Search in sources :

Example 11 with TableException

use of org.apache.flink.table.api.TableException in project flink by apache.

the class StreamExecMatch 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();
    checkOrderKeys(inputRowType);
    final EventComparator<RowData> eventComparator = createEventComparator(config, inputRowType);
    final Transformation<RowData> timestampedInputTransform = translateOrder(inputTransform, inputRowType);
    final Tuple2<Pattern<RowData, RowData>, List<String>> cepPatternAndNames = translatePattern(matchSpec, config.getTableConfig(), planner.getRelBuilder(), inputRowType);
    final Pattern<RowData, RowData> cepPattern = cepPatternAndNames.f0;
    // TODO remove this once it is supported in CEP library
    if (NFACompiler.canProduceEmptyMatches(cepPattern)) {
        throw new TableException("Patterns that can produce empty matches are not supported. There must be at least one non-optional state.");
    }
    // TODO remove this once it is supported in CEP library
    if (cepPattern.getQuantifier().hasProperty(Quantifier.QuantifierProperty.GREEDY)) {
        throw new TableException("Greedy quantifiers are not allowed as the last element of a Pattern yet. " + "Finish your pattern with either a simple variable or reluctant quantifier.");
    }
    if (matchSpec.isAllRows()) {
        throw new TableException("All rows per match mode is not supported yet.");
    }
    final int[] partitionKeys = matchSpec.getPartition().getFieldIndices();
    final SortSpec.SortFieldSpec timeOrderField = matchSpec.getOrderKeys().getFieldSpec(0);
    final LogicalType timeOrderFieldType = inputRowType.getTypeAt(timeOrderField.getFieldIndex());
    final boolean isProctime = TypeCheckUtils.isProcTime(timeOrderFieldType);
    final InternalTypeInfo<RowData> inputTypeInfo = (InternalTypeInfo<RowData>) inputTransform.getOutputType();
    final TypeSerializer<RowData> inputSerializer = inputTypeInfo.createSerializer(planner.getExecEnv().getConfig());
    final NFACompiler.NFAFactory<RowData> nfaFactory = NFACompiler.compileFactory(cepPattern, false);
    final MatchCodeGenerator generator = new MatchCodeGenerator(new CodeGeneratorContext(config.getTableConfig()), planner.getRelBuilder(), // nullableInput
    false, JavaScalaConversionUtil.toScala(cepPatternAndNames.f1), JavaScalaConversionUtil.toScala(Optional.empty()), CodeGenUtils.DEFAULT_COLLECTOR_TERM());
    generator.bindInput(inputRowType, CodeGenUtils.DEFAULT_INPUT1_TERM(), JavaScalaConversionUtil.toScala(Optional.empty()));
    final PatternProcessFunctionRunner patternProcessFunction = generator.generateOneRowPerMatchExpression((RowType) getOutputType(), partitionKeys, matchSpec.getMeasures());
    final CepOperator<RowData, RowData, RowData> operator = new CepOperator<>(inputSerializer, isProctime, nfaFactory, eventComparator, cepPattern.getAfterMatchSkipStrategy(), patternProcessFunction, null);
    final OneInputTransformation<RowData, RowData> transform = ExecNodeUtil.createOneInputTransformation(timestampedInputTransform, createTransformationMeta(MATCH_TRANSFORMATION, config), operator, InternalTypeInfo.of(getOutputType()), timestampedInputTransform.getParallelism());
    final RowDataKeySelector selector = KeySelectorUtil.getRowDataSelector(partitionKeys, inputTypeInfo);
    transform.setStateKeySelector(selector);
    transform.setStateKeyType(selector.getProducedType());
    if (inputsContainSingleton()) {
        transform.setParallelism(1);
        transform.setMaxParallelism(1);
    }
    return transform;
}
Also used : 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) NFACompiler(org.apache.flink.cep.nfa.compiler.NFACompiler) RowData(org.apache.flink.table.data.RowData) RowDataKeySelector(org.apache.flink.table.runtime.keyselector.RowDataKeySelector) List(java.util.List) ArrayList(java.util.ArrayList) Pattern(org.apache.flink.cep.pattern.Pattern) TableException(org.apache.flink.table.api.TableException) CodeGeneratorContext(org.apache.flink.table.planner.codegen.CodeGeneratorContext) InternalTypeInfo(org.apache.flink.table.runtime.typeutils.InternalTypeInfo) MatchCodeGenerator(org.apache.flink.table.planner.codegen.MatchCodeGenerator) PatternProcessFunctionRunner(org.apache.flink.table.runtime.operators.match.PatternProcessFunctionRunner) CepOperator(org.apache.flink.cep.operator.CepOperator) SortSpec(org.apache.flink.table.planner.plan.nodes.exec.spec.SortSpec)

Example 12 with TableException

use of org.apache.flink.table.api.TableException 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 13 with TableException

use of org.apache.flink.table.api.TableException 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 14 with TableException

use of org.apache.flink.table.api.TableException in project flink by apache.

the class StreamExecPythonOverAggregate method getPythonOverWindowAggregateFunctionOperator.

@SuppressWarnings("unchecked")
private OneInputStreamOperator<RowData, RowData> getPythonOverWindowAggregateFunctionOperator(ExecNodeConfig config, Configuration pythonConfig, RowType inputRowType, RowType outputRowType, int rowTiemIdx, long lowerBoundary, boolean isRowsClause, int[] udafInputOffsets, PythonFunctionInfo[] pythonFunctionInfos, long minIdleStateRetentionTime, long maxIdleStateRetentionTime) {
    RowType userDefinedFunctionInputType = (RowType) Projection.of(udafInputOffsets).project(inputRowType);
    RowType userDefinedFunctionOutputType = (RowType) Projection.range(inputRowType.getFieldCount(), outputRowType.getFieldCount()).project(outputRowType);
    GeneratedProjection generatedProjection = ProjectionCodeGenerator.generateProjection(CodeGeneratorContext.apply(config.getTableConfig()), "UdafInputProjection", inputRowType, userDefinedFunctionInputType, udafInputOffsets);
    if (isRowsClause) {
        String className;
        if (rowTiemIdx != -1) {
            className = ARROW_PYTHON_OVER_WINDOW_ROWS_ROW_TIME_AGGREGATE_FUNCTION_OPERATOR_NAME;
        } else {
            className = ARROW_PYTHON_OVER_WINDOW_ROWS_PROC_TIME_AGGREGATE_FUNCTION_OPERATOR_NAME;
        }
        Class<?> clazz = CommonPythonUtil.loadClass(className);
        try {
            Constructor<?> ctor = clazz.getConstructor(Configuration.class, long.class, long.class, PythonFunctionInfo[].class, RowType.class, RowType.class, RowType.class, int.class, long.class, GeneratedProjection.class);
            return (OneInputStreamOperator<RowData, RowData>) ctor.newInstance(pythonConfig, minIdleStateRetentionTime, maxIdleStateRetentionTime, pythonFunctionInfos, inputRowType, userDefinedFunctionInputType, userDefinedFunctionOutputType, rowTiemIdx, lowerBoundary, generatedProjection);
        } catch (NoSuchMethodException | InstantiationException | IllegalAccessException | InvocationTargetException e) {
            throw new TableException("Python Arrow Over Rows Window Function Operator constructed failed.", e);
        }
    } else {
        String className;
        if (rowTiemIdx != -1) {
            className = ARROW_PYTHON_OVER_WINDOW_RANGE_ROW_TIME_AGGREGATE_FUNCTION_OPERATOR_NAME;
        } else {
            className = ARROW_PYTHON_OVER_WINDOW_RANGE_PROC_TIME_AGGREGATE_FUNCTION_OPERATOR_NAME;
        }
        Class<?> clazz = CommonPythonUtil.loadClass(className);
        try {
            Constructor<?> ctor = clazz.getConstructor(Configuration.class, PythonFunctionInfo[].class, RowType.class, RowType.class, RowType.class, int.class, long.class, GeneratedProjection.class);
            return (OneInputStreamOperator<RowData, RowData>) ctor.newInstance(pythonConfig, pythonFunctionInfos, inputRowType, userDefinedFunctionInputType, userDefinedFunctionOutputType, rowTiemIdx, lowerBoundary, generatedProjection);
        } catch (NoSuchMethodException | InstantiationException | IllegalAccessException | InvocationTargetException e) {
            throw new TableException("Python Arrow Over Range Window Function Operator constructed failed.", e);
        }
    }
}
Also used : PythonFunctionInfo(org.apache.flink.table.functions.python.PythonFunctionInfo) TableException(org.apache.flink.table.api.TableException) RowType(org.apache.flink.table.types.logical.RowType) InvocationTargetException(java.lang.reflect.InvocationTargetException) OneInputStreamOperator(org.apache.flink.streaming.api.operators.OneInputStreamOperator) GeneratedProjection(org.apache.flink.table.runtime.generated.GeneratedProjection)

Example 15 with TableException

use of org.apache.flink.table.api.TableException 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)

Aggregations

TableException (org.apache.flink.table.api.TableException)163 RowData (org.apache.flink.table.data.RowData)35 RowType (org.apache.flink.table.types.logical.RowType)35 Transformation (org.apache.flink.api.dag.Transformation)28 ArrayList (java.util.ArrayList)27 ExecEdge (org.apache.flink.table.planner.plan.nodes.exec.ExecEdge)24 LogicalType (org.apache.flink.table.types.logical.LogicalType)24 List (java.util.List)22 DataType (org.apache.flink.table.types.DataType)19 OneInputTransformation (org.apache.flink.streaming.api.transformations.OneInputTransformation)18 ValidationException (org.apache.flink.table.api.ValidationException)17 IOException (java.io.IOException)13 AggregateCall (org.apache.calcite.rel.core.AggregateCall)13 ValueLiteralExpression (org.apache.flink.table.expressions.ValueLiteralExpression)13 RowDataKeySelector (org.apache.flink.table.runtime.keyselector.RowDataKeySelector)13 Optional (java.util.Optional)11 Configuration (org.apache.flink.configuration.Configuration)11 StreamExecutionEnvironment (org.apache.flink.streaming.api.environment.StreamExecutionEnvironment)11 Constructor (java.lang.reflect.Constructor)10 Arrays (java.util.Arrays)9