Search in sources :

Example 11 with BundleProcessor

use of org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor in project beam by apache.

the class RemoteExecutionTest method testExecution.

@Test
public void testExecution() throws Exception {
    launchSdkHarness(PipelineOptionsFactory.create());
    Pipeline p = Pipeline.create();
    p.apply("impulse", Impulse.create()).apply("create", ParDo.of(new DoFn<byte[], String>() {

        @ProcessElement
        public void process(ProcessContext ctxt) {
            ctxt.output("zero");
            ctxt.output("one");
            ctxt.output("two");
        }
    })).apply("len", ParDo.of(new DoFn<String, Long>() {

        @ProcessElement
        public void process(ProcessContext ctxt) {
            ctxt.output((long) ctxt.element().length());
        }
    })).apply("addKeys", WithKeys.of("foo")).setCoder(KvCoder.of(StringUtf8Coder.of(), BigEndianLongCoder.of())).apply("gbk", GroupByKey.create());
    RunnerApi.Pipeline pipelineProto = PipelineTranslation.toProto(p);
    FusedPipeline fused = GreedyPipelineFuser.fuse(pipelineProto);
    checkState(fused.getFusedStages().size() == 1, "Expected exactly one fused stage");
    ExecutableStage stage = fused.getFusedStages().iterator().next();
    ExecutableProcessBundleDescriptor descriptor = ProcessBundleDescriptors.fromExecutableStage("my_stage", stage, dataServer.getApiServiceDescriptor());
    BundleProcessor processor = controlClient.getProcessor(descriptor.getProcessBundleDescriptor(), descriptor.getRemoteInputDestinations());
    Map<String, ? super Coder<WindowedValue<?>>> remoteOutputCoders = descriptor.getRemoteOutputCoders();
    Map<String, Collection<? super WindowedValue<?>>> outputValues = new HashMap<>();
    Map<String, RemoteOutputReceiver<?>> outputReceivers = new HashMap<>();
    for (Entry<String, ? super Coder<WindowedValue<?>>> remoteOutputCoder : remoteOutputCoders.entrySet()) {
        List<? super WindowedValue<?>> outputContents = Collections.synchronizedList(new ArrayList<>());
        outputValues.put(remoteOutputCoder.getKey(), outputContents);
        outputReceivers.put(remoteOutputCoder.getKey(), RemoteOutputReceiver.of((Coder) remoteOutputCoder.getValue(), (FnDataReceiver<? super WindowedValue<?>>) outputContents::add));
    }
    try (RemoteBundle bundle = processor.newBundle(outputReceivers, BundleProgressHandler.ignored())) {
        Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow(new byte[0]));
    }
    for (Collection<? super WindowedValue<?>> windowedValues : outputValues.values()) {
        assertThat(windowedValues, containsInAnyOrder(valueInGlobalWindow(byteValueOf("foo", 4)), valueInGlobalWindow(byteValueOf("foo", 3)), valueInGlobalWindow(byteValueOf("foo", 3))));
    }
}
Also used : ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) HashMap(java.util.HashMap) ByteString(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) BundleProcessor(org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor) WindowedValue(org.apache.beam.sdk.util.WindowedValue) ExecutableStage(org.apache.beam.runners.core.construction.graph.ExecutableStage) ExecutableProcessBundleDescriptor(org.apache.beam.runners.fnexecution.control.ProcessBundleDescriptors.ExecutableProcessBundleDescriptor) KvCoder(org.apache.beam.sdk.coders.KvCoder) Coder(org.apache.beam.sdk.coders.Coder) StringUtf8Coder(org.apache.beam.sdk.coders.StringUtf8Coder) BigEndianLongCoder(org.apache.beam.sdk.coders.BigEndianLongCoder) FnDataReceiver(org.apache.beam.sdk.fn.data.FnDataReceiver) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) Pipeline(org.apache.beam.sdk.Pipeline) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) DoFn(org.apache.beam.sdk.transforms.DoFn) Collection(java.util.Collection) PCollection(org.apache.beam.sdk.values.PCollection) Test(org.junit.Test)

Example 12 with BundleProcessor

use of org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor in project beam by apache.

the class RemoteExecutionTest method testBundleProcessorThrowsExecutionExceptionWhenUserCodeThrows.

@Test
public void testBundleProcessorThrowsExecutionExceptionWhenUserCodeThrows() throws Exception {
    launchSdkHarness(PipelineOptionsFactory.create());
    Pipeline p = Pipeline.create();
    p.apply("impulse", Impulse.create()).apply("create", ParDo.of(new DoFn<byte[], KV<String, String>>() {

        @ProcessElement
        public void process(ProcessContext ctxt) throws Exception {
            String element = CoderUtils.decodeFromByteArray(StringUtf8Coder.of(), ctxt.element());
            if (element.equals("X")) {
                throw new Exception("testBundleExecutionFailure");
            }
            ctxt.output(KV.of(element, element));
        }
    })).apply("gbk", GroupByKey.create());
    RunnerApi.Pipeline pipelineProto = PipelineTranslation.toProto(p);
    FusedPipeline fused = GreedyPipelineFuser.fuse(pipelineProto);
    checkState(fused.getFusedStages().size() == 1, "Expected exactly one fused stage");
    ExecutableStage stage = fused.getFusedStages().iterator().next();
    ExecutableProcessBundleDescriptor descriptor = ProcessBundleDescriptors.fromExecutableStage("my_stage", stage, dataServer.getApiServiceDescriptor());
    BundleProcessor processor = controlClient.getProcessor(descriptor.getProcessBundleDescriptor(), descriptor.getRemoteInputDestinations());
    Map<String, ? super Coder<WindowedValue<?>>> remoteOutputCoders = descriptor.getRemoteOutputCoders();
    Map<String, Collection<? super WindowedValue<?>>> outputValues = new HashMap<>();
    Map<String, RemoteOutputReceiver<?>> outputReceivers = new HashMap<>();
    for (Entry<String, ? super Coder<WindowedValue<?>>> remoteOutputCoder : remoteOutputCoders.entrySet()) {
        List<? super WindowedValue<?>> outputContents = Collections.synchronizedList(new ArrayList<>());
        outputValues.put(remoteOutputCoder.getKey(), outputContents);
        outputReceivers.put(remoteOutputCoder.getKey(), RemoteOutputReceiver.of((Coder) remoteOutputCoder.getValue(), (FnDataReceiver<? super WindowedValue<?>>) outputContents::add));
    }
    try (RemoteBundle bundle = processor.newBundle(outputReceivers, BundleProgressHandler.ignored())) {
        Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow(CoderUtils.encodeToByteArray(StringUtf8Coder.of(), "Y")));
    }
    try {
        try (RemoteBundle bundle = processor.newBundle(outputReceivers, BundleProgressHandler.ignored())) {
            Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow(CoderUtils.encodeToByteArray(StringUtf8Coder.of(), "X")));
        }
        // Fail the test if we reach this point and never threw the exception.
        fail();
    } catch (ExecutionException e) {
        assertTrue(e.getMessage().contains("testBundleExecutionFailure"));
    }
    try (RemoteBundle bundle = processor.newBundle(outputReceivers, BundleProgressHandler.ignored())) {
        Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow(CoderUtils.encodeToByteArray(StringUtf8Coder.of(), "Z")));
    }
    for (Collection<? super WindowedValue<?>> windowedValues : outputValues.values()) {
        assertThat(windowedValues, containsInAnyOrder(valueInGlobalWindow(KV.of("Y", "Y")), valueInGlobalWindow(KV.of("Z", "Z"))));
    }
}
Also used : ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) HashMap(java.util.HashMap) ByteString(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) BundleProcessor(org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor) WindowedValue(org.apache.beam.sdk.util.WindowedValue) ExecutableStage(org.apache.beam.runners.core.construction.graph.ExecutableStage) ExecutionException(java.util.concurrent.ExecutionException) ExecutableProcessBundleDescriptor(org.apache.beam.runners.fnexecution.control.ProcessBundleDescriptors.ExecutableProcessBundleDescriptor) KvCoder(org.apache.beam.sdk.coders.KvCoder) Coder(org.apache.beam.sdk.coders.Coder) StringUtf8Coder(org.apache.beam.sdk.coders.StringUtf8Coder) BigEndianLongCoder(org.apache.beam.sdk.coders.BigEndianLongCoder) FnDataReceiver(org.apache.beam.sdk.fn.data.FnDataReceiver) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) KV(org.apache.beam.sdk.values.KV) CoderException(org.apache.beam.sdk.coders.CoderException) ExecutionException(java.util.concurrent.ExecutionException) Pipeline(org.apache.beam.sdk.Pipeline) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) Collection(java.util.Collection) PCollection(org.apache.beam.sdk.values.PCollection) Test(org.junit.Test)

Example 13 with BundleProcessor

use of org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor in project beam by apache.

the class RemoteExecutionTest method testExecutionWithSideInput.

@Test
public void testExecutionWithSideInput() throws Exception {
    launchSdkHarness(PipelineOptionsFactory.create());
    Pipeline p = Pipeline.create();
    addExperiment(p.getOptions().as(ExperimentalOptions.class), "beam_fn_api");
    // TODO(BEAM-10097): Remove experiment once all portable runners support this view type
    addExperiment(p.getOptions().as(ExperimentalOptions.class), "use_runner_v2");
    PCollection<String> input = p.apply("impulse", Impulse.create()).apply("create", ParDo.of(new DoFn<byte[], String>() {

        @ProcessElement
        public void process(ProcessContext ctxt) {
            ctxt.output("zero");
            ctxt.output("one");
            ctxt.output("two");
        }
    })).setCoder(StringUtf8Coder.of());
    PCollectionView<Iterable<String>> iterableView = input.apply("createIterableSideInput", View.asIterable());
    PCollectionView<Map<String, Iterable<String>>> multimapView = input.apply(WithKeys.of("key")).apply("createMultimapSideInput", View.asMultimap());
    input.apply("readSideInput", ParDo.of(new DoFn<String, KV<String, String>>() {

        @ProcessElement
        public void processElement(ProcessContext context) {
            for (String value : context.sideInput(iterableView)) {
                context.output(KV.of(context.element(), value));
            }
            for (Map.Entry<String, Iterable<String>> entry : context.sideInput(multimapView).entrySet()) {
                for (String value : entry.getValue()) {
                    context.output(KV.of(context.element(), entry.getKey() + ":" + value));
                }
            }
        }
    }).withSideInputs(iterableView, multimapView)).setCoder(KvCoder.of(StringUtf8Coder.of(), StringUtf8Coder.of())).apply("gbk", GroupByKey.create());
    RunnerApi.Pipeline pipelineProto = PipelineTranslation.toProto(p);
    FusedPipeline fused = GreedyPipelineFuser.fuse(pipelineProto);
    Optional<ExecutableStage> optionalStage = Iterables.tryFind(fused.getFusedStages(), (ExecutableStage stage) -> !stage.getSideInputs().isEmpty());
    checkState(optionalStage.isPresent(), "Expected a stage with side inputs.");
    ExecutableStage stage = optionalStage.get();
    ExecutableProcessBundleDescriptor descriptor = ProcessBundleDescriptors.fromExecutableStage("test_stage", stage, dataServer.getApiServiceDescriptor(), stateServer.getApiServiceDescriptor());
    BundleProcessor processor = controlClient.getProcessor(descriptor.getProcessBundleDescriptor(), descriptor.getRemoteInputDestinations(), stateDelegator);
    Map<String, Coder> remoteOutputCoders = descriptor.getRemoteOutputCoders();
    Map<String, Collection<WindowedValue<?>>> outputValues = new HashMap<>();
    Map<String, RemoteOutputReceiver<?>> outputReceivers = new HashMap<>();
    for (Entry<String, Coder> remoteOutputCoder : remoteOutputCoders.entrySet()) {
        List<WindowedValue<?>> outputContents = Collections.synchronizedList(new ArrayList<>());
        outputValues.put(remoteOutputCoder.getKey(), outputContents);
        outputReceivers.put(remoteOutputCoder.getKey(), RemoteOutputReceiver.of((Coder<WindowedValue<?>>) remoteOutputCoder.getValue(), outputContents::add));
    }
    StateRequestHandler stateRequestHandler = StateRequestHandlers.forSideInputHandlerFactory(descriptor.getSideInputSpecs(), new SideInputHandlerFactory() {

        @Override
        public <V, W extends BoundedWindow> IterableSideInputHandler<V, W> forIterableSideInput(String pTransformId, String sideInputId, Coder<V> elementCoder, Coder<W> windowCoder) {
            return new IterableSideInputHandler<V, W>() {

                @Override
                public Iterable<V> get(W window) {
                    return (Iterable) Arrays.asList("A", "B", "C");
                }

                @Override
                public Coder<V> elementCoder() {
                    return elementCoder;
                }
            };
        }

        @Override
        public <K, V, W extends BoundedWindow> MultimapSideInputHandler<K, V, W> forMultimapSideInput(String pTransformId, String sideInputId, KvCoder<K, V> elementCoder, Coder<W> windowCoder) {
            return new MultimapSideInputHandler<K, V, W>() {

                @Override
                public Iterable<K> get(W window) {
                    return (Iterable) Arrays.asList("key1", "key2");
                }

                @Override
                public Iterable<V> get(K key, W window) {
                    if ("key1".equals(key)) {
                        return (Iterable) Arrays.asList("H", "I", "J");
                    } else if ("key2".equals(key)) {
                        return (Iterable) Arrays.asList("M", "N", "O");
                    }
                    return Collections.emptyList();
                }

                @Override
                public Coder<K> keyCoder() {
                    return elementCoder.getKeyCoder();
                }

                @Override
                public Coder<V> valueCoder() {
                    return elementCoder.getValueCoder();
                }
            };
        }
    });
    BundleProgressHandler progressHandler = BundleProgressHandler.ignored();
    try (RemoteBundle bundle = processor.newBundle(outputReceivers, stateRequestHandler, progressHandler)) {
        Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow("X"));
        Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow("Y"));
    }
    for (Collection<WindowedValue<?>> windowedValues : outputValues.values()) {
        assertThat(windowedValues, containsInAnyOrder(valueInGlobalWindow(KV.of("X", "A")), valueInGlobalWindow(KV.of("X", "B")), valueInGlobalWindow(KV.of("X", "C")), valueInGlobalWindow(KV.of("X", "key1:H")), valueInGlobalWindow(KV.of("X", "key1:I")), valueInGlobalWindow(KV.of("X", "key1:J")), valueInGlobalWindow(KV.of("X", "key2:M")), valueInGlobalWindow(KV.of("X", "key2:N")), valueInGlobalWindow(KV.of("X", "key2:O")), valueInGlobalWindow(KV.of("Y", "A")), valueInGlobalWindow(KV.of("Y", "B")), valueInGlobalWindow(KV.of("Y", "C")), valueInGlobalWindow(KV.of("Y", "key1:H")), valueInGlobalWindow(KV.of("Y", "key1:I")), valueInGlobalWindow(KV.of("Y", "key1:J")), valueInGlobalWindow(KV.of("Y", "key2:M")), valueInGlobalWindow(KV.of("Y", "key2:N")), valueInGlobalWindow(KV.of("Y", "key2:O"))));
    }
}
Also used : StateRequestHandler(org.apache.beam.runners.fnexecution.state.StateRequestHandler) IsEmptyIterable(org.hamcrest.collection.IsEmptyIterable) ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) HashMap(java.util.HashMap) ExperimentalOptions(org.apache.beam.sdk.options.ExperimentalOptions) ByteString(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) Entry(java.util.Map.Entry) BundleProcessor(org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor) WindowedValue(org.apache.beam.sdk.util.WindowedValue) KV(org.apache.beam.sdk.values.KV) SideInputHandlerFactory(org.apache.beam.runners.fnexecution.state.StateRequestHandlers.SideInputHandlerFactory) ExecutableStage(org.apache.beam.runners.core.construction.graph.ExecutableStage) ExecutableProcessBundleDescriptor(org.apache.beam.runners.fnexecution.control.ProcessBundleDescriptors.ExecutableProcessBundleDescriptor) KvCoder(org.apache.beam.sdk.coders.KvCoder) Coder(org.apache.beam.sdk.coders.Coder) StringUtf8Coder(org.apache.beam.sdk.coders.StringUtf8Coder) BigEndianLongCoder(org.apache.beam.sdk.coders.BigEndianLongCoder) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) KV(org.apache.beam.sdk.values.KV) Pipeline(org.apache.beam.sdk.Pipeline) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) IterableSideInputHandler(org.apache.beam.runners.fnexecution.state.StateRequestHandlers.IterableSideInputHandler) Collection(java.util.Collection) PCollection(org.apache.beam.sdk.values.PCollection) MultimapSideInputHandler(org.apache.beam.runners.fnexecution.state.StateRequestHandlers.MultimapSideInputHandler) ImmutableMap(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableMap) Map(java.util.Map) ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) HashMap(java.util.HashMap) ConcurrentMap(java.util.concurrent.ConcurrentMap) Test(org.junit.Test)

Example 14 with BundleProcessor

use of org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor in project beam by apache.

the class RemoteExecutionTest method testMetrics.

@Test
@SuppressWarnings("FutureReturnValueIgnored")
public void testMetrics() throws Exception {
    launchSdkHarness(PipelineOptionsFactory.create());
    MetricsDoFn metricsDoFn = new MetricsDoFn();
    Pipeline p = Pipeline.create();
    PCollection<String> input = p.apply("impulse", Impulse.create()).apply("create", ParDo.of(metricsDoFn)).setCoder(StringUtf8Coder.of());
    SingleOutput<String, String> pardo = ParDo.of(new DoFn<String, String>() {

        @ProcessElement
        public void process(ProcessContext ctxt) {
            // Output the element twice to keep unique numbers in asserts, 6 output elements.
            ctxt.output(ctxt.element());
            ctxt.output(ctxt.element());
        }
    });
    input.apply("processA", pardo).setCoder(StringUtf8Coder.of());
    input.apply("processB", pardo).setCoder(StringUtf8Coder.of());
    RunnerApi.Pipeline pipelineProto = PipelineTranslation.toProto(p);
    FusedPipeline fused = GreedyPipelineFuser.fuse(pipelineProto);
    Optional<ExecutableStage> optionalStage = Iterables.tryFind(fused.getFusedStages(), (ExecutableStage stage) -> true);
    checkState(optionalStage.isPresent(), "Expected a stage with side inputs.");
    ExecutableStage stage = optionalStage.get();
    ExecutableProcessBundleDescriptor descriptor = ProcessBundleDescriptors.fromExecutableStage("test_stage", stage, dataServer.getApiServiceDescriptor(), stateServer.getApiServiceDescriptor());
    BundleProcessor processor = controlClient.getProcessor(descriptor.getProcessBundleDescriptor(), descriptor.getRemoteInputDestinations(), stateDelegator);
    Map<String, Coder> remoteOutputCoders = descriptor.getRemoteOutputCoders();
    Map<String, RemoteOutputReceiver<?>> outputReceivers = new HashMap<>();
    for (Entry<String, Coder> remoteOutputCoder : remoteOutputCoders.entrySet()) {
        List<WindowedValue<?>> outputContents = Collections.synchronizedList(new ArrayList<>());
        outputReceivers.put(remoteOutputCoder.getKey(), RemoteOutputReceiver.of((Coder<WindowedValue<?>>) remoteOutputCoder.getValue(), outputContents::add));
    }
    final String testPTransformId = "create-ParMultiDo-Metrics-";
    BundleProgressHandler progressHandler = new BundleProgressHandler() {

        @Override
        public void onProgress(ProcessBundleProgressResponse response) {
            MetricsDoFn.ALLOW_COMPLETION.get(metricsDoFn.uuid).countDown();
            List<Matcher<MonitoringInfo>> matchers = new ArrayList<>();
            // We expect all user counters except for the ones in @FinishBundle
            // Since non-user metrics are registered at bundle creation time, they will still report
            // values most of which will be 0.
            SimpleMonitoringInfoBuilder builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_SUM_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.PROCESS_USER_COUNTER_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64SumValue(1);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_SUM_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.START_USER_COUNTER_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64SumValue(10);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_SUM_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.FINISH_USER_COUNTER_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            matchers.add(not(MonitoringInfoMatchers.matchSetFields(builder.build())));
            // User Distributions.
            builder.setUrn(MonitoringInfoConstants.Urns.USER_DISTRIBUTION_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.PROCESS_USER_DISTRIBUTION_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64DistributionValue(DistributionData.create(1, 1, 1, 1));
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_DISTRIBUTION_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.START_USER_DISTRIBUTION_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64DistributionValue(DistributionData.create(10, 1, 10, 10));
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_DISTRIBUTION_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.FINISH_USER_DISTRIBUTION_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            matchers.add(not(MonitoringInfoMatchers.matchSetFields(builder.build())));
            assertThat(response.getMonitoringInfosList(), Matchers.hasItems(matchers.toArray(new Matcher[0])));
        }

        @Override
        public void onCompleted(ProcessBundleResponse response) {
            List<Matcher<MonitoringInfo>> matchers = new ArrayList<>();
            // User Counters.
            SimpleMonitoringInfoBuilder builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_SUM_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.PROCESS_USER_COUNTER_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64SumValue(1);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_SUM_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.START_USER_COUNTER_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64SumValue(10);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_SUM_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.FINISH_USER_COUNTER_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64SumValue(100);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            // User Distributions.
            builder.setUrn(MonitoringInfoConstants.Urns.USER_DISTRIBUTION_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.PROCESS_USER_DISTRIBUTION_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64DistributionValue(DistributionData.create(1, 1, 1, 1));
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_DISTRIBUTION_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.START_USER_DISTRIBUTION_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64DistributionValue(DistributionData.create(10, 1, 10, 10));
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.USER_DISTRIBUTION_INT64).setLabel(MonitoringInfoConstants.Labels.NAMESPACE, RemoteExecutionTest.class.getName()).setLabel(MonitoringInfoConstants.Labels.NAME, MetricsDoFn.FINISH_USER_DISTRIBUTION_NAME);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            builder.setInt64DistributionValue(DistributionData.create(100, 1, 100, 100));
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            // The element counter should be counted only once for the pcollection.
            // So there should be only two elements.
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.ELEMENT_COUNT);
            builder.setLabel(MonitoringInfoConstants.Labels.PCOLLECTION, "impulse.out");
            builder.setInt64SumValue(1);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.ELEMENT_COUNT);
            builder.setLabel(MonitoringInfoConstants.Labels.PCOLLECTION, "create/ParMultiDo(Metrics).output");
            builder.setInt64SumValue(3);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            // Verify that the element count is not double counted if two PCollections consume it.
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.ELEMENT_COUNT);
            builder.setLabel(MonitoringInfoConstants.Labels.PCOLLECTION, "processA/ParMultiDo(Anonymous).output");
            builder.setInt64SumValue(6);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.ELEMENT_COUNT);
            builder.setLabel(MonitoringInfoConstants.Labels.PCOLLECTION, "processB/ParMultiDo(Anonymous).output");
            builder.setInt64SumValue(6);
            matchers.add(MonitoringInfoMatchers.matchSetFields(builder.build()));
            // Check for execution time metrics for the testPTransformId
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(MonitoringInfoConstants.Urns.START_BUNDLE_MSECS);
            builder.setType(TypeUrns.SUM_INT64_TYPE);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            matchers.add(allOf(MonitoringInfoMatchers.matchSetFields(builder.build()), MonitoringInfoMatchers.counterValueGreaterThanOrEqualTo(1)));
            // Check for execution time metrics for the testPTransformId
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(Urns.PROCESS_BUNDLE_MSECS);
            builder.setType(TypeUrns.SUM_INT64_TYPE);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            matchers.add(allOf(MonitoringInfoMatchers.matchSetFields(builder.build()), MonitoringInfoMatchers.counterValueGreaterThanOrEqualTo(2)));
            builder = new SimpleMonitoringInfoBuilder();
            builder.setUrn(Urns.FINISH_BUNDLE_MSECS);
            builder.setType(TypeUrns.SUM_INT64_TYPE);
            builder.setLabel(MonitoringInfoConstants.Labels.PTRANSFORM, testPTransformId);
            matchers.add(allOf(MonitoringInfoMatchers.matchSetFields(builder.build()), MonitoringInfoMatchers.counterValueGreaterThanOrEqualTo(3)));
            assertThat(response.getMonitoringInfosList(), Matchers.hasItems(matchers.toArray(new Matcher[0])));
        }
    };
    ExecutorService executor = Executors.newSingleThreadExecutor();
    try (RemoteBundle bundle = processor.newBundle(outputReceivers, StateRequestHandler.unsupported(), progressHandler)) {
        Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow(CoderUtils.encodeToByteArray(StringUtf8Coder.of(), "X")));
        executor.submit(() -> {
            checkState(MetricsDoFn.AFTER_PROCESS.get(metricsDoFn.uuid).await(60, TimeUnit.SECONDS), "Runner waited too long for DoFn to get to AFTER_PROCESS.");
            bundle.requestProgress();
            return (Void) null;
        });
    }
    executor.shutdown();
}
Also used : ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) HashMap(java.util.HashMap) Matcher(org.hamcrest.Matcher) ArrayList(java.util.ArrayList) ByteString(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) SimpleMonitoringInfoBuilder(org.apache.beam.runners.core.metrics.SimpleMonitoringInfoBuilder) BundleProcessor(org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor) WindowedValue(org.apache.beam.sdk.util.WindowedValue) ExecutableStage(org.apache.beam.runners.core.construction.graph.ExecutableStage) ExecutableProcessBundleDescriptor(org.apache.beam.runners.fnexecution.control.ProcessBundleDescriptors.ExecutableProcessBundleDescriptor) ProcessBundleResponse(org.apache.beam.model.fnexecution.v1.BeamFnApi.ProcessBundleResponse) KvCoder(org.apache.beam.sdk.coders.KvCoder) Coder(org.apache.beam.sdk.coders.Coder) StringUtf8Coder(org.apache.beam.sdk.coders.StringUtf8Coder) BigEndianLongCoder(org.apache.beam.sdk.coders.BigEndianLongCoder) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) Pipeline(org.apache.beam.sdk.Pipeline) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) ScheduledExecutorService(java.util.concurrent.ScheduledExecutorService) ExecutorService(java.util.concurrent.ExecutorService) ProcessBundleProgressResponse(org.apache.beam.model.fnexecution.v1.BeamFnApi.ProcessBundleProgressResponse) Test(org.junit.Test)

Example 15 with BundleProcessor

use of org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor in project beam by apache.

the class RemoteExecutionTest method testSplit.

@Test(timeout = 60000L)
public void testSplit() throws Exception {
    launchSdkHarness(PipelineOptionsFactory.create());
    Pipeline p = Pipeline.create();
    p.apply("impulse", Impulse.create()).apply("create", ParDo.of(new DoFn<byte[], String>() {

        @ProcessElement
        public void process(ProcessContext ctxt) {
            ctxt.output("zero");
            ctxt.output(WaitingTillSplitRestrictionTracker.WAIT_TILL_SPLIT);
            ctxt.output("two");
        }
    })).apply("forceSplit", ParDo.of(new DoFn<String, String>() {

        @GetInitialRestriction
        public String getInitialRestriction(@Element String element) {
            return element;
        }

        @NewTracker
        public WaitingTillSplitRestrictionTracker newTracker(@Restriction String restriction) {
            return new WaitingTillSplitRestrictionTracker(restriction);
        }

        @ProcessElement
        public void process(RestrictionTracker<String, Void> tracker, ProcessContext context) {
            while (tracker.tryClaim(null)) {
            }
            context.output(tracker.currentRestriction());
        }
    })).apply("addKeys", WithKeys.of("foo")).setCoder(KvCoder.of(StringUtf8Coder.of(), StringUtf8Coder.of())).apply("gbk", GroupByKey.create());
    RunnerApi.Pipeline pipeline = PipelineTranslation.toProto(p);
    // Expand any splittable DoFns within the graph to enable sizing and splitting of bundles.
    RunnerApi.Pipeline pipelineWithSdfExpanded = ProtoOverrides.updateTransform(PTransformTranslation.PAR_DO_TRANSFORM_URN, pipeline, SplittableParDoExpander.createSizedReplacement());
    FusedPipeline fused = GreedyPipelineFuser.fuse(pipelineWithSdfExpanded);
    // Find the fused stage with the SDF ProcessSizedElementAndRestriction transform
    Optional<ExecutableStage> optionalStage = Iterables.tryFind(fused.getFusedStages(), (ExecutableStage stage) -> Iterables.filter(stage.getTransforms(), (PTransformNode node) -> PTransformTranslation.SPLITTABLE_PROCESS_SIZED_ELEMENTS_AND_RESTRICTIONS_URN.equals(node.getTransform().getSpec().getUrn())).iterator().hasNext());
    checkState(optionalStage.isPresent(), "Expected a stage with SDF ProcessSizedElementAndRestriction.");
    ExecutableStage stage = optionalStage.get();
    ExecutableProcessBundleDescriptor descriptor = ProcessBundleDescriptors.fromExecutableStage("my_stage", stage, dataServer.getApiServiceDescriptor());
    BundleProcessor processor = controlClient.getProcessor(descriptor.getProcessBundleDescriptor(), descriptor.getRemoteInputDestinations());
    Map<String, ? super Coder<WindowedValue<?>>> remoteOutputCoders = descriptor.getRemoteOutputCoders();
    Map<String, Collection<? super WindowedValue<?>>> outputValues = new HashMap<>();
    Map<String, RemoteOutputReceiver<?>> outputReceivers = new HashMap<>();
    for (Entry<String, ? super Coder<WindowedValue<?>>> remoteOutputCoder : remoteOutputCoders.entrySet()) {
        List<? super WindowedValue<?>> outputContents = Collections.synchronizedList(new ArrayList<>());
        outputValues.put(remoteOutputCoder.getKey(), outputContents);
        outputReceivers.put(remoteOutputCoder.getKey(), RemoteOutputReceiver.of((Coder) remoteOutputCoder.getValue(), (FnDataReceiver<? super WindowedValue<?>>) outputContents::add));
    }
    List<ProcessBundleSplitResponse> splitResponses = new ArrayList<>();
    List<ProcessBundleResponse> checkpointResponses = new ArrayList<>();
    List<String> requestsFinalization = new ArrayList<>();
    ScheduledExecutorService executor = Executors.newSingleThreadScheduledExecutor();
    ScheduledFuture<Object> future;
    // Execute the remote bundle.
    try (RemoteBundle bundle = processor.newBundle(outputReceivers, Collections.emptyMap(), StateRequestHandler.unsupported(), BundleProgressHandler.ignored(), splitResponses::add, checkpointResponses::add, requestsFinalization::add)) {
        Iterables.getOnlyElement(bundle.getInputReceivers().values()).accept(valueInGlobalWindow(sdfSizedElementAndRestrictionForTest(WaitingTillSplitRestrictionTracker.WAIT_TILL_SPLIT)));
        // Keep sending splits until the bundle terminates.
        future = (ScheduledFuture) executor.scheduleWithFixedDelay(() -> bundle.split(0.5), 0L, 100L, TimeUnit.MILLISECONDS);
    }
    future.cancel(false);
    executor.shutdown();
    assertTrue(requestsFinalization.isEmpty());
    assertTrue(checkpointResponses.isEmpty());
    // We only validate the last split response since it is the only one that could possibly
    // contain the SDF split, all others will be a reduction in the ChannelSplit range.
    assertFalse(splitResponses.isEmpty());
    ProcessBundleSplitResponse splitResponse = splitResponses.get(splitResponses.size() - 1);
    ChannelSplit channelSplit = Iterables.getOnlyElement(splitResponse.getChannelSplitsList());
    // There is only one outcome for the final split that can happen since the SDF is blocking the
    // bundle from completing and hence needed to be split.
    assertEquals(-1L, channelSplit.getLastPrimaryElement());
    assertEquals(1L, channelSplit.getFirstResidualElement());
    assertEquals(1, splitResponse.getPrimaryRootsCount());
    assertEquals(1, splitResponse.getResidualRootsCount());
    assertThat(Iterables.getOnlyElement(outputValues.values()), containsInAnyOrder(valueInGlobalWindow(KV.of("foo", WaitingTillSplitRestrictionTracker.PRIMARY))));
}
Also used : ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) HashMap(java.util.HashMap) PTransformNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode) ArrayList(java.util.ArrayList) ByteString(org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) BundleProcessor(org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor) WindowedValue(org.apache.beam.sdk.util.WindowedValue) ExecutableStage(org.apache.beam.runners.core.construction.graph.ExecutableStage) ExecutableProcessBundleDescriptor(org.apache.beam.runners.fnexecution.control.ProcessBundleDescriptors.ExecutableProcessBundleDescriptor) ProcessBundleResponse(org.apache.beam.model.fnexecution.v1.BeamFnApi.ProcessBundleResponse) KvCoder(org.apache.beam.sdk.coders.KvCoder) Coder(org.apache.beam.sdk.coders.Coder) StringUtf8Coder(org.apache.beam.sdk.coders.StringUtf8Coder) BigEndianLongCoder(org.apache.beam.sdk.coders.BigEndianLongCoder) ScheduledExecutorService(java.util.concurrent.ScheduledExecutorService) FnDataReceiver(org.apache.beam.sdk.fn.data.FnDataReceiver) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) Pipeline(org.apache.beam.sdk.Pipeline) FusedPipeline(org.apache.beam.runners.core.construction.graph.FusedPipeline) ProcessBundleSplitResponse(org.apache.beam.model.fnexecution.v1.BeamFnApi.ProcessBundleSplitResponse) DoFn(org.apache.beam.sdk.transforms.DoFn) Collection(java.util.Collection) PCollection(org.apache.beam.sdk.values.PCollection) ChannelSplit(org.apache.beam.model.fnexecution.v1.BeamFnApi.ProcessBundleSplitResponse.ChannelSplit) Test(org.junit.Test)

Aggregations

BundleProcessor (org.apache.beam.runners.fnexecution.control.SdkHarnessClient.BundleProcessor)26 Test (org.junit.Test)26 ByteString (org.apache.beam.vendor.grpc.v1p43p2.com.google.protobuf.ByteString)23 CloseableFnDataReceiver (org.apache.beam.sdk.fn.data.CloseableFnDataReceiver)14 CompletableFuture (java.util.concurrent.CompletableFuture)13 InstructionResponse (org.apache.beam.model.fnexecution.v1.BeamFnApi.InstructionResponse)13 ProcessBundleResponse (org.apache.beam.model.fnexecution.v1.BeamFnApi.ProcessBundleResponse)12 WindowedValue (org.apache.beam.sdk.util.WindowedValue)11 HashMap (java.util.HashMap)10 ConcurrentHashMap (java.util.concurrent.ConcurrentHashMap)10 RunnerApi (org.apache.beam.model.pipeline.v1.RunnerApi)10 ExecutableStage (org.apache.beam.runners.core.construction.graph.ExecutableStage)10 FusedPipeline (org.apache.beam.runners.core.construction.graph.FusedPipeline)10 ExecutableProcessBundleDescriptor (org.apache.beam.runners.fnexecution.control.ProcessBundleDescriptors.ExecutableProcessBundleDescriptor)10 Pipeline (org.apache.beam.sdk.Pipeline)10 BigEndianLongCoder (org.apache.beam.sdk.coders.BigEndianLongCoder)10 Coder (org.apache.beam.sdk.coders.Coder)10 KvCoder (org.apache.beam.sdk.coders.KvCoder)10 StringUtf8Coder (org.apache.beam.sdk.coders.StringUtf8Coder)10 Collection (java.util.Collection)9