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

Example 11 with Transformation

use of org.apache.flink.api.dag.Transformation 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 12 with Transformation

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

Example 13 with Transformation

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

Example 14 with Transformation

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

the class StreamExecIncrementalGroupAggregate 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 AggregateInfoList partialLocalAggInfoList = AggregateUtil.createPartialAggInfoList(partialLocalAggInputType, JavaScalaConversionUtil.toScala(Arrays.asList(partialOriginalAggCalls)), partialAggCallNeedRetractions, partialAggNeedRetraction, false);
    final GeneratedAggsHandleFunction partialAggsHandler = generateAggsHandler("PartialGroupAggsHandler", partialLocalAggInfoList, partialAggGrouping.length, partialLocalAggInfoList.getAccTypes(), config, planner.getRelBuilder(), // the partial aggregate accumulators will be buffered, so need copy
    true);
    final AggregateInfoList incrementalAggInfo = AggregateUtil.createIncrementalAggInfoList(partialLocalAggInputType, JavaScalaConversionUtil.toScala(Arrays.asList(partialOriginalAggCalls)), partialAggCallNeedRetractions, partialAggNeedRetraction);
    final GeneratedAggsHandleFunction finalAggsHandler = generateAggsHandler("FinalGroupAggsHandler", incrementalAggInfo, 0, partialLocalAggInfoList.getAccTypes(), config, planner.getRelBuilder(), // the final aggregate accumulators is not buffered
    false);
    final RowDataKeySelector partialKeySelector = KeySelectorUtil.getRowDataSelector(partialAggGrouping, InternalTypeInfo.of(inputEdge.getOutputType()));
    final RowDataKeySelector finalKeySelector = KeySelectorUtil.getRowDataSelector(finalAggGrouping, partialKeySelector.getProducedType());
    final MiniBatchIncrementalGroupAggFunction aggFunction = new MiniBatchIncrementalGroupAggFunction(partialAggsHandler, finalAggsHandler, finalKeySelector, config.getStateRetentionTime());
    final OneInputStreamOperator<RowData, RowData> operator = new KeyedMapBundleOperator<>(aggFunction, AggregateUtil.createMiniBatchTrigger(config));
    // partitioned aggregation
    final OneInputTransformation<RowData, RowData> transform = ExecNodeUtil.createOneInputTransformation(inputTransform, createTransformationMeta(INCREMENTAL_GROUP_AGGREGATE_TRANSFORMATION, config), operator, InternalTypeInfo.of(getOutputType()), inputTransform.getParallelism());
    // set KeyType and Selector for state
    transform.setStateKeySelector(partialKeySelector);
    transform.setStateKeyType(partialKeySelector.getProducedType());
    return transform;
}

Example 15 with Transformation

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