Search in sources :

Example 1 with ValueInputStreamSource

use of io.trino.orc.stream.ValueInputStreamSource 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);
}
Also used : ValueInputStream(io.trino.orc.stream.ValueInputStream) OrcColumnId(io.trino.orc.metadata.OrcColumnId) HashMap(java.util.HashMap) InvalidCheckpointException(io.trino.orc.checkpoint.InvalidCheckpointException) ValueInputStreamSource(io.trino.orc.stream.ValueInputStreamSource) InputStreamSource(io.trino.orc.stream.InputStreamSource) Stream(io.trino.orc.metadata.Stream) OrcInputStream(io.trino.orc.stream.OrcInputStream) ValueInputStream(io.trino.orc.stream.ValueInputStream) InputStream(java.io.InputStream) List(java.util.List) ArrayList(java.util.ArrayList) ImmutableList(com.google.common.collect.ImmutableList) ColumnStatistics(io.trino.orc.metadata.statistics.ColumnStatistics) OrcInputStream(io.trino.orc.stream.OrcInputStream) ZoneId(java.time.ZoneId) OrcChunkLoader(io.trino.orc.stream.OrcChunkLoader) ImmutableMap(com.google.common.collect.ImmutableMap) ImmutableMap.toImmutableMap(com.google.common.collect.ImmutableMap.toImmutableMap) ColumnEncoding(io.trino.orc.metadata.ColumnEncoding) InputStreamSources(io.trino.orc.stream.InputStreamSources) StripeFooter(io.trino.orc.metadata.StripeFooter) RowGroupIndex(io.trino.orc.metadata.RowGroupIndex)

Aggregations

ImmutableList (com.google.common.collect.ImmutableList)1 ImmutableMap (com.google.common.collect.ImmutableMap)1 ImmutableMap.toImmutableMap (com.google.common.collect.ImmutableMap.toImmutableMap)1 InvalidCheckpointException (io.trino.orc.checkpoint.InvalidCheckpointException)1 ColumnEncoding (io.trino.orc.metadata.ColumnEncoding)1 OrcColumnId (io.trino.orc.metadata.OrcColumnId)1 RowGroupIndex (io.trino.orc.metadata.RowGroupIndex)1 Stream (io.trino.orc.metadata.Stream)1 StripeFooter (io.trino.orc.metadata.StripeFooter)1 ColumnStatistics (io.trino.orc.metadata.statistics.ColumnStatistics)1 InputStreamSource (io.trino.orc.stream.InputStreamSource)1 InputStreamSources (io.trino.orc.stream.InputStreamSources)1 OrcChunkLoader (io.trino.orc.stream.OrcChunkLoader)1 OrcInputStream (io.trino.orc.stream.OrcInputStream)1 ValueInputStream (io.trino.orc.stream.ValueInputStream)1 ValueInputStreamSource (io.trino.orc.stream.ValueInputStreamSource)1 InputStream (java.io.InputStream)1 ZoneId (java.time.ZoneId)1 ArrayList (java.util.ArrayList)1 HashMap (java.util.HashMap)1