Examples with PartitionField org.apache.iceberg.PartitionField used on opensource projects

Example 6 with PartitionField

use of org.apache.iceberg.PartitionField in project presto by prestodb.

the class PartitionTable method buildRecordCursor.

private RecordCursor buildRecordCursor(Map<StructLikeWrapper, Partition> partitions, List<PartitionField> partitionFields) {
    List<Type> partitionTypes = partitionTypes(partitionFields);
    List<? extends Class<?>> partitionColumnClass = partitionTypes.stream().map(type -> type.typeId().javaClass()).collect(toImmutableList());
    int columnCounts = partitionColumnTypes.size() + 3 + columnMetricTypes.size();
    ImmutableList.Builder<List<Object>> records = ImmutableList.builder();
    for (Partition partition : partitions.values()) {
        List<Object> row = new ArrayList<>(columnCounts);
        // add data for partition columns
        for (int i = 0; i < partitionColumnTypes.size(); i++) {
            row.add(convert(partition.getValues().get(i, partitionColumnClass.get(i)), partitionTypes.get(i)));
        }
        // add the top level metrics.
        row.add(partition.getRecordCount());
        row.add(partition.getFileCount());
        row.add(partition.getSize());
        // add column level metrics
        for (int i = 0; i < columnMetricTypes.size(); i++) {
            if (!partition.hasValidColumnMetrics()) {
                row.add(null);
                continue;
            }
            Integer fieldId = nonPartitionPrimitiveColumns.get(i).fieldId();
            Type.PrimitiveType type = idToTypeMapping.get(fieldId);
            Object min = convert(partition.getMinValues().get(fieldId), type);
            Object max = convert(partition.getMaxValues().get(fieldId), type);
            Long nullCount = partition.getNullCounts().get(fieldId);
            row.add(getColumnMetricBlock(columnMetricTypes.get(i), min, max, nullCount));
        }
        records.add(row);
    }
    return new InMemoryRecordSet(resultTypes, records.build()).cursor();
}

Example 7 with PartitionField

use of org.apache.iceberg.PartitionField in project presto by prestodb.

the class TableStatisticsMaker method makeTableStatistics.

private TableStatistics makeTableStatistics(IcebergTableHandle tableHandle, Constraint constraint) {
    if (!tableHandle.getSnapshotId().isPresent() || constraint.getSummary().isNone()) {
        return TableStatistics.empty();
    }
    TupleDomain<IcebergColumnHandle> intersection = constraint.getSummary().transform(IcebergColumnHandle.class::cast).intersect(tableHandle.getPredicate());
    if (intersection.isNone()) {
        return TableStatistics.empty();
    }
    List<Types.NestedField> columns = icebergTable.schema().columns();
    Map<Integer, Type.PrimitiveType> idToTypeMapping = columns.stream().filter(column -> column.type().isPrimitiveType()).collect(Collectors.toMap(Types.NestedField::fieldId, column -> column.type().asPrimitiveType()));
    List<PartitionField> partitionFields = icebergTable.spec().fields();
    Set<Integer> identityPartitionIds = getIdentityPartitions(icebergTable.spec()).keySet().stream().map(PartitionField::sourceId).collect(toSet());
    List<Types.NestedField> nonPartitionPrimitiveColumns = columns.stream().filter(column -> !identityPartitionIds.contains(column.fieldId()) && column.type().isPrimitiveType()).collect(toImmutableList());
    List<Type> icebergPartitionTypes = partitionTypes(partitionFields, idToTypeMapping);
    List<IcebergColumnHandle> columnHandles = getColumns(icebergTable.schema(), typeManager);
    Map<Integer, IcebergColumnHandle> idToColumnHandle = columnHandles.stream().collect(toImmutableMap(IcebergColumnHandle::getId, identity()));
    ImmutableMap.Builder<Integer, ColumnFieldDetails> idToDetailsBuilder = ImmutableMap.builder();
    for (int index = 0; index < partitionFields.size(); index++) {
        PartitionField field = partitionFields.get(index);
        Type type = icebergPartitionTypes.get(index);
        idToDetailsBuilder.put(field.sourceId(), new ColumnFieldDetails(field, idToColumnHandle.get(field.sourceId()), type, toPrestoType(type, typeManager), type.typeId().javaClass()));
    }
    Map<Integer, ColumnFieldDetails> idToDetails = idToDetailsBuilder.build();
    TableScan tableScan = icebergTable.newScan().filter(toIcebergExpression(intersection)).useSnapshot(tableHandle.getSnapshotId().get()).includeColumnStats();
    Partition summary = null;
    try (CloseableIterable<FileScanTask> fileScanTasks = tableScan.planFiles()) {
        for (FileScanTask fileScanTask : fileScanTasks) {
            DataFile dataFile = fileScanTask.file();
            if (!dataFileMatches(dataFile, constraint, idToTypeMapping, partitionFields, idToDetails)) {
                continue;
            }
            if (summary == null) {
                summary = new Partition(idToTypeMapping, nonPartitionPrimitiveColumns, dataFile.partition(), dataFile.recordCount(), dataFile.fileSizeInBytes(), toMap(idToTypeMapping, dataFile.lowerBounds()), toMap(idToTypeMapping, dataFile.upperBounds()), dataFile.nullValueCounts(), dataFile.columnSizes());
            } else {
                summary.incrementFileCount();
                summary.incrementRecordCount(dataFile.recordCount());
                summary.incrementSize(dataFile.fileSizeInBytes());
                updateSummaryMin(summary, partitionFields, toMap(idToTypeMapping, dataFile.lowerBounds()), dataFile.nullValueCounts(), dataFile.recordCount());
                updateSummaryMax(summary, partitionFields, toMap(idToTypeMapping, dataFile.upperBounds()), dataFile.nullValueCounts(), dataFile.recordCount());
                summary.updateNullCount(dataFile.nullValueCounts());
                updateColumnSizes(summary, dataFile.columnSizes());
            }
        }
    } catch (IOException e) {
        throw new UncheckedIOException(e);
    }
    if (summary == null) {
        return TableStatistics.empty();
    }
    double recordCount = summary.getRecordCount();
    TableStatistics.Builder result = TableStatistics.builder();
    result.setRowCount(Estimate.of(recordCount));
    result.setTotalSize(Estimate.of(summary.getSize()));
    for (IcebergColumnHandle columnHandle : idToColumnHandle.values()) {
        int fieldId = columnHandle.getId();
        ColumnStatistics.Builder columnBuilder = new ColumnStatistics.Builder();
        Long nullCount = summary.getNullCounts().get(fieldId);
        if (nullCount != null) {
            columnBuilder.setNullsFraction(Estimate.of(nullCount / recordCount));
        }
        if (summary.getColumnSizes() != null) {
            Long columnSize = summary.getColumnSizes().get(fieldId);
            if (columnSize != null) {
                columnBuilder.setDataSize(Estimate.of(columnSize));
            }
        }
        Object min = summary.getMinValues().get(fieldId);
        Object max = summary.getMaxValues().get(fieldId);
        if (min instanceof Number && max instanceof Number) {
            columnBuilder.setRange(Optional.of(new DoubleRange(((Number) min).doubleValue(), ((Number) max).doubleValue())));
        }
        result.setColumnStatistics(columnHandle, columnBuilder.build());
    }
    return result.build();
}

Example 8 with PartitionField

use of org.apache.iceberg.PartitionField in project presto by prestodb.

the class TableStatisticsMaker method partitionTypes.

public List<Type> partitionTypes(List<PartitionField> partitionFields, Map<Integer, Type.PrimitiveType> idToTypeMapping) {
    ImmutableList.Builder<Type> partitionTypeBuilder = ImmutableList.builder();
    for (PartitionField partitionField : partitionFields) {
        Type.PrimitiveType sourceType = idToTypeMapping.get(partitionField.sourceId());
        Type type = partitionField.transform().getResultType(sourceType);
        partitionTypeBuilder.add(type);
    }
    return partitionTypeBuilder.build();
}

Example 9 with PartitionField

use of org.apache.iceberg.PartitionField in project presto by prestodb.

the class TableStatisticsMaker method updatePartitionedStats.

private void updatePartitionedStats(Partition summary, List<PartitionField> partitionFields, Map<Integer, Object> current, Map<Integer, Object> newStats, Predicate<Integer> predicate) {
    for (PartitionField field : partitionFields) {
        int id = field.sourceId();
        if (summary.getCorruptedStats().contains(id)) {
            continue;
        }
        Object newValue = newStats.get(id);
        if (newValue == null) {
            continue;
        }
        Object oldValue = current.putIfAbsent(id, newValue);
        if (oldValue != null) {
            Comparator<Object> comparator = Comparators.forType(summary.getIdToTypeMapping().get(id));
            if (predicate.test(comparator.compare(oldValue, newValue))) {
                current.put(id, newValue);
            }
        }
    }
}

Example 10 with PartitionField

use of org.apache.iceberg.PartitionField in project presto by prestodb.

the class IcebergSplitSource method getPartitionKeys.

private static Map<Integer, String> getPartitionKeys(FileScanTask scanTask) {
    StructLike partition = scanTask.file().partition();
    PartitionSpec spec = scanTask.spec();
    Map<PartitionField, Integer> fieldToIndex = getIdentityPartitions(spec);
    Map<Integer, String> partitionKeys = new HashMap<>();
    fieldToIndex.forEach((field, index) -> {
        int id = field.sourceId();
        Type type = spec.schema().findType(id);
        Class<?> javaClass = type.typeId().javaClass();
        Object value = partition.get(index, javaClass);
        if (value == null) {
            partitionKeys.put(id, null);
        } else {
            String partitionValue;
            if (type.typeId() == FIXED || type.typeId() == BINARY) {
                // this is safe because Iceberg PartitionData directly wraps the byte array
                partitionValue = new String(((ByteBuffer) value).array(), UTF_8);
            } else {
                partitionValue = value.toString();
            }
            partitionKeys.put(id, partitionValue);
        }
    });
    return Collections.unmodifiableMap(partitionKeys);
}