Search in sources :

Example 36 with LogicalType

use of org.apache.flink.table.types.logical.LogicalType in project flink by apache.

the class StreamExecMatch method checkOrderKeys.

private void checkOrderKeys(RowType inputRowType) {
    SortSpec orderKeys = matchSpec.getOrderKeys();
    if (orderKeys.getFieldSize() == 0) {
        throw new TableException("You must specify either rowtime or proctime for order by.");
    }
    SortSpec.SortFieldSpec timeOrderField = orderKeys.getFieldSpec(0);
    int timeOrderFieldIdx = timeOrderField.getFieldIndex();
    LogicalType timeOrderFieldType = inputRowType.getTypeAt(timeOrderFieldIdx);
    // need to identify time between others order fields. Time needs to be first sort element
    if (!TypeCheckUtils.isRowTime(timeOrderFieldType) && !TypeCheckUtils.isProcTime(timeOrderFieldType)) {
        throw new TableException("You must specify either rowtime or proctime for order by as the first one.");
    }
    // time ordering needs to be ascending
    if (!orderKeys.getAscendingOrders()[0]) {
        throw new TableException("Primary sort order of a streaming table must be ascending on time.");
    }
}
Also used : TableException(org.apache.flink.table.api.TableException) LogicalType(org.apache.flink.table.types.logical.LogicalType) SortSpec(org.apache.flink.table.planner.plan.nodes.exec.spec.SortSpec)

Example 37 with LogicalType

use of org.apache.flink.table.types.logical.LogicalType in project flink by apache.

the class StreamExecOverAggregate method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    if (overSpec.getGroups().size() > 1) {
        throw new TableException("All aggregates must be computed on the same window.");
    }
    final OverSpec.GroupSpec group = overSpec.getGroups().get(0);
    final int[] orderKeys = group.getSort().getFieldIndices();
    final boolean[] isAscendingOrders = group.getSort().getAscendingOrders();
    if (orderKeys.length != 1 || isAscendingOrders.length != 1) {
        throw new TableException("The window can only be ordered by a single time column.");
    }
    if (!isAscendingOrders[0]) {
        throw new TableException("The window can only be ordered in ASCENDING mode.");
    }
    final int[] partitionKeys = overSpec.getPartition().getFieldIndices();
    if (partitionKeys.length > 0 && config.getStateRetentionTime() < 0) {
        LOG.warn("No state retention interval configured for a query which accumulates state. " + "Please provide a query configuration with valid retention interval to prevent " + "excessive state size. You may specify a retention time of 0 to not clean up the state.");
    }
    final ExecEdge inputEdge = getInputEdges().get(0);
    final Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
    final RowType inputRowType = (RowType) inputEdge.getOutputType();
    final int orderKey = orderKeys[0];
    final LogicalType orderKeyType = inputRowType.getFields().get(orderKey).getType();
    // check time field && identify window rowtime attribute
    final int rowTimeIdx;
    if (isRowtimeAttribute(orderKeyType)) {
        rowTimeIdx = orderKey;
    } else if (isProctimeAttribute(orderKeyType)) {
        rowTimeIdx = -1;
    } else {
        throw new TableException("OVER windows' ordering in stream mode must be defined on a time attribute.");
    }
    final List<RexLiteral> constants = overSpec.getConstants();
    final List<String> fieldNames = new ArrayList<>(inputRowType.getFieldNames());
    final List<LogicalType> fieldTypes = new ArrayList<>(inputRowType.getChildren());
    IntStream.range(0, constants.size()).forEach(i -> fieldNames.add("TMP" + i));
    for (int i = 0; i < constants.size(); ++i) {
        fieldNames.add("TMP" + i);
        fieldTypes.add(FlinkTypeFactory.toLogicalType(constants.get(i).getType()));
    }
    final RowType aggInputRowType = RowType.of(fieldTypes.toArray(new LogicalType[0]), fieldNames.toArray(new String[0]));
    final CodeGeneratorContext ctx = new CodeGeneratorContext(config.getTableConfig());
    final KeyedProcessFunction<RowData, RowData, RowData> overProcessFunction;
    if (group.getLowerBound().isPreceding() && group.getLowerBound().isUnbounded() && group.getUpperBound().isCurrentRow()) {
        // unbounded OVER window
        overProcessFunction = createUnboundedOverProcessFunction(ctx, group.getAggCalls(), constants, aggInputRowType, inputRowType, rowTimeIdx, group.isRows(), config, planner.getRelBuilder());
    } else if (group.getLowerBound().isPreceding() && !group.getLowerBound().isUnbounded() && group.getUpperBound().isCurrentRow()) {
        final Object boundValue = OverAggregateUtil.getBoundary(overSpec, group.getLowerBound());
        if (boundValue instanceof BigDecimal) {
            throw new TableException("the specific value is decimal which haven not supported yet.");
        }
        // bounded OVER window
        final long precedingOffset = -1 * (long) boundValue + (group.isRows() ? 1 : 0);
        overProcessFunction = createBoundedOverProcessFunction(ctx, group.getAggCalls(), constants, aggInputRowType, inputRowType, rowTimeIdx, group.isRows(), precedingOffset, config, planner.getRelBuilder());
    } else {
        throw new TableException("OVER RANGE FOLLOWING windows are not supported yet.");
    }
    final KeyedProcessOperator<RowData, RowData, RowData> operator = new KeyedProcessOperator<>(overProcessFunction);
    OneInputTransformation<RowData, RowData> transform = ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationMeta(OVER_AGGREGATE_TRANSFORMATION, config), operator, InternalTypeInfo.of(getOutputType()), inputTransform.getParallelism());
    // set KeyType and Selector for state
    final RowDataKeySelector selector = KeySelectorUtil.getRowDataSelector(partitionKeys, InternalTypeInfo.of(inputRowType));
    transform.setStateKeySelector(selector);
    transform.setStateKeyType(selector.getProducedType());
    return transform;
}
Also used : RexLiteral(org.apache.calcite.rex.RexLiteral) OneInputTransformation(org.apache.flink.streaming.api.transformations.OneInputTransformation) Transformation(org.apache.flink.api.dag.Transformation) ExecEdge(org.apache.flink.table.planner.plan.nodes.exec.ExecEdge) ArrayList(java.util.ArrayList) RowType(org.apache.flink.table.types.logical.RowType) LogicalType(org.apache.flink.table.types.logical.LogicalType) RowData(org.apache.flink.table.data.RowData) RowDataKeySelector(org.apache.flink.table.runtime.keyselector.RowDataKeySelector) KeyedProcessOperator(org.apache.flink.streaming.api.operators.KeyedProcessOperator) TableException(org.apache.flink.table.api.TableException) CodeGeneratorContext(org.apache.flink.table.planner.codegen.CodeGeneratorContext) OverSpec(org.apache.flink.table.planner.plan.nodes.exec.spec.OverSpec) BigDecimal(java.math.BigDecimal)

Example 38 with LogicalType

use of org.apache.flink.table.types.logical.LogicalType in project flink by apache.

the class StreamExecPythonOverAggregate method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    if (overSpec.getGroups().size() > 1) {
        throw new TableException("All aggregates must be computed on the same window.");
    }
    final OverSpec.GroupSpec group = overSpec.getGroups().get(0);
    final int[] orderKeys = group.getSort().getFieldIndices();
    final boolean[] isAscendingOrders = group.getSort().getAscendingOrders();
    if (orderKeys.length != 1 || isAscendingOrders.length != 1) {
        throw new TableException("The window can only be ordered by a single time column.");
    }
    if (!isAscendingOrders[0]) {
        throw new TableException("The window can only be ordered in ASCENDING mode.");
    }
    final int[] partitionKeys = overSpec.getPartition().getFieldIndices();
    if (partitionKeys.length > 0 && config.getStateRetentionTime() < 0) {
        LOG.warn("No state retention interval configured for a query which accumulates state. " + "Please provide a query configuration with valid retention interval to prevent " + "excessive state size. You may specify a retention time of 0 to not clean up the state.");
    }
    final ExecEdge inputEdge = getInputEdges().get(0);
    final Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
    final RowType inputRowType = (RowType) inputEdge.getOutputType();
    final int orderKey = orderKeys[0];
    final LogicalType orderKeyType = inputRowType.getFields().get(orderKey).getType();
    // check time field && identify window rowtime attribute
    final int rowTimeIdx;
    if (isRowtimeAttribute(orderKeyType)) {
        rowTimeIdx = orderKey;
    } else if (isProctimeAttribute(orderKeyType)) {
        rowTimeIdx = -1;
    } else {
        throw new TableException("OVER windows' ordering in stream mode must be defined on a time attribute.");
    }
    if (group.getLowerBound().isPreceding() && group.getLowerBound().isUnbounded()) {
        throw new TableException("Python UDAF is not supported to be used in UNBOUNDED PRECEDING OVER windows.");
    } else if (!group.getUpperBound().isCurrentRow()) {
        throw new TableException("Python UDAF is not supported to be used in UNBOUNDED FOLLOWING OVER windows.");
    }
    Object boundValue = OverAggregateUtil.getBoundary(overSpec, group.getLowerBound());
    if (boundValue instanceof BigDecimal) {
        throw new TableException("the specific value is decimal which haven not supported yet.");
    }
    long precedingOffset = -1 * (long) boundValue;
    Configuration pythonConfig = CommonPythonUtil.getMergedConfig(planner.getExecEnv(), config.getTableConfig());
    OneInputTransformation<RowData, RowData> transform = createPythonOneInputTransformation(inputTransform, inputRowType, InternalTypeInfo.of(getOutputType()).toRowType(), rowTimeIdx, group.getAggCalls().toArray(new AggregateCall[0]), precedingOffset, group.isRows(), config.getStateRetentionTime(), config.getMaxIdleStateRetentionTime(), pythonConfig, config);
    if (CommonPythonUtil.isPythonWorkerUsingManagedMemory(pythonConfig)) {
        transform.declareManagedMemoryUseCaseAtSlotScope(ManagedMemoryUseCase.PYTHON);
    }
    // set KeyType and Selector for state
    final RowDataKeySelector selector = KeySelectorUtil.getRowDataSelector(partitionKeys, InternalTypeInfo.of(inputRowType));
    transform.setStateKeySelector(selector);
    transform.setStateKeyType(selector.getProducedType());
    return transform;
}
Also used : TableException(org.apache.flink.table.api.TableException) OneInputTransformation(org.apache.flink.streaming.api.transformations.OneInputTransformation) Transformation(org.apache.flink.api.dag.Transformation) ExecEdge(org.apache.flink.table.planner.plan.nodes.exec.ExecEdge) Configuration(org.apache.flink.configuration.Configuration) RowType(org.apache.flink.table.types.logical.RowType) LogicalType(org.apache.flink.table.types.logical.LogicalType) OverSpec(org.apache.flink.table.planner.plan.nodes.exec.spec.OverSpec) BigDecimal(java.math.BigDecimal) AggregateCall(org.apache.calcite.rel.core.AggregateCall) RowData(org.apache.flink.table.data.RowData) RowDataKeySelector(org.apache.flink.table.runtime.keyselector.RowDataKeySelector)

Example 39 with LogicalType

use of org.apache.flink.table.types.logical.LogicalType in project flink by apache.

the class StreamExecTemporalSort method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    // time ordering needs to be ascending
    if (sortSpec.getFieldSize() == 0 || !sortSpec.getFieldSpec(0).getIsAscendingOrder()) {
        throw new TableException("Sort: Primary sort order of a streaming table must be ascending on time.\n" + "please re-check sort statement according to the description above");
    }
    ExecEdge inputEdge = getInputEdges().get(0);
    Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
    RowType inputType = (RowType) inputEdge.getOutputType();
    LogicalType timeType = inputType.getTypeAt(sortSpec.getFieldSpec(0).getFieldIndex());
    if (isRowtimeAttribute(timeType)) {
        return createSortRowTime(inputType, inputTransform, config);
    } else if (isProctimeAttribute(timeType)) {
        return createSortProcTime(inputType, inputTransform, config);
    } else {
        throw new TableException(String.format("Sort: Internal Error\n" + "First field in temporal sort is not a time attribute, %s is given.", timeType));
    }
}
Also used : TableException(org.apache.flink.table.api.TableException) RowData(org.apache.flink.table.data.RowData) OneInputTransformation(org.apache.flink.streaming.api.transformations.OneInputTransformation) Transformation(org.apache.flink.api.dag.Transformation) ExecEdge(org.apache.flink.table.planner.plan.nodes.exec.ExecEdge) RowType(org.apache.flink.table.types.logical.RowType) LogicalType(org.apache.flink.table.types.logical.LogicalType)

Example 40 with LogicalType

use of org.apache.flink.table.types.logical.LogicalType in project flink by apache.

the class StreamExecWindowAggregate method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    final ExecEdge inputEdge = getInputEdges().get(0);
    final Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
    final RowType inputRowType = (RowType) inputEdge.getOutputType();
    final ZoneId shiftTimeZone = TimeWindowUtil.getShiftTimeZone(windowing.getTimeAttributeType(), config.getLocalTimeZone());
    final SliceAssigner sliceAssigner = createSliceAssigner(windowing, shiftTimeZone);
    // Hopping window requires additional COUNT(*) to determine whether to register next timer
    // through whether the current fired window is empty, see SliceSharedWindowAggProcessor.
    final AggregateInfoList aggInfoList = AggregateUtil.deriveStreamWindowAggregateInfoList(inputRowType, JavaScalaConversionUtil.toScala(Arrays.asList(aggCalls)), windowing.getWindow(), // isStateBackendDataViews
    true);
    final GeneratedNamespaceAggsHandleFunction<Long> generatedAggsHandler = createAggsHandler(sliceAssigner, aggInfoList, config, planner.getRelBuilder(), inputRowType.getChildren(), shiftTimeZone);
    final RowDataKeySelector selector = KeySelectorUtil.getRowDataSelector(grouping, InternalTypeInfo.of(inputRowType));
    final LogicalType[] accTypes = convertToLogicalTypes(aggInfoList.getAccTypes());
    final OneInputStreamOperator<RowData, RowData> windowOperator = SlicingWindowAggOperatorBuilder.builder().inputSerializer(new RowDataSerializer(inputRowType)).shiftTimeZone(shiftTimeZone).keySerializer((PagedTypeSerializer<RowData>) selector.getProducedType().toSerializer()).assigner(sliceAssigner).countStarIndex(aggInfoList.getIndexOfCountStar()).aggregate(generatedAggsHandler, new RowDataSerializer(accTypes)).build();
    final OneInputTransformation<RowData, RowData> transform = ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationMeta(WINDOW_AGGREGATE_TRANSFORMATION, config), SimpleOperatorFactory.of(windowOperator), InternalTypeInfo.of(getOutputType()), inputTransform.getParallelism(), WINDOW_AGG_MEMORY_RATIO);
    // set KeyType and Selector for state
    transform.setStateKeySelector(selector);
    transform.setStateKeyType(selector.getProducedType());
    return transform;
}
Also used : SliceAssigner(org.apache.flink.table.runtime.operators.window.slicing.SliceAssigner) OneInputTransformation(org.apache.flink.streaming.api.transformations.OneInputTransformation) Transformation(org.apache.flink.api.dag.Transformation) AggregateInfoList(org.apache.flink.table.planner.plan.utils.AggregateInfoList) ExecEdge(org.apache.flink.table.planner.plan.nodes.exec.ExecEdge) ZoneId(java.time.ZoneId) RowType(org.apache.flink.table.types.logical.RowType) LogicalType(org.apache.flink.table.types.logical.LogicalType) RowData(org.apache.flink.table.data.RowData) RowDataKeySelector(org.apache.flink.table.runtime.keyselector.RowDataKeySelector) RowDataSerializer(org.apache.flink.table.runtime.typeutils.RowDataSerializer)

Aggregations

LogicalType (org.apache.flink.table.types.logical.LogicalType)192 DataType (org.apache.flink.table.types.DataType)53 RowType (org.apache.flink.table.types.logical.RowType)53 RowData (org.apache.flink.table.data.RowData)45 List (java.util.List)29 ArrayList (java.util.ArrayList)28 TableException (org.apache.flink.table.api.TableException)25 TimestampType (org.apache.flink.table.types.logical.TimestampType)25 Internal (org.apache.flink.annotation.Internal)21 IntType (org.apache.flink.table.types.logical.IntType)21 Map (java.util.Map)20 ValidationException (org.apache.flink.table.api.ValidationException)20 ArrayType (org.apache.flink.table.types.logical.ArrayType)19 DecimalType (org.apache.flink.table.types.logical.DecimalType)19 LocalZonedTimestampType (org.apache.flink.table.types.logical.LocalZonedTimestampType)17 Test (org.junit.Test)17 BigIntType (org.apache.flink.table.types.logical.BigIntType)16 LegacyTypeInformationType (org.apache.flink.table.types.logical.LegacyTypeInformationType)16 GenericRowData (org.apache.flink.table.data.GenericRowData)15 Arrays (java.util.Arrays)14