Examples with SingleOutputStreamOperator org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator used on opensource projects

Example 11 with SingleOutputStreamOperator

use of org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator in project beam by apache.

the class FlinkStreamingPortablePipelineTranslator method translateExecutableStage.

private <InputT, OutputT> void translateExecutableStage(String id, RunnerApi.Pipeline pipeline, StreamingTranslationContext context) {
    // TODO: Fail on splittable DoFns.
    // TODO: Special-case single outputs to avoid multiplexing PCollections.
    RunnerApi.Components components = pipeline.getComponents();
    RunnerApi.PTransform transform = components.getTransformsOrThrow(id);
    Map<String, String> outputs = transform.getOutputsMap();
    final RunnerApi.ExecutableStagePayload stagePayload;
    try {
        stagePayload = RunnerApi.ExecutableStagePayload.parseFrom(transform.getSpec().getPayload());
    } catch (IOException e) {
        throw new RuntimeException(e);
    }
    String inputPCollectionId = stagePayload.getInput();
    final TransformedSideInputs transformedSideInputs;
    if (stagePayload.getSideInputsCount() > 0) {
        transformedSideInputs = transformSideInputs(stagePayload, components, context);
    } else {
        transformedSideInputs = new TransformedSideInputs(Collections.emptyMap(), null);
    }
    Map<TupleTag<?>, OutputTag<WindowedValue<?>>> tagsToOutputTags = Maps.newLinkedHashMap();
    Map<TupleTag<?>, Coder<WindowedValue<?>>> tagsToCoders = Maps.newLinkedHashMap();
    // TODO: does it matter which output we designate as "main"
    final TupleTag<OutputT> mainOutputTag = outputs.isEmpty() ? null : new TupleTag(outputs.keySet().iterator().next());
    // associate output tags with ids, output manager uses these Integer ids to serialize state
    BiMap<String, Integer> outputIndexMap = createOutputMap(outputs.keySet());
    Map<String, Coder<WindowedValue<?>>> outputCoders = Maps.newHashMap();
    Map<TupleTag<?>, Integer> tagsToIds = Maps.newHashMap();
    Map<String, TupleTag<?>> collectionIdToTupleTag = Maps.newHashMap();
    // order output names for deterministic mapping
    for (String localOutputName : new TreeMap<>(outputIndexMap).keySet()) {
        String collectionId = outputs.get(localOutputName);
        Coder<WindowedValue<?>> windowCoder = (Coder) instantiateCoder(collectionId, components);
        outputCoders.put(localOutputName, windowCoder);
        TupleTag<?> tupleTag = new TupleTag<>(localOutputName);
        CoderTypeInformation<WindowedValue<?>> typeInformation = new CoderTypeInformation(windowCoder, context.getPipelineOptions());
        tagsToOutputTags.put(tupleTag, new OutputTag<>(localOutputName, typeInformation));
        tagsToCoders.put(tupleTag, windowCoder);
        tagsToIds.put(tupleTag, outputIndexMap.get(localOutputName));
        collectionIdToTupleTag.put(collectionId, tupleTag);
    }
    final SingleOutputStreamOperator<WindowedValue<OutputT>> outputStream;
    DataStream<WindowedValue<InputT>> inputDataStream = context.getDataStreamOrThrow(inputPCollectionId);
    CoderTypeInformation<WindowedValue<OutputT>> outputTypeInformation = !outputs.isEmpty() ? new CoderTypeInformation(outputCoders.get(mainOutputTag.getId()), context.getPipelineOptions()) : null;
    ArrayList<TupleTag<?>> additionalOutputTags = Lists.newArrayList();
    for (TupleTag<?> tupleTag : tagsToCoders.keySet()) {
        if (!mainOutputTag.getId().equals(tupleTag.getId())) {
            additionalOutputTags.add(tupleTag);
        }
    }
    final Coder<WindowedValue<InputT>> windowedInputCoder = instantiateCoder(inputPCollectionId, components);
    final boolean stateful = stagePayload.getUserStatesCount() > 0 || stagePayload.getTimersCount() > 0;
    final boolean hasSdfProcessFn = stagePayload.getComponents().getTransformsMap().values().stream().anyMatch(pTransform -> pTransform.getSpec().getUrn().equals(PTransformTranslation.SPLITTABLE_PROCESS_SIZED_ELEMENTS_AND_RESTRICTIONS_URN));
    Coder keyCoder = null;
    KeySelector<WindowedValue<InputT>, ?> keySelector = null;
    if (stateful || hasSdfProcessFn) {
        // Stateful/SDF stages are only allowed of KV input.
        Coder valueCoder = ((WindowedValue.FullWindowedValueCoder) windowedInputCoder).getValueCoder();
        if (!(valueCoder instanceof KvCoder)) {
            throw new IllegalStateException(String.format(Locale.ENGLISH, "The element coder for stateful DoFn '%s' must be KvCoder but is: %s", inputPCollectionId, valueCoder.getClass().getSimpleName()));
        }
        if (stateful) {
            keyCoder = ((KvCoder) valueCoder).getKeyCoder();
            keySelector = new KvToByteBufferKeySelector(keyCoder, new SerializablePipelineOptions(context.getPipelineOptions()));
        } else {
            // as the key.
            if (!(((KvCoder) valueCoder).getKeyCoder() instanceof KvCoder)) {
                throw new IllegalStateException(String.format(Locale.ENGLISH, "The element coder for splittable DoFn '%s' must be KVCoder(KvCoder, DoubleCoder) but is: %s", inputPCollectionId, valueCoder.getClass().getSimpleName()));
            }
            keyCoder = ((KvCoder) ((KvCoder) valueCoder).getKeyCoder()).getKeyCoder();
            keySelector = new SdfByteBufferKeySelector(keyCoder, new SerializablePipelineOptions(context.getPipelineOptions()));
        }
        inputDataStream = inputDataStream.keyBy(keySelector);
    }
    DoFnOperator.MultiOutputOutputManagerFactory<OutputT> outputManagerFactory = new DoFnOperator.MultiOutputOutputManagerFactory<>(mainOutputTag, tagsToOutputTags, tagsToCoders, tagsToIds, new SerializablePipelineOptions(context.getPipelineOptions()));
    DoFnOperator<InputT, OutputT> doFnOperator = new ExecutableStageDoFnOperator<>(transform.getUniqueName(), windowedInputCoder, Collections.emptyMap(), mainOutputTag, additionalOutputTags, outputManagerFactory, transformedSideInputs.unionTagToView, new ArrayList<>(transformedSideInputs.unionTagToView.values()), getSideInputIdToPCollectionViewMap(stagePayload, components), context.getPipelineOptions(), stagePayload, context.getJobInfo(), FlinkExecutableStageContextFactory.getInstance(), collectionIdToTupleTag, getWindowingStrategy(inputPCollectionId, components), keyCoder, keySelector);
    final String operatorName = generateNameFromStagePayload(stagePayload);
    if (transformedSideInputs.unionTagToView.isEmpty()) {
        outputStream = inputDataStream.transform(operatorName, outputTypeInformation, doFnOperator);
    } else {
        DataStream<RawUnionValue> sideInputStream = transformedSideInputs.unionedSideInputs.broadcast();
        if (stateful || hasSdfProcessFn) {
            // We have to manually construct the two-input transform because we're not
            // allowed to have only one input keyed, normally. Since Flink 1.5.0 it's
            // possible to use the Broadcast State Pattern which provides a more elegant
            // way to process keyed main input with broadcast state, but it's not feasible
            // here because it breaks the DoFnOperator abstraction.
            TwoInputTransformation<WindowedValue<KV<?, InputT>>, RawUnionValue, WindowedValue<OutputT>> rawFlinkTransform = new TwoInputTransformation(inputDataStream.getTransformation(), sideInputStream.getTransformation(), transform.getUniqueName(), doFnOperator, outputTypeInformation, inputDataStream.getParallelism());
            rawFlinkTransform.setStateKeyType(((KeyedStream) inputDataStream).getKeyType());
            rawFlinkTransform.setStateKeySelectors(((KeyedStream) inputDataStream).getKeySelector(), null);
            outputStream = new SingleOutputStreamOperator(inputDataStream.getExecutionEnvironment(), // we have to cheat around the ctor being protected
            rawFlinkTransform) {
            };
        } else {
            outputStream = inputDataStream.connect(sideInputStream).transform(operatorName, outputTypeInformation, doFnOperator);
        }
    }
    // Assign a unique but consistent id to re-map operator state
    outputStream.uid(transform.getUniqueName());
    if (mainOutputTag != null) {
        context.addDataStream(outputs.get(mainOutputTag.getId()), outputStream);
    }
    for (TupleTag<?> tupleTag : additionalOutputTags) {
        context.addDataStream(outputs.get(tupleTag.getId()), outputStream.getSideOutput(tagsToOutputTags.get(tupleTag)));
    }
}

Example 12 with SingleOutputStreamOperator

use of org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator in project flink by apache.

the class TypeFillTest method test.

@Test
public void test() {
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    try {
        env.addSource(new TestSource<Integer>()).print();
        fail();
    } catch (Exception ignored) {
    }
    DataStream<Long> source = env.generateSequence(1, 10);
    try {
        source.map(new TestMap<Long, Long>()).print();
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.flatMap(new TestFlatMap<Long, Long>()).print();
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.connect(source).map(new TestCoMap<Long, Long, Integer>()).print();
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.connect(source).flatMap(new TestCoFlatMap<Long, Long, Integer>()).print();
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.keyBy(new TestKeySelector<Long, String>()).print();
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.connect(source).keyBy(new TestKeySelector<Long, String>(), new TestKeySelector<>());
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.coGroup(source).where(new TestKeySelector<>()).equalTo(new TestKeySelector<>());
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.join(source).where(new TestKeySelector<>()).equalTo(new TestKeySelector<>());
        fail();
    } catch (Exception ignored) {
    }
    try {
        source.keyBy((in) -> in).intervalJoin(source.keyBy((in) -> in)).between(Time.milliseconds(10L), Time.milliseconds(10L)).process(new TestProcessJoinFunction<>()).print();
        fail();
    } catch (Exception ignored) {
    }
    env.addSource(new TestSource<Integer>()).returns(Integer.class);
    source.map(new TestMap<Long, Long>()).returns(Long.class).print();
    source.flatMap(new TestFlatMap<Long, Long>()).returns(new TypeHint<Long>() {
    }).print();
    source.connect(source).map(new TestCoMap<Long, Long, Integer>()).returns(BasicTypeInfo.INT_TYPE_INFO).print();
    source.connect(source).flatMap(new TestCoFlatMap<Long, Long, Integer>()).returns(BasicTypeInfo.INT_TYPE_INFO).print();
    source.connect(source).keyBy(new TestKeySelector<>(), new TestKeySelector<>(), Types.STRING);
    source.coGroup(source).where(new TestKeySelector<>(), Types.STRING).equalTo(new TestKeySelector<>(), Types.STRING);
    source.join(source).where(new TestKeySelector<>(), Types.STRING).equalTo(new TestKeySelector<>(), Types.STRING);
    source.keyBy((in) -> in).intervalJoin(source.keyBy((in) -> in)).between(Time.milliseconds(10L), Time.milliseconds(10L)).process(new TestProcessJoinFunction<Long, Long, String>()).returns(Types.STRING);
    source.keyBy((in) -> in).intervalJoin(source.keyBy((in) -> in)).between(Time.milliseconds(10L), Time.milliseconds(10L)).process(new TestProcessJoinFunction<>(), Types.STRING);
    assertEquals(BasicTypeInfo.LONG_TYPE_INFO, source.map(new TestMap<Long, Long>()).returns(Long.class).getType());
    SingleOutputStreamOperator<String> map = source.map(new MapFunction<Long, String>() {

        @Override
        public String map(Long value) throws Exception {
            return null;
        }
    });
    map.print();
    try {
        map.returns(String.class);
        fail();
    } catch (Exception ignored) {
    }
}

Example 13 with SingleOutputStreamOperator

use of org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator in project flink by apache.

the class StateDescriptorPassingTest method testReduceWindowAllState.

@Test
public void testReduceWindowAllState() {
    final StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    env.registerTypeWithKryoSerializer(File.class, JavaSerializer.class);
    // simulate ingestion time
    DataStream<File> src = env.fromElements(new File("/")).assignTimestampsAndWatermarks(WatermarkStrategy.<File>forMonotonousTimestamps().withTimestampAssigner((file, ts) -> System.currentTimeMillis()));
    SingleOutputStreamOperator<?> result = src.windowAll(TumblingEventTimeWindows.of(Time.milliseconds(1000))).reduce(new ReduceFunction<File>() {

        @Override
        public File reduce(File value1, File value2) {
            return null;
        }
    });
    validateStateDescriptorConfigured(result);
}

Example 14 with SingleOutputStreamOperator

use of org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator in project flink by apache.

the class StateDescriptorPassingTest method testProcessAllWindowState.

@Test
public void testProcessAllWindowState() {
    final StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    env.registerTypeWithKryoSerializer(File.class, JavaSerializer.class);
    // simulate ingestion time
    DataStream<File> src = env.fromElements(new File("/")).assignTimestampsAndWatermarks(WatermarkStrategy.<File>forMonotonousTimestamps().withTimestampAssigner((file, ts) -> System.currentTimeMillis()));
    SingleOutputStreamOperator<?> result = src.windowAll(TumblingEventTimeWindows.of(Time.milliseconds(1000))).process(new ProcessAllWindowFunction<File, String, TimeWindow>() {

        @Override
        public void process(Context ctx, Iterable<File> input, Collector<String> out) {
        }
    });
    validateListStateDescriptorConfigured(result);
}

Example 15 with SingleOutputStreamOperator

use of org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator in project flink by apache.

the class SortingBoundedInputITCase method testTwoInputOperator.

@Test
public void testTwoInputOperator() {
    long numberOfRecords = 500_000;
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    Configuration config = new Configuration();
    config.set(ExecutionOptions.RUNTIME_MODE, RuntimeExecutionMode.BATCH);
    env.configure(config, this.getClass().getClassLoader());
    DataStreamSource<Tuple2<Integer, byte[]>> elements1 = env.fromParallelCollection(new InputGenerator(numberOfRecords), new TupleTypeInfo<>(BasicTypeInfo.INT_TYPE_INFO, PrimitiveArrayTypeInfo.BYTE_PRIMITIVE_ARRAY_TYPE_INFO));
    DataStreamSource<Tuple2<Integer, byte[]>> elements2 = env.fromParallelCollection(new InputGenerator(numberOfRecords), new TupleTypeInfo<>(BasicTypeInfo.INT_TYPE_INFO, PrimitiveArrayTypeInfo.BYTE_PRIMITIVE_ARRAY_TYPE_INFO));
    SingleOutputStreamOperator<Long> counts = elements1.connect(elements2).keyBy(element -> element.f0, element -> element.f0).transform("Asserting operator", BasicTypeInfo.LONG_TYPE_INFO, new AssertingTwoInputOperator());
    long sum = CollectionUtil.iteratorToList(DataStreamUtils.collect(counts)).stream().mapToLong(l -> l).sum();
    assertThat(sum, equalTo(numberOfRecords * 2));
}