Examples with ColumnStatistics io.trino.orc.metadata.statistics.ColumnStatistics used on opensource projects

Example 11 with ColumnStatistics

use of io.trino.orc.metadata.statistics.ColumnStatistics in project trino by trinodb.

the class StructColumnWriter method finishRowGroup.

@Override
public Map<OrcColumnId, ColumnStatistics> finishRowGroup() {
    checkState(!closed);
    ColumnStatistics statistics = new ColumnStatistics((long) nonNullValueCount, 0, null, null, null, null, null, null, null, null, null);
    rowGroupColumnStatistics.add(statistics);
    nonNullValueCount = 0;
    ImmutableMap.Builder<OrcColumnId, ColumnStatistics> columnStatistics = ImmutableMap.builder();
    columnStatistics.put(columnId, statistics);
    structFields.stream().map(ColumnWriter::finishRowGroup).forEach(columnStatistics::putAll);
    return columnStatistics.buildOrThrow();
}

Example 12 with ColumnStatistics

use of io.trino.orc.metadata.statistics.ColumnStatistics in project trino by trinodb.

the class TimestampColumnWriter method finishRowGroup.

@Override
public Map<OrcColumnId, ColumnStatistics> finishRowGroup() {
    checkState(!closed);
    ColumnStatistics statistics = statisticsBuilder.buildColumnStatistics();
    rowGroupColumnStatistics.add(statistics);
    statisticsBuilder = statisticsBuilderSupplier.get();
    return ImmutableMap.of(columnId, statistics);
}

Example 13 with ColumnStatistics

use of io.trino.orc.metadata.statistics.ColumnStatistics in project trino by trinodb.

the class TestTupleDomainOrcPredicate method stringColumnStats.

private static ColumnStatistics stringColumnStats(Long numberOfValues, String minimum, String maximum) {
    Slice minimumSlice = minimum == null ? null : utf8Slice(minimum);
    Slice maximumSlice = maximum == null ? null : utf8Slice(maximum);
    // sum and minAverageValueSizeInBytes are not used in this test; they could be arbitrary numbers
    return new ColumnStatistics(numberOfValues, 10L, null, null, null, new StringStatistics(minimumSlice, maximumSlice, 100L), null, null, null, null, null);
}

Example 14 with ColumnStatistics

use of io.trino.orc.metadata.statistics.ColumnStatistics 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 15 with ColumnStatistics

use of io.trino.orc.metadata.statistics.ColumnStatistics in project trino by trinodb.

the class StripeReader method getRowGroupStatistics.

private static ColumnMetadata<ColumnStatistics> getRowGroupStatistics(ColumnMetadata<OrcType> types, Map<StreamId, List<RowGroupIndex>> columnIndexes, int rowGroup) {
    requireNonNull(columnIndexes, "columnIndexes is null");
    checkArgument(rowGroup >= 0, "rowGroup is negative");
    Map<Integer, List<RowGroupIndex>> rowGroupIndexesByColumn = columnIndexes.entrySet().stream().collect(toImmutableMap(entry -> entry.getKey().getColumnId().getId(), Entry::getValue));
    List<ColumnStatistics> statistics = new ArrayList<>(types.size());
    for (int columnIndex = 0; columnIndex < types.size(); columnIndex++) {
        List<RowGroupIndex> rowGroupIndexes = rowGroupIndexesByColumn.get(columnIndex);
        if (rowGroupIndexes != null) {
            statistics.add(rowGroupIndexes.get(rowGroup).getColumnStatistics());
        } else {
            statistics.add(null);
        }
    }
    return new ColumnMetadata<>(statistics);
}