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

Example 21 with Stream

use of io.prestosql.orc.metadata.Stream 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 22 with Stream

use of io.prestosql.orc.metadata.Stream 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();
}

Example 23 with Stream

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

the class Checkpoints method getStreamCheckpoints.

public static Map<StreamId, StreamCheckpoint> getStreamCheckpoints(Set<OrcColumnId> columns, ColumnMetadata<OrcType> columnTypes, boolean compressed, int rowGroupId, ColumnMetadata<ColumnEncoding> columnEncodings, Map<StreamId, Stream> streams, Map<StreamId, List<RowGroupIndex>> columnIndexes) throws InvalidCheckpointException {
    ImmutableSetMultimap.Builder<OrcColumnId, StreamKind> streamKindsBuilder = ImmutableSetMultimap.builder();
    for (Stream stream : streams.values()) {
        streamKindsBuilder.put(stream.getColumnId(), stream.getStreamKind());
    }
    SetMultimap<OrcColumnId, StreamKind> streamKinds = streamKindsBuilder.build();
    ImmutableMap.Builder<StreamId, StreamCheckpoint> checkpoints = ImmutableMap.builder();
    for (Map.Entry<StreamId, List<RowGroupIndex>> entry : columnIndexes.entrySet()) {
        OrcColumnId columnId = entry.getKey().getColumnId();
        if (!columns.contains(columnId)) {
            continue;
        }
        List<Integer> positionsList = entry.getValue().get(rowGroupId).getPositions();
        ColumnEncodingKind columnEncoding = columnEncodings.get(columnId).getColumnEncodingKind();
        OrcTypeKind columnType = columnTypes.get(columnId).getOrcTypeKind();
        Set<StreamKind> availableStreams = streamKinds.get(columnId);
        ColumnPositionsList columnPositionsList = new ColumnPositionsList(columnId, columnType, positionsList);
        switch(columnType) {
            case BOOLEAN:
                checkpoints.putAll(getBooleanColumnCheckpoints(columnId, compressed, availableStreams, columnPositionsList));
                break;
            case BYTE:
                checkpoints.putAll(getByteColumnCheckpoints(columnId, compressed, availableStreams, columnPositionsList));
                break;
            case SHORT:
            case INT:
            case LONG:
            case DATE:
                checkpoints.putAll(getLongColumnCheckpoints(columnId, columnEncoding, compressed, availableStreams, columnPositionsList));
                break;
            case FLOAT:
                checkpoints.putAll(getFloatColumnCheckpoints(columnId, compressed, availableStreams, columnPositionsList));
                break;
            case DOUBLE:
                checkpoints.putAll(getDoubleColumnCheckpoints(columnId, compressed, availableStreams, columnPositionsList));
                break;
            case TIMESTAMP:
                checkpoints.putAll(getTimestampColumnCheckpoints(columnId, columnEncoding, compressed, availableStreams, columnPositionsList));
                break;
            case BINARY:
            case STRING:
            case VARCHAR:
            case CHAR:
                checkpoints.putAll(getSliceColumnCheckpoints(columnId, columnEncoding, compressed, availableStreams, columnPositionsList));
                break;
            case LIST:
            case MAP:
                checkpoints.putAll(getListOrMapColumnCheckpoints(columnId, columnEncoding, compressed, availableStreams, columnPositionsList));
                break;
            case STRUCT:
                checkpoints.putAll(getStructColumnCheckpoints(columnId, compressed, availableStreams, columnPositionsList));
                break;
            case DECIMAL:
                checkpoints.putAll(getDecimalColumnCheckpoints(columnId, columnEncoding, compressed, availableStreams, columnPositionsList));
                break;
            default:
                throw new IllegalArgumentException("Unsupported column type " + columnType);
        }
    }
    return checkpoints.build();
}

Example 24 with Stream

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

the class SliceDirectColumnWriter method getIndexStreams.

@Override
public List<StreamDataOutput> getIndexStreams(CompressedMetadataWriter metadataWriter) throws IOException {
    checkState(closed);
    ImmutableList.Builder<RowGroupIndex> rowGroupIndexes = ImmutableList.builder();
    List<LongStreamCheckpoint> lengthCheckpoints = lengthStream.getCheckpoints();
    List<ByteArrayStreamCheckpoint> dataCheckpoints = dataStream.getCheckpoints();
    Optional<List<BooleanStreamCheckpoint>> presentCheckpoints = presentStream.getCheckpoints();
    for (int i = 0; i < rowGroupColumnStatistics.size(); i++) {
        int groupId = i;
        ColumnStatistics columnStatistics = rowGroupColumnStatistics.get(groupId);
        LongStreamCheckpoint lengthCheckpoint = lengthCheckpoints.get(groupId);
        ByteArrayStreamCheckpoint dataCheckpoint = dataCheckpoints.get(groupId);
        Optional<BooleanStreamCheckpoint> presentCheckpoint = presentCheckpoints.map(checkpoints -> checkpoints.get(groupId));
        List<Integer> positions = createSliceColumnPositionList(compressed, lengthCheckpoint, dataCheckpoint, presentCheckpoint);
        rowGroupIndexes.add(new RowGroupIndex(positions, columnStatistics));
    }
    Slice slice = metadataWriter.writeRowIndexes(rowGroupIndexes.build());
    Stream stream = new Stream(columnId, StreamKind.ROW_INDEX, slice.length(), false);
    return ImmutableList.of(new StreamDataOutput(slice, stream));
}

Example 25 with Stream

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

the class StructColumnWriter method getIndexStreams.

@Override
public List<StreamDataOutput> getIndexStreams(CompressedMetadataWriter metadataWriter) throws IOException {
    checkState(closed);
    ImmutableList.Builder<RowGroupIndex> rowGroupIndexes = ImmutableList.builder();
    Optional<List<BooleanStreamCheckpoint>> presentCheckpoints = presentStream.getCheckpoints();
    for (int i = 0; i < rowGroupColumnStatistics.size(); i++) {
        int groupId = i;
        ColumnStatistics columnStatistics = rowGroupColumnStatistics.get(groupId);
        Optional<BooleanStreamCheckpoint> presentCheckpoint = presentCheckpoints.map(checkpoints -> checkpoints.get(groupId));
        List<Integer> positions = createStructColumnPositionList(compressed, presentCheckpoint);
        rowGroupIndexes.add(new RowGroupIndex(positions, columnStatistics));
    }
    Slice slice = metadataWriter.writeRowIndexes(rowGroupIndexes.build());
    Stream stream = new Stream(columnId, StreamKind.ROW_INDEX, slice.length(), false);
    ImmutableList.Builder<StreamDataOutput> indexStreams = ImmutableList.builder();
    indexStreams.add(new StreamDataOutput(slice, stream));
    for (ColumnWriter structField : structFields) {
        indexStreams.addAll(structField.getIndexStreams(metadataWriter));
    }
    return indexStreams.build();
}