Search in sources :

Example 91 with OutputTag

use of org.apache.flink.util.OutputTag in project flink-mirror by flink-ci.

the class IterateITCase method testmultipleHeadsTailsSimple.

@Test
public void testmultipleHeadsTailsSimple() {
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    DataStream<Integer> source1 = env.fromElements(1, 2, 3, 4, 5).shuffle().map(noOpIntMap).name("ParallelizeMapShuffle");
    DataStream<Integer> source2 = env.fromElements(1, 2, 3, 4, 5).map(noOpIntMap).name("ParallelizeMapRebalance");
    IterativeStream<Integer> iter1 = source1.union(source2).iterate();
    DataStream<Integer> head1 = iter1.map(noOpIntMap).name("IterRebalanceMap").setParallelism(parallelism / 2);
    DataStream<Integer> head2 = iter1.map(noOpIntMap).name("IterForwardMap");
    DataStreamSink<Integer> head3 = iter1.map(noOpIntMap).setParallelism(parallelism / 2).addSink(new ReceiveCheckNoOpSink<Integer>());
    DataStreamSink<Integer> head4 = iter1.map(noOpIntMap).addSink(new ReceiveCheckNoOpSink<Integer>());
    OutputTag<Integer> even = new OutputTag<Integer>("even") {
    };
    OutputTag<Integer> odd = new OutputTag<Integer>("odd") {
    };
    SingleOutputStreamOperator<Object> source3 = env.fromElements(1, 2, 3, 4, 5).map(noOpIntMap).name("EvenOddSourceMap").process(new ProcessFunction<Integer, Object>() {

        @Override
        public void processElement(Integer value, Context ctx, Collector<Object> out) throws Exception {
            if (value % 2 == 0) {
                ctx.output(even, value);
            } else {
                ctx.output(odd, value);
            }
        }
    });
    iter1.closeWith(source3.getSideOutput(even).union(head1.rebalance().map(noOpIntMap).broadcast(), head2.shuffle()));
    StreamGraph graph = env.getStreamGraph();
    JobGraph jg = graph.getJobGraph();
    assertEquals(1, graph.getIterationSourceSinkPairs().size());
    Tuple2<StreamNode, StreamNode> sourceSinkPair = graph.getIterationSourceSinkPairs().iterator().next();
    StreamNode itSource = sourceSinkPair.f0;
    StreamNode itSink = sourceSinkPair.f1;
    assertEquals(4, itSource.getOutEdges().size());
    assertEquals(3, itSink.getInEdges().size());
    assertEquals(itSource.getParallelism(), itSink.getParallelism());
    for (StreamEdge edge : itSource.getOutEdges()) {
        if (graph.getTargetVertex(edge).getOperatorName().equals("IterRebalanceMap")) {
            assertTrue(edge.getPartitioner() instanceof RebalancePartitioner);
        } else if (graph.getTargetVertex(edge).getOperatorName().equals("IterForwardMap")) {
            assertTrue(edge.getPartitioner() instanceof ForwardPartitioner);
        }
    }
    for (StreamEdge edge : itSink.getInEdges()) {
        if (graph.getStreamNode(edge.getSourceId()).getOperatorName().equals("ParallelizeMapShuffle")) {
            assertTrue(edge.getPartitioner() instanceof ShufflePartitioner);
        }
        if (graph.getStreamNode(edge.getSourceId()).getOperatorName().equals("ParallelizeMapForward")) {
            assertTrue(edge.getPartitioner() instanceof ForwardPartitioner);
        }
        if (graph.getStreamNode(edge.getSourceId()).getOperatorName().equals("EvenOddSourceMap")) {
            assertTrue(edge.getPartitioner() instanceof ForwardPartitioner);
        }
    }
    // Test co-location
    JobVertex itSource1 = null;
    JobVertex itSink1 = null;
    for (JobVertex vertex : jg.getVertices()) {
        if (vertex.getName().contains("IterationSource")) {
            itSource1 = vertex;
        } else if (vertex.getName().contains("IterationSink")) {
            itSink1 = vertex;
        }
    }
    assertTrue(itSource1.getCoLocationGroup() != null);
    assertEquals(itSource1.getCoLocationGroup(), itSink1.getCoLocationGroup());
}
Also used : RebalancePartitioner(org.apache.flink.streaming.runtime.partitioner.RebalancePartitioner) StreamEdge(org.apache.flink.streaming.api.graph.StreamEdge) InvalidProgramException(org.apache.flink.api.common.InvalidProgramException) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) ShufflePartitioner(org.apache.flink.streaming.runtime.partitioner.ShufflePartitioner) OutputTag(org.apache.flink.util.OutputTag) StreamGraph(org.apache.flink.streaming.api.graph.StreamGraph) StreamExecutionEnvironment(org.apache.flink.streaming.api.environment.StreamExecutionEnvironment) StreamNode(org.apache.flink.streaming.api.graph.StreamNode) ForwardPartitioner(org.apache.flink.streaming.runtime.partitioner.ForwardPartitioner) Test(org.junit.Test)

Example 92 with OutputTag

use of org.apache.flink.util.OutputTag in project flink-mirror by flink-ci.

the class IterateITCase method testmultipleHeadsTailsWithTailPartitioning.

@Test
public void testmultipleHeadsTailsWithTailPartitioning() {
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    DataStream<Integer> source1 = env.fromElements(1, 2, 3, 4, 5).shuffle().map(noOpIntMap);
    DataStream<Integer> source2 = env.fromElements(1, 2, 3, 4, 5).map(noOpIntMap);
    IterativeStream<Integer> iter1 = source1.union(source2).iterate();
    DataStream<Integer> head1 = iter1.map(noOpIntMap).name("map1");
    DataStream<Integer> head2 = iter1.map(noOpIntMap).setParallelism(parallelism / 2).name("shuffle").rebalance();
    DataStreamSink<Integer> head3 = iter1.map(noOpIntMap).setParallelism(parallelism / 2).addSink(new ReceiveCheckNoOpSink<Integer>());
    DataStreamSink<Integer> head4 = iter1.map(noOpIntMap).addSink(new ReceiveCheckNoOpSink<Integer>());
    OutputTag<Integer> even = new OutputTag<Integer>("even") {
    };
    OutputTag<Integer> odd = new OutputTag<Integer>("odd") {
    };
    SingleOutputStreamOperator<Object> source3 = env.fromElements(1, 2, 3, 4, 5).map(noOpIntMap).name("split").process(new ProcessFunction<Integer, Object>() {

        @Override
        public void processElement(Integer value, Context ctx, Collector<Object> out) throws Exception {
            if (value % 2 == 0) {
                ctx.output(even, value);
            } else {
                ctx.output(odd, value);
            }
        }
    });
    iter1.closeWith(source3.getSideOutput(even).union(head1.map(noOpIntMap).name("bc").broadcast(), head2.map(noOpIntMap).shuffle()));
    StreamGraph graph = env.getStreamGraph();
    JobGraph jg = graph.getJobGraph();
    assertEquals(1, graph.getIterationSourceSinkPairs().size());
    Tuple2<StreamNode, StreamNode> sourceSinkPair = graph.getIterationSourceSinkPairs().iterator().next();
    StreamNode itSource = sourceSinkPair.f0;
    StreamNode itSink = sourceSinkPair.f1;
    assertEquals(4, itSource.getOutEdges().size());
    assertEquals(3, itSink.getInEdges().size());
    assertEquals(itSource.getParallelism(), itSink.getParallelism());
    for (StreamEdge edge : itSource.getOutEdges()) {
        if (graph.getTargetVertex(edge).getOperatorName().equals("map1")) {
            assertTrue(edge.getPartitioner() instanceof ForwardPartitioner);
            assertEquals(4, graph.getTargetVertex(edge).getParallelism());
        } else if (graph.getTargetVertex(edge).getOperatorName().equals("shuffle")) {
            assertTrue(edge.getPartitioner() instanceof RebalancePartitioner);
            assertEquals(2, graph.getTargetVertex(edge).getParallelism());
        }
    }
    for (StreamEdge edge : itSink.getInEdges()) {
        String tailName = graph.getSourceVertex(edge).getOperatorName();
        if (tailName.equals("split")) {
            assertTrue(edge.getPartitioner() instanceof ForwardPartitioner);
        } else if (tailName.equals("bc")) {
            assertTrue(edge.getPartitioner() instanceof BroadcastPartitioner);
        } else if (tailName.equals("shuffle")) {
            assertTrue(edge.getPartitioner() instanceof ShufflePartitioner);
        }
    }
    // Test co-location
    JobVertex itSource1 = null;
    JobVertex itSink1 = null;
    for (JobVertex vertex : jg.getVertices()) {
        if (vertex.getName().contains("IterationSource")) {
            itSource1 = vertex;
        } else if (vertex.getName().contains("IterationSink")) {
            itSink1 = vertex;
        }
    }
    assertTrue(itSource1.getCoLocationGroup() != null);
    assertTrue(itSink1.getCoLocationGroup() != null);
    assertEquals(itSource1.getCoLocationGroup(), itSink1.getCoLocationGroup());
}
Also used : RebalancePartitioner(org.apache.flink.streaming.runtime.partitioner.RebalancePartitioner) OutputTag(org.apache.flink.util.OutputTag) StreamGraph(org.apache.flink.streaming.api.graph.StreamGraph) StreamNode(org.apache.flink.streaming.api.graph.StreamNode) ForwardPartitioner(org.apache.flink.streaming.runtime.partitioner.ForwardPartitioner) StreamEdge(org.apache.flink.streaming.api.graph.StreamEdge) InvalidProgramException(org.apache.flink.api.common.InvalidProgramException) BroadcastPartitioner(org.apache.flink.streaming.runtime.partitioner.BroadcastPartitioner) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) ShufflePartitioner(org.apache.flink.streaming.runtime.partitioner.ShufflePartitioner) StreamExecutionEnvironment(org.apache.flink.streaming.api.environment.StreamExecutionEnvironment) Test(org.junit.Test)

Example 93 with OutputTag

use of org.apache.flink.util.OutputTag in project flink-mirror by flink-ci.

the class OperatorChain method createStreamOutput.

private RecordWriterOutput<OUT> createStreamOutput(RecordWriter<SerializationDelegate<StreamRecord<OUT>>> recordWriter, StreamEdge edge, StreamConfig upStreamConfig, Environment taskEnvironment) {
    // OutputTag, return null if not sideOutput
    OutputTag sideOutputTag = edge.getOutputTag();
    TypeSerializer outSerializer;
    if (edge.getOutputTag() != null) {
        // side output
        outSerializer = upStreamConfig.getTypeSerializerSideOut(edge.getOutputTag(), taskEnvironment.getUserCodeClassLoader().asClassLoader());
    } else {
        // main output
        outSerializer = upStreamConfig.getTypeSerializerOut(taskEnvironment.getUserCodeClassLoader().asClassLoader());
    }
    return closer.register(new RecordWriterOutput<OUT>(recordWriter, outSerializer, sideOutputTag, edge.supportsUnalignedCheckpoints()));
}
Also used : TypeSerializer(org.apache.flink.api.common.typeutils.TypeSerializer) OutputTag(org.apache.flink.util.OutputTag)

Example 94 with OutputTag

use of org.apache.flink.util.OutputTag in project flink-mirror by flink-ci.

the class OperatorChain method createChainedSources.

@SuppressWarnings("rawtypes")
private Map<StreamConfig.SourceInputConfig, ChainedSource> createChainedSources(StreamTask<OUT, OP> containingTask, StreamConfig.InputConfig[] configuredInputs, Map<Integer, StreamConfig> chainedConfigs, ClassLoader userCodeClassloader, List<StreamOperatorWrapper<?, ?>> allOpWrappers) {
    if (Arrays.stream(configuredInputs).noneMatch(input -> input instanceof StreamConfig.SourceInputConfig)) {
        return Collections.emptyMap();
    }
    checkState(mainOperatorWrapper.getStreamOperator() instanceof MultipleInputStreamOperator, "Creating chained input is only supported with MultipleInputStreamOperator and MultipleInputStreamTask");
    Map<StreamConfig.SourceInputConfig, ChainedSource> chainedSourceInputs = new HashMap<>();
    MultipleInputStreamOperator<?> multipleInputOperator = (MultipleInputStreamOperator<?>) mainOperatorWrapper.getStreamOperator();
    List<Input> operatorInputs = multipleInputOperator.getInputs();
    int sourceInputGateIndex = Arrays.stream(containingTask.getEnvironment().getAllInputGates()).mapToInt(IndexedInputGate::getInputGateIndex).max().orElse(-1) + 1;
    for (int inputId = 0; inputId < configuredInputs.length; inputId++) {
        if (!(configuredInputs[inputId] instanceof StreamConfig.SourceInputConfig)) {
            continue;
        }
        StreamConfig.SourceInputConfig sourceInput = (StreamConfig.SourceInputConfig) configuredInputs[inputId];
        int sourceEdgeId = sourceInput.getInputEdge().getSourceId();
        StreamConfig sourceInputConfig = chainedConfigs.get(sourceEdgeId);
        OutputTag outputTag = sourceInput.getInputEdge().getOutputTag();
        WatermarkGaugeExposingOutput chainedSourceOutput = createChainedSourceOutput(containingTask, sourceInputConfig, userCodeClassloader, getFinishedOnRestoreInputOrDefault(operatorInputs.get(inputId)), multipleInputOperator.getMetricGroup(), outputTag);
        SourceOperator<?, ?> sourceOperator = (SourceOperator<?, ?>) createOperator(containingTask, sourceInputConfig, userCodeClassloader, (WatermarkGaugeExposingOutput<StreamRecord<OUT>>) chainedSourceOutput, allOpWrappers, true);
        chainedSourceInputs.put(sourceInput, new ChainedSource(chainedSourceOutput, this.isTaskDeployedAsFinished() ? new StreamTaskFinishedOnRestoreSourceInput<>(sourceOperator, sourceInputGateIndex++, inputId) : new StreamTaskSourceInput<>(sourceOperator, sourceInputGateIndex++, inputId)));
    }
    return chainedSourceInputs;
}
Also used : HashMap(java.util.HashMap) StreamConfig(org.apache.flink.streaming.api.graph.StreamConfig) BoundedMultiInput(org.apache.flink.streaming.api.operators.BoundedMultiInput) StreamTaskSourceInput(org.apache.flink.streaming.runtime.io.StreamTaskSourceInput) Input(org.apache.flink.streaming.api.operators.Input) MultipleInputStreamOperator(org.apache.flink.streaming.api.operators.MultipleInputStreamOperator) SourceOperator(org.apache.flink.streaming.api.operators.SourceOperator) OutputTag(org.apache.flink.util.OutputTag)

Example 95 with OutputTag

use of org.apache.flink.util.OutputTag in project flink-mirror by flink-ci.

the class SideOutputITCase method testProcessdWindowFunctionSideOutput.

@Test
public void testProcessdWindowFunctionSideOutput() throws Exception {
    TestListResultSink<Integer> resultSink = new TestListResultSink<>();
    TestListResultSink<String> sideOutputResultSink = new TestListResultSink<>();
    StreamExecutionEnvironment see = StreamExecutionEnvironment.getExecutionEnvironment();
    see.setParallelism(3);
    DataStream<Integer> dataStream = see.fromCollection(elements);
    OutputTag<String> sideOutputTag = new OutputTag<String>("side") {
    };
    SingleOutputStreamOperator<Integer> windowOperator = dataStream.assignTimestampsAndWatermarks(new TestWatermarkAssigner()).keyBy(new TestKeySelector()).window(SlidingEventTimeWindows.of(Time.milliseconds(1), Time.milliseconds(1))).process(new ProcessWindowFunction<Integer, Integer, Integer, TimeWindow>() {

        private static final long serialVersionUID = 1L;

        @Override
        public void process(Integer integer, Context context, Iterable<Integer> elements, Collector<Integer> out) throws Exception {
            out.collect(integer);
            context.output(sideOutputTag, "sideout-" + String.valueOf(integer));
        }
    });
    windowOperator.getSideOutput(sideOutputTag).addSink(sideOutputResultSink);
    windowOperator.addSink(resultSink);
    see.execute();
    assertEquals(Arrays.asList("sideout-1", "sideout-2", "sideout-5"), sideOutputResultSink.getSortedResult());
    assertEquals(Arrays.asList(1, 2, 5), resultSink.getSortedResult());
}
Also used : TimeWindow(org.apache.flink.streaming.api.windowing.windows.TimeWindow) ExpectedException(org.junit.rules.ExpectedException) TestListResultSink(org.apache.flink.test.streaming.runtime.util.TestListResultSink) OutputTag(org.apache.flink.util.OutputTag) StreamExecutionEnvironment(org.apache.flink.streaming.api.environment.StreamExecutionEnvironment) Test(org.junit.Test)

Aggregations

OutputTag (org.apache.flink.util.OutputTag)111 Test (org.junit.Test)97 StreamExecutionEnvironment (org.apache.flink.streaming.api.environment.StreamExecutionEnvironment)86 TestListResultSink (org.apache.flink.test.streaming.runtime.util.TestListResultSink)57 ExpectedException (org.junit.rules.ExpectedException)57 List (java.util.List)24 StreamRecord (org.apache.flink.streaming.runtime.streamrecord.StreamRecord)24 ArrayList (java.util.ArrayList)20 SingleOutputStreamOperator (org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator)18 HashMap (java.util.HashMap)17 DataStream (org.apache.flink.streaming.api.datastream.DataStream)17 JobGraph (org.apache.flink.runtime.jobgraph.JobGraph)16 Collector (org.apache.flink.util.Collector)16 Arrays (java.util.Arrays)15 Tuple2 (org.apache.flink.api.java.tuple.Tuple2)15 Assert.assertEquals (org.junit.Assert.assertEquals)13 Objects (java.util.Objects)12 Optional (java.util.Optional)12 JobVertex (org.apache.flink.runtime.jobgraph.JobVertex)12 Map (java.util.Map)10