Examples with InputStreamSources io.trino.orc.stream.InputStreamSources used on opensource projects

Example 1 with InputStreamSources

use of io.trino.orc.stream.InputStreamSources in project trino by trinodb.

the class StripeReader method readStripe.

public Stripe readStripe(StripeInformation stripe, AggregatedMemoryContext memoryUsage) throws IOException {
    // read the stripe footer
    StripeFooter stripeFooter = readStripeFooter(stripe, memoryUsage);
    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 (rowsInRowGroup.isPresent() && stripe.getNumberOfRows() > rowsInRowGroup.getAsInt()) {
        // 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, memoryUsage);
        // read the bloom filter for each column
        Map<OrcColumnId, List<BloomFilter>> bloomFilterIndexes = readBloomFilterIndexes(streams, streamsData);
        // read the row index for each column
        Map<StreamId, List<RowGroupIndex>> columnIndexes = readColumnIndexes(streams, streamsData, bloomFilterIndexes);
        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
            memoryUsage.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.buildOrThrow();
    // read the file regions
    Map<StreamId, OrcChunkLoader> streamsData = readDiskRanges(stripe.getOffset(), diskRanges, memoryUsage);
    long minAverageRowBytes = 0;
    for (Entry<StreamId, Stream> entry : streams.entrySet()) {
        if (entry.getKey().getStreamKind() == ROW_INDEX) {
            List<RowGroupIndex> 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.buildOrThrow()));
    return new Stripe(stripe.getNumberOfRows(), fileTimeZone, columnEncodings, ImmutableList.of(rowGroup), dictionaryStreamSources);
}

Example 2 with InputStreamSources

use of io.trino.orc.stream.InputStreamSources in project trino by trinodb.

the class OrcRecordReader method advanceToNextStripe.

private void advanceToNextStripe() throws IOException {
    currentStripeMemoryContext.close();
    currentStripeMemoryContext = memoryUsage.newAggregatedMemoryContext();
    rowGroups = ImmutableList.<RowGroup>of().iterator();
    if (currentStripe >= 0) {
        if (stripeStatisticsValidation.isPresent()) {
            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, currentStripeMemoryContext);
    if (stripe != null) {
        // Give readers access to dictionary streams
        InputStreamSources dictionaryStreamSources = stripe.getDictionaryStreamSources();
        ColumnMetadata<ColumnEncoding> columnEncodings = stripe.getColumnEncodings();
        ZoneId fileTimeZone = stripe.getFileTimeZone();
        for (ColumnReader column : columnReaders) {
            if (column != null) {
                column.startStripe(fileTimeZone, dictionaryStreamSources, columnEncodings);
            }
        }
        rowGroups = stripe.getRowGroups().iterator();
    }
    orcDataSourceMemoryUsage.setBytes(orcDataSource.getRetainedSize());
}

Example 3 with InputStreamSources

use of io.trino.orc.stream.InputStreamSources in project trino by trinodb.

the class OrcRecordReader method advanceToNextRowGroup.

private boolean advanceToNextRowGroup() throws IOException {
    nextRowInGroup = 0;
    if (currentRowGroup >= 0) {
        if (rowGroupStatisticsValidation.isPresent()) {
            StatisticsValidation statisticsValidation = rowGroupStatisticsValidation.get();
            long offset = stripes.get(currentStripe).getOffset();
            writeValidation.get().validateRowGroupStatistics(orcDataSource.getId(), offset, currentRowGroup, statisticsValidation.build().get());
            statisticsValidation.reset();
        }
    }
    while (!rowGroups.hasNext() && currentStripe < stripes.size()) {
        advanceToNextStripe();
        currentRowGroup = -1;
    }
    if (!rowGroups.hasNext()) {
        currentGroupRowCount = 0;
        return false;
    }
    currentRowGroup++;
    RowGroup currentRowGroup = rowGroups.next();
    currentGroupRowCount = currentRowGroup.getRowCount();
    if (currentRowGroup.getMinAverageRowBytes() > 0) {
        maxBatchSize = toIntExact(min(maxBatchSize, max(1, 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 (ColumnReader column : columnReaders) {
        if (column != null) {
            column.startRowGroup(rowGroupStreamSources);
        }
    }
    return true;
}

Example 4 with InputStreamSources

use of io.trino.orc.stream.InputStreamSources in project trino by trinodb.

the class StripeReader method createDictionaryStreamSources.

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

Example 5 with InputStreamSources

use of io.trino.orc.stream.InputStreamSources in project trino by trinodb.

the class StripeReader method createRowGroup.

private static RowGroup createRowGroup(int groupId, int rowOffset, int rowCount, long minAverageRowBytes, Map<StreamId, ValueInputStream<?>> valueStreams, Map<StreamId, StreamCheckpoint> checkpoints) {
    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.buildOrThrow());
    return new RowGroup(groupId, rowOffset, rowCount, minAverageRowBytes, rowGroupStreams);
}