}