Examples with ExecNodeConfig org.apache.flink.table.planner.plan.nodes.exec.ExecNodeConfig used on opensource projects

Example 1 with ExecNodeConfig

use of org.apache.flink.table.planner.plan.nodes.exec.ExecNodeConfig in project flink by apache.

the class StreamExecGroupWindowAggregate method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    final boolean isCountWindow;
    if (window instanceof TumblingGroupWindow) {
        isCountWindow = hasRowIntervalType(((TumblingGroupWindow) window).size());
    } else if (window instanceof SlidingGroupWindow) {
        isCountWindow = hasRowIntervalType(((SlidingGroupWindow) window).size());
    } else {
        isCountWindow = false;
    }
    if (isCountWindow && grouping.length > 0 && config.getStateRetentionTime() < 0) {
        LOGGER.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 inputTimeFieldIndex;
    if (isRowtimeAttribute(window.timeAttribute())) {
        inputTimeFieldIndex = timeFieldIndex(FlinkTypeFactory.INSTANCE().buildRelNodeRowType(inputRowType), planner.getRelBuilder(), window.timeAttribute());
        if (inputTimeFieldIndex < 0) {
            throw new TableException("Group window must defined on a time attribute, " + "but the time attribute can't be found.\n" + "This should never happen. Please file an issue.");
        }
    } else {
        inputTimeFieldIndex = -1;
    }
    final ZoneId shiftTimeZone = TimeWindowUtil.getShiftTimeZone(window.timeAttribute().getOutputDataType().getLogicalType(), config.getLocalTimeZone());
    final boolean[] aggCallNeedRetractions = new boolean[aggCalls.length];
    Arrays.fill(aggCallNeedRetractions, needRetraction);
    final AggregateInfoList aggInfoList = transformToStreamAggregateInfoList(inputRowType, JavaScalaConversionUtil.toScala(Arrays.asList(aggCalls)), aggCallNeedRetractions, needRetraction, // isStateBackendDataViews
    true, // needDistinctInfo
    true);
    final GeneratedClass<?> aggCodeGenerator = createAggsHandler(aggInfoList, config, planner.getRelBuilder(), inputRowType.getChildren(), shiftTimeZone);
    final LogicalType[] aggResultTypes = extractLogicalTypes(aggInfoList.getActualValueTypes());
    final LogicalType[] windowPropertyTypes = Arrays.stream(namedWindowProperties).map(p -> p.getProperty().getResultType()).toArray(LogicalType[]::new);
    final EqualiserCodeGenerator generator = new EqualiserCodeGenerator(ArrayUtils.addAll(aggResultTypes, windowPropertyTypes));
    final GeneratedRecordEqualiser equaliser = generator.generateRecordEqualiser("WindowValueEqualiser");
    final LogicalType[] aggValueTypes = extractLogicalTypes(aggInfoList.getActualValueTypes());
    final LogicalType[] accTypes = extractLogicalTypes(aggInfoList.getAccTypes());
    final int inputCountIndex = aggInfoList.getIndexOfCountStar();
    final WindowOperator<?, ?> operator = createWindowOperator(config, aggCodeGenerator, equaliser, accTypes, windowPropertyTypes, aggValueTypes, inputRowType.getChildren().toArray(new LogicalType[0]), inputTimeFieldIndex, shiftTimeZone, inputCountIndex);
    final OneInputTransformation<RowData, RowData> transform = ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationMeta(GROUP_WINDOW_AGGREGATE_TRANSFORMATION, config), operator, InternalTypeInfo.of(getOutputType()), inputTransform.getParallelism());
    // set KeyType and Selector for state
    final RowDataKeySelector selector = KeySelectorUtil.getRowDataSelector(grouping, InternalTypeInfo.of(inputRowType));
    transform.setStateKeySelector(selector);
    transform.setStateKeyType(selector.getProducedType());
    return transform;
}

Example 2 with ExecNodeConfig

use of org.apache.flink.table.planner.plan.nodes.exec.ExecNodeConfig in project flink by apache.

the class StreamExecLegacyTableSourceScan method createConversionTransformationIfNeeded.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> createConversionTransformationIfNeeded(StreamExecutionEnvironment streamExecEnv, ExecNodeConfig config, Transformation<?> sourceTransform, @Nullable RexNode rowtimeExpression) {
    final RowType outputType = (RowType) getOutputType();
    final Transformation<RowData> transformation;
    final int[] fieldIndexes = computeIndexMapping(true);
    if (needInternalConversion(fieldIndexes)) {
        final String extractElement, resetElement;
        if (ScanUtil.hasTimeAttributeField(fieldIndexes)) {
            String elementTerm = OperatorCodeGenerator.ELEMENT();
            extractElement = String.format("ctx.%s = %s;", elementTerm, elementTerm);
            resetElement = String.format("ctx.%s = null;", elementTerm);
        } else {
            extractElement = "";
            resetElement = "";
        }
        final CodeGeneratorContext ctx = new CodeGeneratorContext(config.getTableConfig()).setOperatorBaseClass(TableStreamOperator.class);
        // the produced type may not carry the correct precision user defined in DDL, because
        // it may be converted from legacy type. Fix precision using logical schema from DDL.
        // Code generation requires the correct precision of input fields.
        final DataType fixedProducedDataType = TableSourceUtil.fixPrecisionForProducedDataType(tableSource, outputType);
        transformation = ScanUtil.convertToInternalRow(ctx, (Transformation<Object>) sourceTransform, fieldIndexes, fixedProducedDataType, outputType, qualifiedName, (detailName, simplifyName) -> createFormattedTransformationName(detailName, simplifyName, config), (description) -> createFormattedTransformationDescription(description, config), JavaScalaConversionUtil.toScala(Optional.ofNullable(rowtimeExpression)), extractElement, resetElement);
    } else {
        transformation = (Transformation<RowData>) sourceTransform;
    }
    final RelDataType relDataType = FlinkTypeFactory.INSTANCE().buildRelNodeRowType(outputType);
    final DataStream<RowData> ingestedTable = new DataStream<>(streamExecEnv, transformation);
    final Optional<RowtimeAttributeDescriptor> rowtimeDesc = JavaScalaConversionUtil.toJava(TableSourceUtil.getRowtimeAttributeDescriptor(tableSource, relDataType));
    final DataStream<RowData> withWatermarks = rowtimeDesc.map(desc -> {
        int rowtimeFieldIdx = relDataType.getFieldNames().indexOf(desc.getAttributeName());
        WatermarkStrategy strategy = desc.getWatermarkStrategy();
        if (strategy instanceof PeriodicWatermarkAssigner) {
            PeriodicWatermarkAssignerWrapper watermarkGenerator = new PeriodicWatermarkAssignerWrapper((PeriodicWatermarkAssigner) strategy, rowtimeFieldIdx);
            return ingestedTable.assignTimestampsAndWatermarks(watermarkGenerator);
        } else if (strategy instanceof PunctuatedWatermarkAssigner) {
            PunctuatedWatermarkAssignerWrapper watermarkGenerator = new PunctuatedWatermarkAssignerWrapper((PunctuatedWatermarkAssigner) strategy, rowtimeFieldIdx, tableSource.getProducedDataType());
            return ingestedTable.assignTimestampsAndWatermarks(watermarkGenerator);
        } else {
            // underlying DataStream.
            return ingestedTable;
        }
    }).orElse(// No need to generate watermarks if no rowtime
    ingestedTable);
    // attribute is specified.
    return withWatermarks.getTransformation();
}

Example 3 with ExecNodeConfig

use of org.apache.flink.table.planner.plan.nodes.exec.ExecNodeConfig in project flink by apache.

the class BatchExecPythonOverAggregate method getPythonOverWindowAggregateFunctionOperator.

@SuppressWarnings("unchecked")
private OneInputStreamOperator<RowData, RowData> getPythonOverWindowAggregateFunctionOperator(ExecNodeConfig config, Configuration pythonConfig, RowType inputRowType, RowType outputRowType, boolean[] isRangeWindows, int[] udafInputOffsets, PythonFunctionInfo[] pythonFunctionInfos) {
    Class<?> clazz = CommonPythonUtil.loadClass(ARROW_PYTHON_OVER_WINDOW_AGGREGATE_FUNCTION_OPERATOR_NAME);
    RowType udfInputType = (RowType) Projection.of(udafInputOffsets).project(inputRowType);
    RowType udfOutputType = (RowType) Projection.range(inputRowType.getFieldCount(), outputRowType.getFieldCount()).project(outputRowType);
    PartitionSpec partitionSpec = overSpec.getPartition();
    List<OverSpec.GroupSpec> groups = overSpec.getGroups();
    SortSpec sortSpec = groups.get(groups.size() - 1).getSort();
    try {
        Constructor<?> ctor = clazz.getConstructor(Configuration.class, PythonFunctionInfo[].class, RowType.class, RowType.class, RowType.class, long[].class, long[].class, boolean[].class, int[].class, int.class, boolean.class, GeneratedProjection.class, GeneratedProjection.class, GeneratedProjection.class);
        return (OneInputStreamOperator<RowData, RowData>) ctor.newInstance(pythonConfig, pythonFunctionInfos, inputRowType, udfInputType, udfOutputType, lowerBoundary.stream().mapToLong(i -> i).toArray(), upperBoundary.stream().mapToLong(i -> i).toArray(), isRangeWindows, aggWindowIndex.stream().mapToInt(i -> i).toArray(), sortSpec.getFieldIndices()[0], sortSpec.getAscendingOrders()[0], ProjectionCodeGenerator.generateProjection(CodeGeneratorContext.apply(config.getTableConfig()), "UdafInputProjection", inputRowType, udfInputType, udafInputOffsets), ProjectionCodeGenerator.generateProjection(CodeGeneratorContext.apply(config.getTableConfig()), "GroupKey", inputRowType, (RowType) Projection.of(partitionSpec.getFieldIndices()).project(inputRowType), partitionSpec.getFieldIndices()), ProjectionCodeGenerator.generateProjection(CodeGeneratorContext.apply(config.getTableConfig()), "GroupSet", inputRowType, (RowType) Projection.of(partitionSpec.getFieldIndices()).project(inputRowType), partitionSpec.getFieldIndices()));
    } catch (NoSuchMethodException | InstantiationException | IllegalAccessException | InvocationTargetException e) {
        throw new TableException("Python BatchArrowPythonOverWindowAggregateFunctionOperator constructed failed.", e);
    }
}

Example 4 with ExecNodeConfig

use of org.apache.flink.table.planner.plan.nodes.exec.ExecNodeConfig in project flink by apache.

the class BatchExecMultipleInput method translateToPlanInternal.

@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    final List<Transformation<?>> inputTransforms = new ArrayList<>();
    for (ExecEdge inputEdge : getInputEdges()) {
        inputTransforms.add(inputEdge.translateToPlan(planner));
    }
    final Transformation<?> outputTransform = rootNode.translateToPlan(planner);
    final int[] readOrders = getInputProperties().stream().map(InputProperty::getPriority).mapToInt(i -> i).toArray();
    final TableOperatorWrapperGenerator generator = new TableOperatorWrapperGenerator(inputTransforms, outputTransform, readOrders);
    generator.generate();
    final List<Pair<Transformation<?>, InputSpec>> inputTransformAndInputSpecPairs = generator.getInputTransformAndInputSpecPairs();
    final MultipleInputTransformation<RowData> multipleInputTransform = new MultipleInputTransformation<>(createTransformationName(config), new BatchMultipleInputStreamOperatorFactory(inputTransformAndInputSpecPairs.stream().map(Pair::getValue).collect(Collectors.toList()), generator.getHeadWrappers(), generator.getTailWrapper()), InternalTypeInfo.of(getOutputType()), generator.getParallelism());
    multipleInputTransform.setDescription(createTransformationDescription(config));
    inputTransformAndInputSpecPairs.forEach(input -> multipleInputTransform.addInput(input.getKey()));
    if (generator.getMaxParallelism() > 0) {
        multipleInputTransform.setMaxParallelism(generator.getMaxParallelism());
    }
    // set resources
    multipleInputTransform.setResources(generator.getMinResources(), generator.getPreferredResources());
    final int memoryWeight = generator.getManagedMemoryWeight();
    final long memoryBytes = (long) memoryWeight << 20;
    ExecNodeUtil.setManagedMemoryWeight(multipleInputTransform, memoryBytes);
    // set chaining strategy for source chaining
    multipleInputTransform.setChainingStrategy(ChainingStrategy.HEAD_WITH_SOURCES);
    return multipleInputTransform;
}

Example 5 with ExecNodeConfig

use of org.apache.flink.table.planner.plan.nodes.exec.ExecNodeConfig in project flink by apache.

the class BatchExecPythonGroupWindowAggregate method createPythonOneInputTransformation.

private OneInputTransformation<RowData, RowData> createPythonOneInputTransformation(Transformation<RowData> inputTransform, RowType inputRowType, RowType outputRowType, int maxLimitSize, long windowSize, long slideSize, Configuration pythonConfig, ExecNodeConfig config) {
    int[] namePropertyTypeArray = Arrays.stream(namedWindowProperties).mapToInt(p -> {
        WindowProperty property = p.getProperty();
        if (property instanceof WindowStart) {
            return 0;
        }
        if (property instanceof WindowEnd) {
            return 1;
        }
        if (property instanceof RowtimeAttribute) {
            return 2;
        }
        throw new TableException("Unexpected property " + property);
    }).toArray();
    Tuple2<int[], PythonFunctionInfo[]> aggInfos = CommonPythonUtil.extractPythonAggregateFunctionInfosFromAggregateCall(aggCalls);
    int[] pythonUdafInputOffsets = aggInfos.f0;
    PythonFunctionInfo[] pythonFunctionInfos = aggInfos.f1;
    OneInputStreamOperator<RowData, RowData> pythonOperator = getPythonGroupWindowAggregateFunctionOperator(config, pythonConfig, inputRowType, outputRowType, maxLimitSize, windowSize, slideSize, namePropertyTypeArray, pythonUdafInputOffsets, pythonFunctionInfos);
    return ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationName(config), createTransformationDescription(config), pythonOperator, InternalTypeInfo.of(outputRowType), inputTransform.getParallelism());
}