Examples with DeltaLakeFileStatistics io.trino.plugin.deltalake.transactionlog.statistics.DeltaLakeFileStatistics used on opensource projects

Example 1 with DeltaLakeFileStatistics

use of io.trino.plugin.deltalake.transactionlog.statistics.DeltaLakeFileStatistics in project trino by trinodb.

the class TestDeltaLakeWriter method testMergeFloatNaNStatistics.

@Test
public void testMergeFloatNaNStatistics() {
    String columnName = "t_float";
    PrimitiveType type = new PrimitiveType(Type.Repetition.REQUIRED, PrimitiveType.PrimitiveTypeName.FLOAT, columnName);
    List<ColumnChunkMetaData> metadata = ImmutableList.of(createMetaData(columnName, type, 10, Statistics.getBuilderForReading(type).withMin(getFloatByteArray(0.01f)).withMax(getFloatByteArray(1.0f)).withNumNulls(6).build()), createMetaData(columnName, type, 10, Statistics.getBuilderForReading(type).withMin(getFloatByteArray(Float.NaN)).withMax(getFloatByteArray(1.0f)).withNumNulls(6).build()), createMetaData(columnName, type, 10, Statistics.getBuilderForReading(type).withMin(getFloatByteArray(-2.001f)).withMax(getFloatByteArray(0.0f)).withNumNulls(7).build()));
    DeltaLakeColumnHandle floatColumn = new DeltaLakeColumnHandle(columnName, REAL, REGULAR);
    DeltaLakeFileStatistics fileStats = mergeStats(buildMultimap(columnName, metadata), ImmutableMap.of(columnName, REAL), 20);
    assertEquals(fileStats.getNumRecords(), Optional.of(20L));
    assertEquals(fileStats.getMinColumnValue(floatColumn), Optional.empty());
    assertEquals(fileStats.getMaxColumnValue(floatColumn), Optional.empty());
    assertEquals(fileStats.getNullCount(columnName), Optional.empty());
}

Example 2 with DeltaLakeFileStatistics

use of io.trino.plugin.deltalake.transactionlog.statistics.DeltaLakeFileStatistics in project trino by trinodb.

the class TestDeltaLakeWriter method testMergeStringStatistics.

@Test
public void testMergeStringStatistics() {
    String columnName = "t_string";
    PrimitiveType type = new PrimitiveType(Type.Repetition.REQUIRED, PrimitiveType.PrimitiveTypeName.BINARY, columnName);
    List<ColumnChunkMetaData> metadata = ImmutableList.of(createMetaData(columnName, type, 10, Statistics.getBuilderForReading(type).withMin("aba".getBytes(UTF_8)).withMax("ab⌘".getBytes(UTF_8)).withNumNulls(6).build()), createMetaData(columnName, type, 10, Statistics.getBuilderForReading(type).withMin("aba".getBytes(UTF_8)).withMax("abc".getBytes(UTF_8)).withNumNulls(6).build()));
    DeltaLakeColumnHandle varcharColumn = new DeltaLakeColumnHandle(columnName, VarcharType.createUnboundedVarcharType(), REGULAR);
    DeltaLakeFileStatistics fileStats = mergeStats(buildMultimap(columnName, metadata), ImmutableMap.of(columnName, createUnboundedVarcharType()), 20);
    assertEquals(fileStats.getNumRecords(), Optional.of(20L));
    assertEquals(fileStats.getMinColumnValue(varcharColumn), Optional.of(utf8Slice("aba")));
    assertEquals(fileStats.getMaxColumnValue(varcharColumn), Optional.of(utf8Slice("ab⌘")));
    assertEquals(fileStats.getNullCount(columnName), Optional.of(12L));
}

Example 3 with DeltaLakeFileStatistics

use of io.trino.plugin.deltalake.transactionlog.statistics.DeltaLakeFileStatistics 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 4 with DeltaLakeFileStatistics

use of io.trino.plugin.deltalake.transactionlog.statistics.DeltaLakeFileStatistics in project trino by trinodb.

the class AbstractTestDeltaLakeCreateTableStatistics method testDateRecords.

@Test
public void testDateRecords() throws IOException {
    String columnName = "t_date";
    DeltaLakeColumnHandle columnHandle = new DeltaLakeColumnHandle(columnName, DateType.DATE, REGULAR);
    try (TestTable table = new TestTable("test_date_records_", ImmutableList.of(columnName), "VALUES DATE '2011-08-08', DATE '2012-08-08', DATE '2013-08-08', DATE '2013-08-09'")) {
        List<AddFileEntry> addFileEntries = getAddFileEntries(table.getName());
        AddFileEntry entry = getOnlyElement(addFileEntries);
        assertThat(entry.getStats()).isPresent();
        DeltaLakeFileStatistics fileStatistics = entry.getStats().get();
        assertEquals(fileStatistics.getNumRecords(), Optional.of(4L));
        assertEquals(fileStatistics.getMinColumnValue(columnHandle), Optional.of(LocalDate.parse("2011-08-08").toEpochDay()));
        assertEquals(fileStatistics.getMaxColumnValue(columnHandle), Optional.of(LocalDate.parse("2013-08-09").toEpochDay()));
        assertEquals(fileStatistics.getNullCount(columnName), Optional.of(0L));
    }
}

Example 5 with DeltaLakeFileStatistics

use of io.trino.plugin.deltalake.transactionlog.statistics.DeltaLakeFileStatistics in project trino by trinodb.

the class AbstractTestDeltaLakeCreateTableStatistics method testPartitionedTable.

@Test
public void testPartitionedTable() throws IOException {
    String columnName = "t_string";
    DeltaLakeColumnHandle columnHandle = new DeltaLakeColumnHandle(columnName, createUnboundedVarcharType(), REGULAR);
    String partitionColumn = "t_int";
    try (TestTable table = new TestTable("test_partitioned_table_", ImmutableList.of(columnName, partitionColumn), ImmutableList.of(partitionColumn), "VALUES ('a', 1), ('b', 1), ('c', 1), ('c', 2), ('d', 2), ('e', 2), (null, 1)")) {
        List<AddFileEntry> addFileEntries = getAddFileEntries(table.getName());
        assertEquals(addFileEntries.size(), 2);
        for (AddFileEntry addFileEntry : addFileEntries) {
            assertThat(addFileEntry.getStats()).isPresent();
            DeltaLakeFileStatistics fileStatistics = addFileEntry.getStats().get();
            if (addFileEntry.getPartitionValues().get(partitionColumn).equals("1")) {
                assertEquals(fileStatistics.getMinColumnValue(columnHandle), Optional.of(utf8Slice("a")));
                assertEquals(fileStatistics.getMaxColumnValue(columnHandle), Optional.of(utf8Slice("c")));
                assertEquals(fileStatistics.getNumRecords(), Optional.of(4L));
                assertEquals(fileStatistics.getNullCount(columnName), Optional.of(1L));
            } else if (addFileEntry.getPartitionValues().get(partitionColumn).equals("2")) {
                assertEquals(fileStatistics.getMinColumnValue(columnHandle), Optional.of(utf8Slice("c")));
                assertEquals(fileStatistics.getMaxColumnValue(columnHandle), Optional.of(utf8Slice("e")));
                assertEquals(fileStatistics.getNumRecords(), Optional.of(3L));
                assertEquals(fileStatistics.getNullCount(columnName), Optional.of(0L));
            }
        }
    }
}