Examples with DoubleRange io.trino.spi.statistics.DoubleRange used on opensource projects

Example 1 with DoubleRange

use of io.trino.spi.statistics.DoubleRange in project trino by trinodb.

the class HiveMetastoreBackedDeltaLakeMetastore method getTableStatistics.

@Override
public TableStatistics getTableStatistics(ConnectorSession session, DeltaLakeTableHandle tableHandle, Constraint constraint) {
    TableSnapshot tableSnapshot = getSnapshot(tableHandle.getSchemaTableName(), session);
    double numRecords = 0L;
    MetadataEntry metadata = transactionLogAccess.getMetadataEntry(tableSnapshot, session).orElseThrow(() -> new TrinoException(DELTA_LAKE_INVALID_SCHEMA, "Metadata not found in transaction log for " + tableHandle.getTableName()));
    List<ColumnMetadata> columnMetadata = DeltaLakeSchemaSupport.extractSchema(metadata, typeManager);
    List<DeltaLakeColumnHandle> columns = columnMetadata.stream().map(columnMeta -> new DeltaLakeColumnHandle(columnMeta.getName(), columnMeta.getType(), metadata.getCanonicalPartitionColumns().contains(columnMeta.getName()) ? PARTITION_KEY : REGULAR)).collect(toImmutableList());
    Map<DeltaLakeColumnHandle, Double> nullCounts = new HashMap<>();
    columns.forEach(column -> nullCounts.put(column, 0.0));
    Map<DeltaLakeColumnHandle, Double> minValues = new HashMap<>();
    Map<DeltaLakeColumnHandle, Double> maxValues = new HashMap<>();
    Map<DeltaLakeColumnHandle, Set<String>> partitioningColumnsDistinctValues = new HashMap<>();
    columns.stream().filter(column -> column.getColumnType() == PARTITION_KEY).forEach(column -> partitioningColumnsDistinctValues.put(column, new HashSet<>()));
    if (tableHandle.getEnforcedPartitionConstraint().isNone() || tableHandle.getNonPartitionConstraint().isNone() || constraint.getSummary().isNone()) {
        return createZeroStatistics(columns);
    }
    Set<String> predicatedColumnNames = tableHandle.getNonPartitionConstraint().getDomains().orElseThrow().keySet().stream().map(DeltaLakeColumnHandle::getName).collect(toImmutableSet());
    List<ColumnMetadata> predicatedColumns = columnMetadata.stream().filter(column -> predicatedColumnNames.contains(column.getName())).collect(toImmutableList());
    for (AddFileEntry addEntry : transactionLogAccess.getActiveFiles(tableSnapshot, session)) {
        Optional<? extends DeltaLakeFileStatistics> fileStatistics = addEntry.getStats();
        if (fileStatistics.isEmpty()) {
            // Open source Delta Lake does not collect stats
            return TableStatistics.empty();
        }
        DeltaLakeFileStatistics stats = fileStatistics.get();
        if (!partitionMatchesPredicate(addEntry.getCanonicalPartitionValues(), tableHandle.getEnforcedPartitionConstraint().getDomains().orElseThrow())) {
            continue;
        }
        TupleDomain<DeltaLakeColumnHandle> statisticsPredicate = createStatisticsPredicate(addEntry, predicatedColumns, tableHandle.getMetadataEntry().getCanonicalPartitionColumns());
        if (!tableHandle.getNonPartitionConstraint().overlaps(statisticsPredicate)) {
            continue;
        }
        if (stats.getNumRecords().isEmpty()) {
            // Not clear if it's possible for stats to be present with no row count, but bail out if that happens
            return TableStatistics.empty();
        }
        numRecords += stats.getNumRecords().get();
        for (DeltaLakeColumnHandle column : columns) {
            if (column.getColumnType() == PARTITION_KEY) {
                Optional<String> partitionValue = addEntry.getCanonicalPartitionValues().get(column.getName());
                if (partitionValue.isEmpty()) {
                    nullCounts.merge(column, (double) stats.getNumRecords().get(), Double::sum);
                } else {
                    // NULL is not counted as a distinct value
                    // Code below assumes that values returned by addEntry.getCanonicalPartitionValues() are normalized,
                    // it may not be true in case of real, doubles, timestamps etc
                    partitioningColumnsDistinctValues.get(column).add(partitionValue.get());
                }
            } else {
                Optional<Long> maybeNullCount = stats.getNullCount(column.getName());
                if (maybeNullCount.isPresent()) {
                    nullCounts.put(column, nullCounts.get(column) + maybeNullCount.get());
                } else {
                    // If any individual file fails to report null counts, fail to calculate the total for the table
                    nullCounts.put(column, NaN);
                }
            }
            // Math.min returns NaN if any operand is NaN
            stats.getMinColumnValue(column).map(parsedValue -> toStatsRepresentation(column.getType(), parsedValue)).filter(OptionalDouble::isPresent).map(OptionalDouble::getAsDouble).ifPresent(parsedValueAsDouble -> minValues.merge(column, parsedValueAsDouble, Math::min));
            stats.getMaxColumnValue(column).map(parsedValue -> toStatsRepresentation(column.getType(), parsedValue)).filter(OptionalDouble::isPresent).map(OptionalDouble::getAsDouble).ifPresent(parsedValueAsDouble -> maxValues.merge(column, parsedValueAsDouble, Math::max));
        }
    }
    if (numRecords == 0) {
        return createZeroStatistics(columns);
    }
    TableStatistics.Builder statsBuilder = new TableStatistics.Builder().setRowCount(Estimate.of(numRecords));
    Optional<DeltaLakeStatistics> statistics = Optional.empty();
    if (isExtendedStatisticsEnabled(session)) {
        statistics = statisticsAccess.readDeltaLakeStatistics(session, tableHandle.getLocation());
    }
    for (DeltaLakeColumnHandle column : columns) {
        ColumnStatistics.Builder columnStatsBuilder = new ColumnStatistics.Builder();
        Double nullCount = nullCounts.get(column);
        columnStatsBuilder.setNullsFraction(nullCount.isNaN() ? Estimate.unknown() : Estimate.of(nullCount / numRecords));
        Double maxValue = maxValues.get(column);
        Double minValue = minValues.get(column);
        if (isValidInRange(maxValue) && isValidInRange(minValue)) {
            columnStatsBuilder.setRange(new DoubleRange(minValue, maxValue));
        } else if (isValidInRange(maxValue)) {
            columnStatsBuilder.setRange(new DoubleRange(NEGATIVE_INFINITY, maxValue));
        } else if (isValidInRange(minValue)) {
            columnStatsBuilder.setRange(new DoubleRange(minValue, POSITIVE_INFINITY));
        }
        // extend statistics with NDV
        if (column.getColumnType() == PARTITION_KEY) {
            columnStatsBuilder.setDistinctValuesCount(Estimate.of(partitioningColumnsDistinctValues.get(column).size()));
        }
        if (statistics.isPresent()) {
            DeltaLakeColumnStatistics deltaLakeColumnStatistics = statistics.get().getColumnStatistics().get(column.getName());
            if (deltaLakeColumnStatistics != null && column.getColumnType() != PARTITION_KEY) {
                columnStatsBuilder.setDistinctValuesCount(Estimate.of(deltaLakeColumnStatistics.getNdvSummary().cardinality()));
            }
        }
        statsBuilder.setColumnStatistics(column, columnStatsBuilder.build());
    }
    return statsBuilder.build();
}

Example 2 with DoubleRange

use of io.trino.spi.statistics.DoubleRange in project trino by trinodb.

the class BlackHoleMetadata method getTableStatistics.

@Override
public TableStatistics getTableStatistics(ConnectorSession session, ConnectorTableHandle tableHandle, Constraint constraint) {
    BlackHoleTableHandle table = (BlackHoleTableHandle) tableHandle;
    TableStatistics.Builder tableStats = TableStatistics.builder();
    double rows = (double) table.getSplitCount() * table.getPagesPerSplit() * table.getRowsPerPage();
    tableStats.setRowCount(Estimate.of(rows));
    for (BlackHoleColumnHandle column : table.getColumnHandles()) {
        ColumnStatistics.Builder stats = ColumnStatistics.builder().setDistinctValuesCount(Estimate.of(1)).setNullsFraction(Estimate.of(0));
        if (isNumericType(column.getColumnType())) {
            stats.setRange(new DoubleRange(0, 0));
        }
        tableStats.setColumnStatistics(column, stats.build());
    }
    return tableStats.build();
}

Example 3 with DoubleRange

use of io.trino.spi.statistics.DoubleRange in project trino by trinodb.

the class TestMetastoreHiveStatisticsProvider method testGetTableStatistics.

@Test
public void testGetTableStatistics() {
    String partitionName = "p1=string1/p2=1234";
    PartitionStatistics statistics = PartitionStatistics.builder().setBasicStatistics(new HiveBasicStatistics(OptionalLong.empty(), OptionalLong.of(1000), OptionalLong.empty(), OptionalLong.empty())).setColumnStatistics(ImmutableMap.of(COLUMN, createIntegerColumnStatistics(OptionalLong.of(-100), OptionalLong.of(100), OptionalLong.of(500), OptionalLong.of(300)))).build();
    MetastoreHiveStatisticsProvider statisticsProvider = new MetastoreHiveStatisticsProvider((session, table, hivePartitions) -> ImmutableMap.of(partitionName, statistics));
    HiveColumnHandle columnHandle = createBaseColumn(COLUMN, 2, HIVE_LONG, BIGINT, REGULAR, Optional.empty());
    TableStatistics expected = TableStatistics.builder().setRowCount(Estimate.of(1000)).setColumnStatistics(PARTITION_COLUMN_1, ColumnStatistics.builder().setDataSize(Estimate.of(7000)).setNullsFraction(Estimate.of(0)).setDistinctValuesCount(Estimate.of(1)).build()).setColumnStatistics(PARTITION_COLUMN_2, ColumnStatistics.builder().setRange(new DoubleRange(1234, 1234)).setNullsFraction(Estimate.of(0)).setDistinctValuesCount(Estimate.of(1)).build()).setColumnStatistics(columnHandle, ColumnStatistics.builder().setRange(new DoubleRange(-100, 100)).setNullsFraction(Estimate.of(0.5)).setDistinctValuesCount(Estimate.of(300)).build()).build();
    assertEquals(statisticsProvider.getTableStatistics(SESSION, TABLE, ImmutableMap.of("p1", PARTITION_COLUMN_1, "p2", PARTITION_COLUMN_2, COLUMN, columnHandle), ImmutableMap.of("p1", VARCHAR, "p2", BIGINT, COLUMN, BIGINT), ImmutableList.of(partition(partitionName))), expected);
}

Example 4 with DoubleRange

use of io.trino.spi.statistics.DoubleRange in project trino by trinodb.

the class TestTpcdsMetadataStatistics method testTableStatsDetails.

@Test
public void testTableStatsDetails() {
    SchemaTableName schemaTableName = new SchemaTableName("sf1", Table.CALL_CENTER.getName());
    ConnectorTableHandle tableHandle = metadata.getTableHandle(session, schemaTableName);
    TableStatistics tableStatistics = metadata.getTableStatistics(session, tableHandle, alwaysTrue());
    estimateAssertion.assertClose(tableStatistics.getRowCount(), Estimate.of(6), "Row count does not match");
    // all columns have stats
    Map<String, ColumnHandle> columnHandles = metadata.getColumnHandles(session, tableHandle);
    for (ColumnHandle column : columnHandles.values()) {
        assertTrue(tableStatistics.getColumnStatistics().containsKey(column));
        assertNotNull(tableStatistics.getColumnStatistics().get(column));
    }
    // identifier
    assertColumnStatistics(tableStatistics.getColumnStatistics().get(columnHandles.get(CallCenterColumn.CC_CALL_CENTER_SK.getName())), ColumnStatistics.builder().setNullsFraction(Estimate.of(0)).setDistinctValuesCount(Estimate.of(6)).setRange(new DoubleRange(1, 6)).build());
    // varchar
    assertColumnStatistics(tableStatistics.getColumnStatistics().get(columnHandles.get(CallCenterColumn.CC_CALL_CENTER_ID.getName())), ColumnStatistics.builder().setNullsFraction(Estimate.of(0)).setDistinctValuesCount(Estimate.of(3)).setDataSize(Estimate.of(48.0)).build());
    // char
    assertColumnStatistics(tableStatistics.getColumnStatistics().get(columnHandles.get(CallCenterColumn.CC_ZIP.getName())), ColumnStatistics.builder().setNullsFraction(Estimate.of(0)).setDistinctValuesCount(Estimate.of(1)).setDataSize(Estimate.of(5.0)).build());
    // decimal
    assertColumnStatistics(tableStatistics.getColumnStatistics().get(columnHandles.get(CallCenterColumn.CC_GMT_OFFSET.getName())), ColumnStatistics.builder().setNullsFraction(Estimate.of(0)).setDistinctValuesCount(Estimate.of(1)).setRange(new DoubleRange(-5, -5)).build());
    // date
    assertColumnStatistics(tableStatistics.getColumnStatistics().get(columnHandles.get(CallCenterColumn.CC_REC_START_DATE.getName())), ColumnStatistics.builder().setNullsFraction(Estimate.of(0)).setDistinctValuesCount(Estimate.of(4)).setRange(new DoubleRange(10227L, 11688L)).build());
    // only null values
    assertColumnStatistics(tableStatistics.getColumnStatistics().get(columnHandles.get(CallCenterColumn.CC_CLOSED_DATE_SK.getName())), ColumnStatistics.builder().setNullsFraction(Estimate.of(1)).setDistinctValuesCount(Estimate.of(0)).build());
}

Example 5 with DoubleRange

use of io.trino.spi.statistics.DoubleRange in project trino by trinodb.

the class MetastoreHiveStatisticsProvider method calculateRangeForPartitioningKey.

@VisibleForTesting
static Optional<DoubleRange> calculateRangeForPartitioningKey(HiveColumnHandle column, Type type, List<HivePartition> partitions) {
    List<OptionalDouble> convertedValues = partitions.stream().map(HivePartition::getKeys).map(keys -> keys.get(column)).filter(value -> !value.isNull()).map(NullableValue::getValue).map(value -> convertPartitionValueToDouble(type, value)).collect(toImmutableList());
    if (convertedValues.stream().noneMatch(OptionalDouble::isPresent)) {
        return Optional.empty();
    }
    double[] values = convertedValues.stream().peek(convertedValue -> checkState(convertedValue.isPresent(), "convertedValue is missing")).mapToDouble(OptionalDouble::getAsDouble).toArray();
    verify(values.length != 0, "No values");
    if (DoubleStream.of(values).anyMatch(Double::isNaN)) {
        return Optional.empty();
    }
    double min = DoubleStream.of(values).min().orElseThrow();
    double max = DoubleStream.of(values).max().orElseThrow();
    return Optional.of(new DoubleRange(min, max));
}