Search in sources :

Example 1 with IsBounded

use of org.apache.beam.model.pipeline.v1.RunnerApi.IsBounded in project beam by apache.

the class PCollectionTranslationTest method testEncodeDecodeFields.

@Test
public void testEncodeDecodeFields() throws Exception {
    SdkComponents sdkComponents = SdkComponents.create();
    sdkComponents.registerEnvironment(Environments.createDockerEnvironment("java"));
    RunnerApi.PCollection protoCollection = PCollectionTranslation.toProto(testCollection, sdkComponents);
    RehydratedComponents protoComponents = RehydratedComponents.forComponents(sdkComponents.toComponents());
    Coder<?> decodedCoder = protoComponents.getCoder(protoCollection.getCoderId());
    WindowingStrategy<?, ?> decodedStrategy = protoComponents.getWindowingStrategy(protoCollection.getWindowingStrategyId());
    IsBounded decodedIsBounded = PCollectionTranslation.isBounded(protoCollection);
    assertThat(decodedCoder, equalTo(testCollection.getCoder()));
    assertThat(decodedStrategy, equalTo(testCollection.getWindowingStrategy().withEnvironmentId(sdkComponents.getOnlyEnvironmentId()).fixDefaults()));
    assertThat(decodedIsBounded, equalTo(testCollection.isBounded()));
}
Also used : RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) IsBounded(org.apache.beam.sdk.values.PCollection.IsBounded) Test(org.junit.Test)

Example 2 with IsBounded

use of org.apache.beam.model.pipeline.v1.RunnerApi.IsBounded in project beam by apache.

the class CreateExecutableStageNodeFunction method apply.

@Override
public Node apply(MutableNetwork<Node, Edge> input) {
    for (Node node : input.nodes()) {
        if (node instanceof RemoteGrpcPortNode || node instanceof ParallelInstructionNode || node instanceof InstructionOutputNode) {
            continue;
        }
        throw new IllegalArgumentException(String.format("Network contains unknown type of node: %s", input));
    }
    // Fix all non output nodes to have named edges.
    for (Node node : input.nodes()) {
        if (node instanceof InstructionOutputNode) {
            continue;
        }
        for (Node successor : input.successors(node)) {
            for (Edge edge : input.edgesConnecting(node, successor)) {
                if (edge instanceof DefaultEdge) {
                    input.removeEdge(edge);
                    input.addEdge(node, successor, MultiOutputInfoEdge.create(new MultiOutputInfo().setTag(idGenerator.getId())));
                }
            }
        }
    }
    RunnerApi.Components.Builder componentsBuilder = RunnerApi.Components.newBuilder();
    componentsBuilder.mergeFrom(this.pipeline.getComponents());
    // Default to use the Java environment if pipeline doesn't have environment specified.
    if (pipeline.getComponents().getEnvironmentsMap().isEmpty()) {
        String envId = Environments.JAVA_SDK_HARNESS_ENVIRONMENT.getUrn() + idGenerator.getId();
        componentsBuilder.putEnvironments(envId, Environments.JAVA_SDK_HARNESS_ENVIRONMENT);
    }
    // By default, use GlobalWindow for all languages.
    // For java, if there is a IntervalWindowCoder, then use FixedWindow instead.
    // TODO: should get real WindowingStategy from pipeline proto.
    String globalWindowingStrategyId = "generatedGlobalWindowingStrategy" + idGenerator.getId();
    String intervalWindowEncodingWindowingStrategyId = "generatedIntervalWindowEncodingWindowingStrategy" + idGenerator.getId();
    SdkComponents sdkComponents = SdkComponents.create(pipeline.getComponents(), null);
    try {
        registerWindowingStrategy(globalWindowingStrategyId, WindowingStrategy.globalDefault(), componentsBuilder, sdkComponents);
        registerWindowingStrategy(intervalWindowEncodingWindowingStrategyId, WindowingStrategy.of(FixedWindows.of(Duration.standardSeconds(1))), componentsBuilder, sdkComponents);
    } catch (IOException exc) {
        throw new RuntimeException("Could not convert default windowing stratey to proto", exc);
    }
    Map<Node, String> nodesToPCollections = new HashMap<>();
    ImmutableMap.Builder<String, NameContext> ptransformIdToNameContexts = ImmutableMap.builder();
    ImmutableMap.Builder<String, Iterable<SideInputInfo>> ptransformIdToSideInputInfos = ImmutableMap.builder();
    ImmutableMap.Builder<String, Iterable<PCollectionView<?>>> ptransformIdToPCollectionViews = ImmutableMap.builder();
    // A field of ExecutableStage which includes the PCollection goes to worker side.
    Set<PCollectionNode> executableStageOutputs = new HashSet<>();
    // A field of ExecutableStage which includes the PCollection goes to runner side.
    Set<PCollectionNode> executableStageInputs = new HashSet<>();
    for (InstructionOutputNode node : Iterables.filter(input.nodes(), InstructionOutputNode.class)) {
        InstructionOutput instructionOutput = node.getInstructionOutput();
        String coderId = "generatedCoder" + idGenerator.getId();
        String windowingStrategyId;
        try (ByteString.Output output = ByteString.newOutput()) {
            try {
                Coder<?> javaCoder = CloudObjects.coderFromCloudObject(CloudObject.fromSpec(instructionOutput.getCodec()));
                Coder<?> elementCoder = ((WindowedValueCoder<?>) javaCoder).getValueCoder();
                sdkComponents.registerCoder(elementCoder);
                RunnerApi.Coder coderProto = CoderTranslation.toProto(elementCoder, sdkComponents);
                componentsBuilder.putCoders(coderId, coderProto);
                // For now, Dataflow runner harness only deal with FixedWindow.
                if (javaCoder instanceof FullWindowedValueCoder) {
                    FullWindowedValueCoder<?> windowedValueCoder = (FullWindowedValueCoder<?>) javaCoder;
                    Coder<?> windowCoder = windowedValueCoder.getWindowCoder();
                    if (windowCoder instanceof IntervalWindowCoder) {
                        windowingStrategyId = intervalWindowEncodingWindowingStrategyId;
                    } else if (windowCoder instanceof GlobalWindow.Coder) {
                        windowingStrategyId = globalWindowingStrategyId;
                    } else {
                        throw new UnsupportedOperationException(String.format("Dataflow portable runner harness doesn't support windowing with %s", windowCoder));
                    }
                } else {
                    throw new UnsupportedOperationException("Dataflow portable runner harness only supports FullWindowedValueCoder");
                }
            } catch (IOException e) {
                throw new IllegalArgumentException(String.format("Unable to encode coder %s for output %s", instructionOutput.getCodec(), instructionOutput), e);
            } catch (Exception e) {
                // Coder probably wasn't a java coder
                OBJECT_MAPPER.writeValue(output, instructionOutput.getCodec());
                componentsBuilder.putCoders(coderId, RunnerApi.Coder.newBuilder().setSpec(RunnerApi.FunctionSpec.newBuilder().setPayload(output.toByteString())).build());
                // For non-java coder, hope it's GlobalWindows by default.
                // TODO(BEAM-6231): Actually discover the right windowing strategy.
                windowingStrategyId = globalWindowingStrategyId;
            }
        } catch (IOException e) {
            throw new IllegalArgumentException(String.format("Unable to encode coder %s for output %s", instructionOutput.getCodec(), instructionOutput), e);
        }
        // TODO(BEAM-6275): Set correct IsBounded on generated PCollections
        String pcollectionId = node.getPcollectionId();
        RunnerApi.PCollection pCollection = RunnerApi.PCollection.newBuilder().setCoderId(coderId).setWindowingStrategyId(windowingStrategyId).setIsBounded(RunnerApi.IsBounded.Enum.BOUNDED).build();
        nodesToPCollections.put(node, pcollectionId);
        componentsBuilder.putPcollections(pcollectionId, pCollection);
        // is set
        if (isExecutableStageOutputPCollection(input, node)) {
            executableStageOutputs.add(PipelineNode.pCollection(pcollectionId, pCollection));
        }
        if (isExecutableStageInputPCollection(input, node)) {
            executableStageInputs.add(PipelineNode.pCollection(pcollectionId, pCollection));
        }
    }
    componentsBuilder.putAllCoders(sdkComponents.toComponents().getCodersMap());
    Set<PTransformNode> executableStageTransforms = new HashSet<>();
    Set<TimerReference> executableStageTimers = new HashSet<>();
    List<UserStateId> userStateIds = new ArrayList<>();
    Set<SideInputReference> executableStageSideInputs = new HashSet<>();
    for (ParallelInstructionNode node : Iterables.filter(input.nodes(), ParallelInstructionNode.class)) {
        ImmutableMap.Builder<String, PCollectionNode> sideInputIds = ImmutableMap.builder();
        ParallelInstruction parallelInstruction = node.getParallelInstruction();
        String ptransformId = "generatedPtransform" + idGenerator.getId();
        ptransformIdToNameContexts.put(ptransformId, NameContext.create(null, parallelInstruction.getOriginalName(), parallelInstruction.getSystemName(), parallelInstruction.getName()));
        RunnerApi.PTransform.Builder pTransform = RunnerApi.PTransform.newBuilder();
        RunnerApi.FunctionSpec.Builder transformSpec = RunnerApi.FunctionSpec.newBuilder();
        List<String> timerIds = new ArrayList<>();
        if (parallelInstruction.getParDo() != null) {
            ParDoInstruction parDoInstruction = parallelInstruction.getParDo();
            CloudObject userFnSpec = CloudObject.fromSpec(parDoInstruction.getUserFn());
            String userFnClassName = userFnSpec.getClassName();
            if (userFnClassName.equals("CombineValuesFn") || userFnClassName.equals("KeyedCombineFn")) {
                transformSpec = transformCombineValuesFnToFunctionSpec(userFnSpec);
                ptransformIdToPCollectionViews.put(ptransformId, Collections.emptyList());
            } else {
                String parDoPTransformId = getString(userFnSpec, PropertyNames.SERIALIZED_FN);
                RunnerApi.PTransform parDoPTransform = pipeline.getComponents().getTransformsOrDefault(parDoPTransformId, null);
                // TODO: only the non-null branch should exist; for migration ease only
                if (parDoPTransform != null) {
                    checkArgument(parDoPTransform.getSpec().getUrn().equals(PTransformTranslation.PAR_DO_TRANSFORM_URN), "Found transform \"%s\" for ParallelDo instruction, " + " but that transform had unexpected URN \"%s\" (expected \"%s\")", parDoPTransformId, parDoPTransform.getSpec().getUrn(), PTransformTranslation.PAR_DO_TRANSFORM_URN);
                    RunnerApi.ParDoPayload parDoPayload;
                    try {
                        parDoPayload = RunnerApi.ParDoPayload.parseFrom(parDoPTransform.getSpec().getPayload());
                    } catch (InvalidProtocolBufferException exc) {
                        throw new RuntimeException("ParDo did not have a ParDoPayload", exc);
                    }
                    // user timers and user state.
                    for (Map.Entry<String, RunnerApi.TimerFamilySpec> entry : parDoPayload.getTimerFamilySpecsMap().entrySet()) {
                        timerIds.add(entry.getKey());
                    }
                    for (Map.Entry<String, RunnerApi.StateSpec> entry : parDoPayload.getStateSpecsMap().entrySet()) {
                        UserStateId.Builder builder = UserStateId.newBuilder();
                        builder.setTransformId(parDoPTransformId);
                        builder.setLocalName(entry.getKey());
                        userStateIds.add(builder.build());
                    }
                    // To facilitate the creation of Set executableStageSideInputs.
                    for (String sideInputTag : parDoPayload.getSideInputsMap().keySet()) {
                        String sideInputPCollectionId = parDoPTransform.getInputsOrThrow(sideInputTag);
                        RunnerApi.PCollection sideInputPCollection = pipeline.getComponents().getPcollectionsOrThrow(sideInputPCollectionId);
                        pTransform.putInputs(sideInputTag, sideInputPCollectionId);
                        PCollectionNode pCollectionNode = PipelineNode.pCollection(sideInputPCollectionId, sideInputPCollection);
                        sideInputIds.put(sideInputTag, pCollectionNode);
                    }
                    // To facilitate the creation of Map(ptransformId -> pCollectionView), which is
                    // required by constructing an ExecutableStageNode.
                    ImmutableList.Builder<PCollectionView<?>> pcollectionViews = ImmutableList.builder();
                    for (Map.Entry<String, RunnerApi.SideInput> sideInputEntry : parDoPayload.getSideInputsMap().entrySet()) {
                        pcollectionViews.add(RegisterNodeFunction.transformSideInputForRunner(pipeline, parDoPTransform, sideInputEntry.getKey(), sideInputEntry.getValue()));
                    }
                    ptransformIdToPCollectionViews.put(ptransformId, pcollectionViews.build());
                    transformSpec.setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).setPayload(parDoPayload.toByteString());
                } else {
                    // legacy path - bytes are the FunctionSpec's payload field, basically, and
                    // SDKs expect it in the PTransform's payload field
                    byte[] userFnBytes = getBytes(userFnSpec, PropertyNames.SERIALIZED_FN);
                    transformSpec.setUrn(ParDoTranslation.CUSTOM_JAVA_DO_FN_URN).setPayload(ByteString.copyFrom(userFnBytes));
                }
                if (parDoInstruction.getSideInputs() != null) {
                    ptransformIdToSideInputInfos.put(ptransformId, forSideInputInfos(parDoInstruction.getSideInputs(), true));
                }
            }
        } else if (parallelInstruction.getRead() != null) {
            ReadInstruction readInstruction = parallelInstruction.getRead();
            CloudObject sourceSpec = CloudObject.fromSpec(CloudSourceUtils.flattenBaseSpecs(readInstruction.getSource()).getSpec());
            // TODO: Need to plumb through the SDK specific function spec.
            transformSpec.setUrn(JAVA_SOURCE_URN);
            try {
                byte[] serializedSource = Base64.getDecoder().decode(getString(sourceSpec, SERIALIZED_SOURCE));
                ByteString sourceByteString = ByteString.copyFrom(serializedSource);
                transformSpec.setPayload(sourceByteString);
            } catch (Exception e) {
                throw new IllegalArgumentException(String.format("Unable to process Read %s", parallelInstruction), e);
            }
        } else if (parallelInstruction.getFlatten() != null) {
            transformSpec.setUrn(PTransformTranslation.FLATTEN_TRANSFORM_URN);
        } else {
            throw new IllegalArgumentException(String.format("Unknown type of ParallelInstruction %s", parallelInstruction));
        }
        // predecessor in a ParDo. This PCollection is called the "main input".
        for (Node predecessorOutput : input.predecessors(node)) {
            pTransform.putInputs("generatedInput" + idGenerator.getId(), nodesToPCollections.get(predecessorOutput));
        }
        for (Edge edge : input.outEdges(node)) {
            Node nodeOutput = input.incidentNodes(edge).target();
            MultiOutputInfoEdge edge2 = (MultiOutputInfoEdge) edge;
            pTransform.putOutputs(edge2.getMultiOutputInfo().getTag(), nodesToPCollections.get(nodeOutput));
        }
        pTransform.setSpec(transformSpec);
        PTransformNode pTransformNode = PipelineNode.pTransform(ptransformId, pTransform.build());
        executableStageTransforms.add(pTransformNode);
        for (String timerId : timerIds) {
            executableStageTimers.add(TimerReference.of(pTransformNode, timerId));
        }
        ImmutableMap<String, PCollectionNode> sideInputIdToPCollectionNodes = sideInputIds.build();
        for (String sideInputTag : sideInputIdToPCollectionNodes.keySet()) {
            SideInputReference sideInputReference = SideInputReference.of(pTransformNode, sideInputTag, sideInputIdToPCollectionNodes.get(sideInputTag));
            executableStageSideInputs.add(sideInputReference);
        }
        executableStageTransforms.add(pTransformNode);
    }
    if (executableStageInputs.size() != 1) {
        throw new UnsupportedOperationException("ExecutableStage only support one input PCollection");
    }
    PCollectionNode executableInput = executableStageInputs.iterator().next();
    RunnerApi.Components executableStageComponents = componentsBuilder.build();
    // Get Environment from ptransform, otherwise, use JAVA_SDK_HARNESS_ENVIRONMENT as default.
    Environment executableStageEnv = getEnvironmentFromPTransform(executableStageComponents, executableStageTransforms);
    if (executableStageEnv == null) {
        executableStageEnv = Environments.JAVA_SDK_HARNESS_ENVIRONMENT;
    }
    Set<UserStateReference> executableStageUserStateReference = new HashSet<>();
    for (UserStateId userStateId : userStateIds) {
        executableStageUserStateReference.add(UserStateReference.fromUserStateId(userStateId, executableStageComponents));
    }
    ExecutableStage executableStage = ImmutableExecutableStage.ofFullComponents(executableStageComponents, executableStageEnv, executableInput, executableStageSideInputs, executableStageUserStateReference, executableStageTimers, executableStageTransforms, executableStageOutputs, DEFAULT_WIRE_CODER_SETTINGS);
    return ExecutableStageNode.create(executableStage, ptransformIdToNameContexts.build(), ptransformIdToSideInputInfos.build(), ptransformIdToPCollectionViews.build());
}
Also used : HashMap(java.util.HashMap) MultiOutputInfo(com.google.api.services.dataflow.model.MultiOutputInfo) ByteString(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString) PCollectionNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PCollectionNode) InstructionOutputNode(org.apache.beam.runners.dataflow.worker.graph.Nodes.InstructionOutputNode) PTransformNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode) ParallelInstructionNode(org.apache.beam.runners.dataflow.worker.graph.Nodes.ParallelInstructionNode) Node(org.apache.beam.runners.dataflow.worker.graph.Nodes.Node) ExecutableStageNode(org.apache.beam.runners.dataflow.worker.graph.Nodes.ExecutableStageNode) PipelineNode(org.apache.beam.runners.core.construction.graph.PipelineNode) RemoteGrpcPortNode(org.apache.beam.runners.dataflow.worker.graph.Nodes.RemoteGrpcPortNode) InstructionOutput(com.google.api.services.dataflow.model.InstructionOutput) PTransformNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode) ArrayList(java.util.ArrayList) Structs.getString(org.apache.beam.runners.dataflow.util.Structs.getString) ByteString(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString) RehydratedComponents(org.apache.beam.runners.core.construction.RehydratedComponents) SdkComponents(org.apache.beam.runners.core.construction.SdkComponents) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) SideInputReference(org.apache.beam.runners.core.construction.graph.SideInputReference) ImmutableExecutableStage(org.apache.beam.runners.core.construction.graph.ImmutableExecutableStage) ExecutableStage(org.apache.beam.runners.core.construction.graph.ExecutableStage) HashSet(java.util.HashSet) DefaultEdge(org.apache.beam.runners.dataflow.worker.graph.Edges.DefaultEdge) MultiOutputInfoEdge(org.apache.beam.runners.dataflow.worker.graph.Edges.MultiOutputInfoEdge) ImmutableMap(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableMap) ParDoInstruction(com.google.api.services.dataflow.model.ParDoInstruction) PCollectionView(org.apache.beam.sdk.values.PCollectionView) Environment(org.apache.beam.model.pipeline.v1.RunnerApi.Environment) GlobalWindow(org.apache.beam.sdk.transforms.windowing.GlobalWindow) ImmutableMap(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableMap) Map(java.util.Map) HashMap(java.util.HashMap) IntervalWindowCoder(org.apache.beam.sdk.transforms.windowing.IntervalWindow.IntervalWindowCoder) RemoteGrpcPortNode(org.apache.beam.runners.dataflow.worker.graph.Nodes.RemoteGrpcPortNode) TimerReference(org.apache.beam.runners.core.construction.graph.TimerReference) ImmutableList(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableList) ParallelInstructionNode(org.apache.beam.runners.dataflow.worker.graph.Nodes.ParallelInstructionNode) SdkComponents(org.apache.beam.runners.core.construction.SdkComponents) ReadInstruction(com.google.api.services.dataflow.model.ReadInstruction) InstructionOutputNode(org.apache.beam.runners.dataflow.worker.graph.Nodes.InstructionOutputNode) FullWindowedValueCoder(org.apache.beam.sdk.util.WindowedValue.FullWindowedValueCoder) NameContext(org.apache.beam.runners.dataflow.worker.counters.NameContext) InvalidProtocolBufferException(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.InvalidProtocolBufferException) IOException(java.io.IOException) PCollectionNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PCollectionNode) InvalidProtocolBufferException(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.InvalidProtocolBufferException) IOException(java.io.IOException) ParallelInstruction(com.google.api.services.dataflow.model.ParallelInstruction) WindowedValueCoder(org.apache.beam.sdk.util.WindowedValue.WindowedValueCoder) FullWindowedValueCoder(org.apache.beam.sdk.util.WindowedValue.FullWindowedValueCoder) CloudObject(org.apache.beam.runners.dataflow.util.CloudObject) UserStateId(org.apache.beam.model.pipeline.v1.RunnerApi.ExecutableStagePayload.UserStateId) UserStateReference(org.apache.beam.runners.core.construction.graph.UserStateReference) Edge(org.apache.beam.runners.dataflow.worker.graph.Edges.Edge) MultiOutputInfoEdge(org.apache.beam.runners.dataflow.worker.graph.Edges.MultiOutputInfoEdge) DefaultEdge(org.apache.beam.runners.dataflow.worker.graph.Edges.DefaultEdge)

Example 3 with IsBounded

use of org.apache.beam.model.pipeline.v1.RunnerApi.IsBounded in project beam by apache.

the class ParDoBoundMultiTranslator method doTranslatePortable.

// static for serializing anonymous functions
private static <InT, OutT> void doTranslatePortable(PipelineNode.PTransformNode transform, QueryablePipeline pipeline, PortableTranslationContext ctx) {
    Map<String, String> outputs = transform.getTransform().getOutputsMap();
    final RunnerApi.ExecutableStagePayload stagePayload;
    try {
        stagePayload = RunnerApi.ExecutableStagePayload.parseFrom(transform.getTransform().getSpec().getPayload());
    } catch (IOException e) {
        throw new RuntimeException(e);
    }
    String inputId = stagePayload.getInput();
    final MessageStream<OpMessage<InT>> inputStream = ctx.getMessageStreamById(inputId);
    // Analyze side inputs
    final List<MessageStream<OpMessage<Iterable<?>>>> sideInputStreams = new ArrayList<>();
    final Map<SideInputId, PCollectionView<?>> sideInputMapping = new HashMap<>();
    final Map<String, PCollectionView<?>> idToViewMapping = new HashMap<>();
    final RunnerApi.Components components = stagePayload.getComponents();
    for (SideInputId sideInputId : stagePayload.getSideInputsList()) {
        final String sideInputCollectionId = components.getTransformsOrThrow(sideInputId.getTransformId()).getInputsOrThrow(sideInputId.getLocalName());
        final WindowingStrategy<?, BoundedWindow> windowingStrategy = WindowUtils.getWindowStrategy(sideInputCollectionId, components);
        final WindowedValue.WindowedValueCoder<?> coder = (WindowedValue.WindowedValueCoder) instantiateCoder(sideInputCollectionId, components);
        // Create a runner-side view
        final PCollectionView<?> view = createPCollectionView(sideInputId, coder, windowingStrategy);
        // Use GBK to aggregate the side inputs and then broadcast it out
        final MessageStream<OpMessage<Iterable<?>>> broadcastSideInput = groupAndBroadcastSideInput(sideInputId, sideInputCollectionId, components.getPcollectionsOrThrow(sideInputCollectionId), (WindowingStrategy) windowingStrategy, coder, ctx);
        sideInputStreams.add(broadcastSideInput);
        sideInputMapping.put(sideInputId, view);
        idToViewMapping.put(getSideInputUniqueId(sideInputId), view);
    }
    final Map<TupleTag<?>, Integer> tagToIndexMap = new HashMap<>();
    final Map<Integer, String> indexToIdMap = new HashMap<>();
    final Map<String, TupleTag<?>> idToTupleTagMap = new HashMap<>();
    // first output as the main output
    final TupleTag<OutT> mainOutputTag = outputs.isEmpty() ? null : new TupleTag(outputs.keySet().iterator().next());
    AtomicInteger index = new AtomicInteger(0);
    outputs.keySet().iterator().forEachRemaining(outputName -> {
        TupleTag<?> tupleTag = new TupleTag<>(outputName);
        tagToIndexMap.put(tupleTag, index.get());
        String collectionId = outputs.get(outputName);
        indexToIdMap.put(index.get(), collectionId);
        idToTupleTagMap.put(collectionId, tupleTag);
        index.incrementAndGet();
    });
    WindowedValue.WindowedValueCoder<InT> windowedInputCoder = WindowUtils.instantiateWindowedCoder(inputId, pipeline.getComponents());
    // TODO: support schema and side inputs for portable runner
    // Note: transform.getTransform() is an ExecutableStage, not ParDo, so we need to extract
    // these info from its components.
    final DoFnSchemaInformation doFnSchemaInformation = null;
    final RunnerApi.PCollection input = pipeline.getComponents().getPcollectionsOrThrow(inputId);
    final PCollection.IsBounded isBounded = SamzaPipelineTranslatorUtils.isBounded(input);
    final Coder<?> keyCoder = StateUtils.isStateful(stagePayload) ? ((KvCoder) ((WindowedValue.FullWindowedValueCoder) windowedInputCoder).getValueCoder()).getKeyCoder() : null;
    final DoFnOp<InT, OutT, RawUnionValue> op = new DoFnOp<>(mainOutputTag, new NoOpDoFn<>(), keyCoder, // input coder not in use
    windowedInputCoder.getValueCoder(), windowedInputCoder, // output coders not in use
    Collections.emptyMap(), new ArrayList<>(sideInputMapping.values()), // used by java runner only
    new ArrayList<>(idToTupleTagMap.values()), WindowUtils.getWindowStrategy(inputId, stagePayload.getComponents()), idToViewMapping, new DoFnOp.MultiOutputManagerFactory(tagToIndexMap), ctx.getTransformFullName(), ctx.getTransformId(), isBounded, true, stagePayload, ctx.getJobInfo(), idToTupleTagMap, doFnSchemaInformation, sideInputMapping);
    final MessageStream<OpMessage<InT>> mergedStreams;
    if (sideInputStreams.isEmpty()) {
        mergedStreams = inputStream;
    } else {
        MessageStream<OpMessage<InT>> mergedSideInputStreams = MessageStream.mergeAll(sideInputStreams).flatMap(new SideInputWatermarkFn());
        mergedStreams = inputStream.merge(Collections.singletonList(mergedSideInputStreams));
    }
    final MessageStream<OpMessage<RawUnionValue>> taggedOutputStream = mergedStreams.flatMapAsync(OpAdapter.adapt(op));
    for (int outputIndex : tagToIndexMap.values()) {
        @SuppressWarnings("unchecked") final MessageStream<OpMessage<OutT>> outputStream = taggedOutputStream.filter(message -> message.getType() != OpMessage.Type.ELEMENT || message.getElement().getValue().getUnionTag() == outputIndex).flatMapAsync(OpAdapter.adapt(new RawUnionValueToValue()));
        ctx.registerMessageStream(indexToIdMap.get(outputIndex), outputStream);
    }
}
Also used : WindowedValue(org.apache.beam.sdk.util.WindowedValue) PCollectionViews(org.apache.beam.sdk.values.PCollectionViews) OpMessage(org.apache.beam.runners.samza.runtime.OpMessage) DoFnSchemaInformation(org.apache.beam.sdk.transforms.DoFnSchemaInformation) RunnerPCollectionView(org.apache.beam.runners.core.construction.RunnerPCollectionView) WatermarkFunction(org.apache.samza.operators.functions.WatermarkFunction) DoFnSignatures(org.apache.beam.sdk.transforms.reflect.DoFnSignatures) AtomicInteger(java.util.concurrent.atomic.AtomicInteger) Map(java.util.Map) Iterators(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.Iterators) StateUtils(org.apache.beam.runners.samza.util.StateUtils) KvCoder(org.apache.beam.sdk.coders.KvCoder) Collection(java.util.Collection) ServiceLoader(java.util.ServiceLoader) Collectors(java.util.stream.Collectors) TransformHierarchy(org.apache.beam.sdk.runners.TransformHierarchy) FlatMapFunction(org.apache.samza.operators.functions.FlatMapFunction) List(java.util.List) ParDo(org.apache.beam.sdk.transforms.ParDo) TypeDescriptors(org.apache.beam.sdk.values.TypeDescriptors) OpEmitter(org.apache.beam.runners.samza.runtime.OpEmitter) PipelineTranslatorUtils.instantiateCoder(org.apache.beam.runners.fnexecution.translation.PipelineTranslatorUtils.instantiateCoder) WindowingStrategy(org.apache.beam.sdk.values.WindowingStrategy) SamzaPipelineTranslatorUtils(org.apache.beam.runners.samza.util.SamzaPipelineTranslatorUtils) KV(org.apache.beam.sdk.values.KV) TypeDescriptor(org.apache.beam.sdk.values.TypeDescriptor) OpAdapter(org.apache.beam.runners.samza.runtime.OpAdapter) DoFnOp(org.apache.beam.runners.samza.runtime.DoFnOp) Coder(org.apache.beam.sdk.coders.Coder) HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) RawUnionValue(org.apache.beam.sdk.transforms.join.RawUnionValue) ViewFn(org.apache.beam.sdk.transforms.ViewFn) TupleTag(org.apache.beam.sdk.values.TupleTag) SamzaDoFnInvokerRegistrar(org.apache.beam.runners.samza.runtime.SamzaDoFnInvokerRegistrar) WindowUtils(org.apache.beam.runners.samza.util.WindowUtils) SideInputId(org.apache.beam.model.pipeline.v1.RunnerApi.ExecutableStagePayload.SideInputId) ParDoTranslation(org.apache.beam.runners.core.construction.ParDoTranslation) MessageStream(org.apache.samza.operators.MessageStream) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) DoFn(org.apache.beam.sdk.transforms.DoFn) QueryablePipeline(org.apache.beam.runners.core.construction.graph.QueryablePipeline) Op(org.apache.beam.runners.samza.runtime.Op) DoFnSignature(org.apache.beam.sdk.transforms.reflect.DoFnSignature) Iterator(java.util.Iterator) IterableCoder(org.apache.beam.sdk.coders.IterableCoder) IOException(java.io.IOException) PCollection(org.apache.beam.sdk.values.PCollection) SamzaPipelineOptions(org.apache.beam.runners.samza.SamzaPipelineOptions) PCollectionView(org.apache.beam.sdk.values.PCollectionView) BoundedWindow(org.apache.beam.sdk.transforms.windowing.BoundedWindow) Instant(org.joda.time.Instant) PipelineNode(org.apache.beam.runners.core.construction.graph.PipelineNode) VoidCoder(org.apache.beam.sdk.coders.VoidCoder) Collections(java.util.Collections) OpMessage(org.apache.beam.runners.samza.runtime.OpMessage) HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) TupleTag(org.apache.beam.sdk.values.TupleTag) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) WindowedValue(org.apache.beam.sdk.util.WindowedValue) MessageStream(org.apache.samza.operators.MessageStream) BoundedWindow(org.apache.beam.sdk.transforms.windowing.BoundedWindow) SideInputId(org.apache.beam.model.pipeline.v1.RunnerApi.ExecutableStagePayload.SideInputId) RawUnionValue(org.apache.beam.sdk.transforms.join.RawUnionValue) RunnerPCollectionView(org.apache.beam.runners.core.construction.RunnerPCollectionView) PCollectionView(org.apache.beam.sdk.values.PCollectionView) DoFnSchemaInformation(org.apache.beam.sdk.transforms.DoFnSchemaInformation) AtomicInteger(java.util.concurrent.atomic.AtomicInteger) DoFnOp(org.apache.beam.runners.samza.runtime.DoFnOp) IOException(java.io.IOException) AtomicInteger(java.util.concurrent.atomic.AtomicInteger) PCollection(org.apache.beam.sdk.values.PCollection)

Aggregations

RunnerApi (org.apache.beam.model.pipeline.v1.RunnerApi)3 IOException (java.io.IOException)2 ArrayList (java.util.ArrayList)2 HashMap (java.util.HashMap)2 Map (java.util.Map)2 PipelineNode (org.apache.beam.runners.core.construction.graph.PipelineNode)2 PCollectionView (org.apache.beam.sdk.values.PCollectionView)2 InstructionOutput (com.google.api.services.dataflow.model.InstructionOutput)1 MultiOutputInfo (com.google.api.services.dataflow.model.MultiOutputInfo)1 ParDoInstruction (com.google.api.services.dataflow.model.ParDoInstruction)1 ParallelInstruction (com.google.api.services.dataflow.model.ParallelInstruction)1 ReadInstruction (com.google.api.services.dataflow.model.ReadInstruction)1 Collection (java.util.Collection)1 Collections (java.util.Collections)1 HashSet (java.util.HashSet)1 Iterator (java.util.Iterator)1 List (java.util.List)1 ServiceLoader (java.util.ServiceLoader)1 AtomicInteger (java.util.concurrent.atomic.AtomicInteger)1 Collectors (java.util.stream.Collectors)1