Examples with Transformation org.apache.flink.api.dag.Transformation used on opensource projects

Example 31 with Transformation

use of org.apache.flink.api.dag.Transformation 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());
}

Example 32 with Transformation

use of org.apache.flink.api.dag.Transformation in project flink by apache.

the class BatchExecPythonGroupWindowAggregate 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 RowType outputRowType = InternalTypeInfo.of(getOutputType()).toRowType();
    final Tuple2<Long, Long> windowSizeAndSlideSize = WindowCodeGenerator.getWindowDef(window);
    final Configuration pythonConfig = CommonPythonUtil.getMergedConfig(planner.getExecEnv(), config.getTableConfig());
    int groupBufferLimitSize = pythonConfig.getInteger(ExecutionConfigOptions.TABLE_EXEC_WINDOW_AGG_BUFFER_SIZE_LIMIT);
    OneInputTransformation<RowData, RowData> transform = createPythonOneInputTransformation(inputTransform, inputRowType, outputRowType, groupBufferLimitSize, windowSizeAndSlideSize.f0, windowSizeAndSlideSize.f1, pythonConfig, config);
    if (CommonPythonUtil.isPythonWorkerUsingManagedMemory(pythonConfig)) {
        transform.declareManagedMemoryUseCaseAtSlotScope(ManagedMemoryUseCase.PYTHON);
    }
    return transform;
}

Example 33 with Transformation

use of org.apache.flink.api.dag.Transformation in project flink by apache.

the class BatchExecRank method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    ExecEdge inputEdge = getInputEdges().get(0);
    Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
    RowType inputType = (RowType) inputEdge.getOutputType();
    // operator needn't cache data
    // The collation for the partition-by and order-by fields is inessential here,
    // we only use the comparator to distinguish fields change.
    RankOperator operator = new RankOperator(ComparatorCodeGenerator.gen(config.getTableConfig(), "PartitionByComparator", inputType, SortUtil.getAscendingSortSpec(partitionFields)), ComparatorCodeGenerator.gen(config.getTableConfig(), "OrderByComparator", inputType, SortUtil.getAscendingSortSpec(sortFields)), rankStart, rankEnd, outputRankNumber);
    return ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationName(config), createTransformationDescription(config), SimpleOperatorFactory.of(operator), InternalTypeInfo.of((RowType) getOutputType()), inputTransform.getParallelism());
}

Example 34 with Transformation

use of org.apache.flink.api.dag.Transformation in project flink by apache.

the class BatchExecSortMergeJoin method translateToPlanInternal.

@Override
@SuppressWarnings("unchecked")
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    ExecEdge leftInputEdge = getInputEdges().get(0);
    ExecEdge rightInputEdge = getInputEdges().get(1);
    // get input types
    RowType leftType = (RowType) leftInputEdge.getOutputType();
    RowType rightType = (RowType) rightInputEdge.getOutputType();
    LogicalType[] keyFieldTypes = IntStream.of(leftKeys).mapToObj(leftType::getTypeAt).toArray(LogicalType[]::new);
    RowType keyType = RowType.of(keyFieldTypes);
    GeneratedJoinCondition condFunc = JoinUtil.generateConditionFunction(config.getTableConfig(), nonEquiCondition, leftType, rightType);
    long externalBufferMemory = config.get(ExecutionConfigOptions.TABLE_EXEC_RESOURCE_EXTERNAL_BUFFER_MEMORY).getBytes();
    long sortMemory = config.get(ExecutionConfigOptions.TABLE_EXEC_RESOURCE_SORT_MEMORY).getBytes();
    int externalBufferNum = 1;
    if (joinType == FlinkJoinType.FULL) {
        externalBufferNum = 2;
    }
    long managedMemory = externalBufferMemory * externalBufferNum + sortMemory * 2;
    SortCodeGenerator leftSortGen = newSortGen(config, leftKeys, leftType);
    SortCodeGenerator rightSortGen = newSortGen(config, rightKeys, rightType);
    int[] keyPositions = IntStream.range(0, leftKeys.length).toArray();
    SortMergeJoinOperator operator = new SortMergeJoinOperator(1.0 * externalBufferMemory / managedMemory, joinType, leftIsSmaller, condFunc, ProjectionCodeGenerator.generateProjection(new CodeGeneratorContext(config.getTableConfig()), "SMJProjection", leftType, keyType, leftKeys), ProjectionCodeGenerator.generateProjection(new CodeGeneratorContext(config.getTableConfig()), "SMJProjection", rightType, keyType, rightKeys), leftSortGen.generateNormalizedKeyComputer("LeftComputer"), leftSortGen.generateRecordComparator("LeftComparator"), rightSortGen.generateNormalizedKeyComputer("RightComputer"), rightSortGen.generateRecordComparator("RightComparator"), newSortGen(config, keyPositions, keyType).generateRecordComparator("KeyComparator"), filterNulls);
    Transformation<RowData> leftInputTransform = (Transformation<RowData>) leftInputEdge.translateToPlan(planner);
    Transformation<RowData> rightInputTransform = (Transformation<RowData>) rightInputEdge.translateToPlan(planner);
    return ExecNodeUtil.createTwoInputTransformation(leftInputTransform, rightInputTransform, createTransformationName(config), createTransformationDescription(config), SimpleOperatorFactory.of(operator), InternalTypeInfo.of(getOutputType()), rightInputTransform.getParallelism(), managedMemory);
}

Example 35 with Transformation

use of org.apache.flink.api.dag.Transformation in project flink by apache.

the class BatchExecSort method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    ExecEdge inputEdge = getInputEdges().get(0);
    Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
    RowType inputType = (RowType) inputEdge.getOutputType();
    SortCodeGenerator codeGen = new SortCodeGenerator(config.getTableConfig(), inputType, sortSpec);
    SortOperator operator = new SortOperator(codeGen.generateNormalizedKeyComputer("BatchExecSortComputer"), codeGen.generateRecordComparator("BatchExecSortComparator"));
    long sortMemory = config.get(ExecutionConfigOptions.TABLE_EXEC_RESOURCE_SORT_MEMORY).getBytes();
    return ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationName(config), createTransformationDescription(config), SimpleOperatorFactory.of(operator), InternalTypeInfo.of((RowType) getOutputType()), inputTransform.getParallelism(), sortMemory);
}