Examples with Pipeline org.apache.beam.model.pipeline.v1.RunnerApi.Pipeline used on opensource projects

Example 26 with Pipeline

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

the class EnvironmentsTest method getEnvironmentPTransform.

@Test
public void getEnvironmentPTransform() throws IOException {
    Pipeline p = Pipeline.create();
    SdkComponents components = SdkComponents.create();
    Environment env = Environments.createDockerEnvironment("java");
    components.registerEnvironment(env);
    ParDoPayload payload = ParDoTranslation.translateParDo(ParDo.of(new DoFn<String, String>() {

        @ProcessElement
        public void process(ProcessContext ctxt) {
        }
    }).withOutputTags(new TupleTag<>(), TupleTagList.empty()), PCollection.createPrimitiveOutputInternal(p, WindowingStrategy.globalDefault(), IsBounded.BOUNDED, StringUtf8Coder.of()), DoFnSchemaInformation.create(), Pipeline.create(), components);
    RehydratedComponents rehydratedComponents = RehydratedComponents.forComponents(components.toComponents());
    PTransform ptransform = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).setPayload(payload.toByteString()).build()).setEnvironmentId(components.getOnlyEnvironmentId()).build();
    Environment env1 = Environments.getEnvironment(ptransform, rehydratedComponents).get();
    assertThat(env1, equalTo(components.toComponents().getEnvironmentsOrThrow(ptransform.getEnvironmentId())));
}

Example 27 with Pipeline

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

the class PCollectionTranslationTest method testEncodeDecodeCycle.

@Test
public void testEncodeDecodeCycle() throws Exception {
    // Encode
    SdkComponents sdkComponents = SdkComponents.create();
    sdkComponents.registerEnvironment(Environments.createDockerEnvironment("java"));
    RunnerApi.PCollection protoCollection = PCollectionTranslation.toProto(testCollection, sdkComponents);
    RehydratedComponents protoComponents = RehydratedComponents.forComponents(sdkComponents.toComponents());
    // Decode
    Pipeline pipeline = Pipeline.create();
    PCollection<?> decodedCollection = PCollectionTranslation.fromProto(protoCollection, pipeline, protoComponents);
    // Verify
    assertThat(decodedCollection.getCoder(), equalTo(testCollection.getCoder()));
    assertThat(decodedCollection.getWindowingStrategy(), equalTo(testCollection.getWindowingStrategy().withEnvironmentId(sdkComponents.getOnlyEnvironmentId()).fixDefaults()));
    assertThat(decodedCollection.isBounded(), equalTo(testCollection.isBounded()));
}

Example 28 with Pipeline

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

the class OutputDeduplicator method ensureSingleProducer.

/**
 * Ensure that no {@link PCollection} output by any of the {@code stages} or {@code
 * unfusedTransforms} is produced by more than one of those stages or transforms.
 *
 * <p>For each {@link PCollection} output by multiple stages and/or transforms, each producer is
 * rewritten to produce a partial {@link PCollection}, which are then flattened together via an
 * introduced Flatten node which produces the original output.
 */
static DeduplicationResult ensureSingleProducer(QueryablePipeline pipeline, Collection<ExecutableStage> stages, Collection<PTransformNode> unfusedTransforms) {
    RunnerApi.Components.Builder unzippedComponents = pipeline.getComponents().toBuilder();
    Multimap<PCollectionNode, StageOrTransform> pcollectionProducers = getProducers(pipeline, stages, unfusedTransforms);
    Multimap<StageOrTransform, PCollectionNode> requiresNewOutput = HashMultimap.create();
    // ExecutableStage must also be rewritten to have updated outputs and transforms.
    for (Map.Entry<PCollectionNode, Collection<StageOrTransform>> collectionProducer : pcollectionProducers.asMap().entrySet()) {
        if (collectionProducer.getValue().size() > 1) {
            for (StageOrTransform producer : collectionProducer.getValue()) {
                requiresNewOutput.put(producer, collectionProducer.getKey());
            }
        }
    }
    Map<ExecutableStage, ExecutableStage> updatedStages = new LinkedHashMap<>();
    Map<String, PTransformNode> updatedTransforms = new LinkedHashMap<>();
    Multimap<String, PCollectionNode> originalToPartial = HashMultimap.create();
    for (Map.Entry<StageOrTransform, Collection<PCollectionNode>> deduplicationTargets : requiresNewOutput.asMap().entrySet()) {
        if (deduplicationTargets.getKey().getStage() != null) {
            StageDeduplication deduplication = deduplicatePCollections(deduplicationTargets.getKey().getStage(), deduplicationTargets.getValue(), unzippedComponents::containsPcollections);
            for (Entry<String, PCollectionNode> originalToPartialReplacement : deduplication.getOriginalToPartialPCollections().entrySet()) {
                originalToPartial.put(originalToPartialReplacement.getKey(), originalToPartialReplacement.getValue());
                unzippedComponents.putPcollections(originalToPartialReplacement.getValue().getId(), originalToPartialReplacement.getValue().getPCollection());
            }
            updatedStages.put(deduplicationTargets.getKey().getStage(), deduplication.getUpdatedStage());
        } else if (deduplicationTargets.getKey().getTransform() != null) {
            PTransformDeduplication deduplication = deduplicatePCollections(deduplicationTargets.getKey().getTransform(), deduplicationTargets.getValue(), unzippedComponents::containsPcollections);
            for (Entry<String, PCollectionNode> originalToPartialReplacement : deduplication.getOriginalToPartialPCollections().entrySet()) {
                originalToPartial.put(originalToPartialReplacement.getKey(), originalToPartialReplacement.getValue());
                unzippedComponents.putPcollections(originalToPartialReplacement.getValue().getId(), originalToPartialReplacement.getValue().getPCollection());
            }
            updatedTransforms.put(deduplicationTargets.getKey().getTransform().getId(), deduplication.getUpdatedTransform());
        } else {
            throw new IllegalStateException(String.format("%s with no %s or %s", StageOrTransform.class.getSimpleName(), ExecutableStage.class.getSimpleName(), PTransformNode.class.getSimpleName()));
        }
    }
    Set<PTransformNode> introducedFlattens = new LinkedHashSet<>();
    for (Map.Entry<String, Collection<PCollectionNode>> partialFlattenTargets : originalToPartial.asMap().entrySet()) {
        String flattenId = SyntheticComponents.uniqueId("unzipped_flatten", unzippedComponents::containsTransforms);
        PTransform flattenPartialPCollections = createFlattenOfPartials(flattenId, partialFlattenTargets.getKey(), partialFlattenTargets.getValue());
        unzippedComponents.putTransforms(flattenId, flattenPartialPCollections);
        introducedFlattens.add(PipelineNode.pTransform(flattenId, flattenPartialPCollections));
    }
    Components components = unzippedComponents.build();
    return DeduplicationResult.of(components, introducedFlattens, updatedStages, updatedTransforms);
}

Example 29 with Pipeline

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

the class FlinkStreamingPortablePipelineTranslator method translateFlatten.

private <T> void translateFlatten(String id, RunnerApi.Pipeline pipeline, StreamingTranslationContext context) {
    RunnerApi.PTransform transform = pipeline.getComponents().getTransformsOrThrow(id);
    Map<String, String> allInputs = transform.getInputsMap();
    if (allInputs.isEmpty()) {
        // create an empty dummy source to satisfy downstream operations
        // we cannot create an empty source in Flink, therefore we have to
        // add the flatMap that simply never forwards the single element
        long shutdownAfterIdleSourcesMs = context.getPipelineOptions().getShutdownSourcesAfterIdleMs();
        DataStreamSource<WindowedValue<byte[]>> dummySource = context.getExecutionEnvironment().addSource(new ImpulseSourceFunction(shutdownAfterIdleSourcesMs));
        DataStream<WindowedValue<T>> result = dummySource.<WindowedValue<T>>flatMap((s, collector) -> {
        // never return anything
        }).returns(new CoderTypeInformation<>(WindowedValue.getFullCoder((Coder<T>) VoidCoder.of(), GlobalWindow.Coder.INSTANCE), context.getPipelineOptions()));
        context.addDataStream(Iterables.getOnlyElement(transform.getOutputsMap().values()), result);
    } else {
        DataStream<T> result = null;
        // Determine DataStreams that we use as input several times. For those, we need to uniquify
        // input streams because Flink seems to swallow watermarks when we have a union of one and
        // the same stream.
        HashMultiset<DataStream<T>> inputCounts = HashMultiset.create();
        for (String input : allInputs.values()) {
            DataStream<T> current = context.getDataStreamOrThrow(input);
            inputCounts.add(current, 1);
        }
        for (String input : allInputs.values()) {
            DataStream<T> current = context.getDataStreamOrThrow(input);
            final int timesRequired = inputCounts.count(current);
            if (timesRequired > 1) {
                current = current.flatMap(new FlatMapFunction<T, T>() {

                    private static final long serialVersionUID = 1L;

                    @Override
                    public void flatMap(T t, Collector<T> collector) {
                        collector.collect(t);
                    }
                });
            }
            result = (result == null) ? current : result.union(current);
        }
        context.addDataStream(Iterables.getOnlyElement(transform.getOutputsMap().values()), result);
    }
}

Example 30 with Pipeline

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

the class FlinkStreamingPortablePipelineTranslator method translateStreamingImpulse.

private void translateStreamingImpulse(String id, RunnerApi.Pipeline pipeline, StreamingTranslationContext context) {
    RunnerApi.PTransform pTransform = pipeline.getComponents().getTransformsOrThrow(id);
    TypeInformation<WindowedValue<byte[]>> typeInfo = new CoderTypeInformation<>(WindowedValue.getFullCoder(ByteArrayCoder.of(), GlobalWindow.Coder.INSTANCE), context.getPipelineOptions());
    ObjectMapper objectMapper = new ObjectMapper();
    final int intervalMillis;
    final int messageCount;
    try {
        JsonNode config = objectMapper.readTree(pTransform.getSpec().getPayload().toByteArray());
        intervalMillis = config.path("interval_ms").asInt(100);
        messageCount = config.path("message_count").asInt(0);
    } catch (IOException e) {
        throw new RuntimeException("Failed to parse configuration for streaming impulse", e);
    }
    SingleOutputStreamOperator<WindowedValue<byte[]>> source = context.getExecutionEnvironment().addSource(new StreamingImpulseSource(intervalMillis, messageCount), StreamingImpulseSource.class.getSimpleName()).returns(typeInfo);
    context.addDataStream(Iterables.getOnlyElement(pTransform.getOutputsMap().values()), source);
}