Examples with JobInfo org.apache.beam.runners.fnexecution.provisioning.JobInfo used on opensource projects

Example 1 with JobInfo

use of org.apache.beam.runners.fnexecution.provisioning.JobInfo in project beam by apache.

the class ReferenceCountingExecutableStageContextFactoryTest method testCreateReuseReleaseCreate.

@Test
public void testCreateReuseReleaseCreate() throws Exception {
    Creator creator = mock(Creator.class);
    ExecutableStageContext c1 = mock(ExecutableStageContext.class);
    ExecutableStageContext c2 = mock(ExecutableStageContext.class);
    ExecutableStageContext c3 = mock(ExecutableStageContext.class);
    ExecutableStageContext c4 = mock(ExecutableStageContext.class);
    when(creator.apply(any(JobInfo.class))).thenReturn(c1).thenReturn(c2).thenReturn(c3).thenReturn(c4);
    ReferenceCountingExecutableStageContextFactory factory = ReferenceCountingExecutableStageContextFactory.create(creator, (x) -> true);
    JobInfo jobA = mock(JobInfo.class);
    when(jobA.jobId()).thenReturn("jobA");
    JobInfo jobB = mock(JobInfo.class);
    when(jobB.jobId()).thenReturn("jobB");
    // 1 open jobA
    ExecutableStageContext ac1A = factory.get(jobA);
    // 1 open jobB
    ExecutableStageContext ac2B = factory.get(jobB);
    Assert.assertSame("Context should be cached and reused.", ac1A, // 2 open jobA
    factory.get(jobA));
    Assert.assertSame("Context should be cached and reused.", ac2B, // 2 open jobB
    factory.get(jobB));
    // 1 open jobA
    factory.release(ac1A);
    Assert.assertSame("Context should be cached and reused.", ac1A, // 2 open jobA
    factory.get(jobA));
    // 1 open jobA
    factory.release(ac1A);
    // 0 open jobA
    factory.release(ac1A);
    // 1 open jobA
    ExecutableStageContext ac3A = factory.get(jobA);
    Assert.assertNotSame("We should get a new instance.", ac1A, ac3A);
    Assert.assertSame("Context should be cached and reused.", ac3A, // 2 open jobA
    factory.get(jobA));
    // 1 open jobA
    factory.release(ac3A);
    // 0 open jobA
    factory.release(ac3A);
    Assert.assertSame("Context should be cached and reused.", ac2B, // 3 open jobB
    factory.get(jobB));
    // 2 open jobB
    factory.release(ac2B);
    // 1 open jobB
    factory.release(ac2B);
    // 0 open jobB
    factory.release(ac2B);
    // 1 open jobB
    ExecutableStageContext ac4B = factory.get(jobB);
    Assert.assertNotSame("We should get a new instance.", ac2B, ac4B);
    // 0 open jobB
    factory.release(ac4B);
}

Example 2 with JobInfo

use of org.apache.beam.runners.fnexecution.provisioning.JobInfo in project beam by apache.

the class SamzaJobInvoker method invokeWithExecutor.

@Override
protected JobInvocation invokeWithExecutor(RunnerApi.Pipeline pipeline, Struct options, @Nullable String retrievalToken, ListeningExecutorService executorService) {
    LOG.trace("Parsing pipeline options");
    final SamzaPortablePipelineOptions samzaOptions = PipelineOptionsTranslation.fromProto(options).as(SamzaPortablePipelineOptions.class);
    final PortablePipelineRunner pipelineRunner;
    if (Strings.isNullOrEmpty(samzaOptions.getOutputExecutablePath())) {
        pipelineRunner = new SamzaPipelineRunner(samzaOptions);
    } else {
        /*
       * To support --output_executable_path where bundles the input pipeline along with all
       * artifacts, etc. required to run the pipeline into a jar that can be executed later.
       */
        pipelineRunner = new PortablePipelineJarCreator(SamzaPipelineRunner.class);
    }
    final String invocationId = String.format("%s_%s", samzaOptions.getJobName(), UUID.randomUUID().toString());
    final JobInfo jobInfo = JobInfo.create(invocationId, samzaOptions.getJobName(), retrievalToken, options);
    return new JobInvocation(jobInfo, executorService, pipeline, pipelineRunner);
}

Example 3 with JobInfo

use of org.apache.beam.runners.fnexecution.provisioning.JobInfo in project beam by apache.

the class SparkPipelineRunner method run.

@Override
public PortablePipelineResult run(RunnerApi.Pipeline pipeline, JobInfo jobInfo) {
    SparkPortablePipelineTranslator translator;
    boolean isStreaming = pipelineOptions.isStreaming() || hasUnboundedPCollections(pipeline);
    if (isStreaming) {
        translator = new SparkStreamingPortablePipelineTranslator();
    } else {
        translator = new SparkBatchPortablePipelineTranslator();
    }
    // Expand any splittable DoFns within the graph to enable sizing and splitting of bundles.
    Pipeline pipelineWithSdfExpanded = ProtoOverrides.updateTransform(PTransformTranslation.PAR_DO_TRANSFORM_URN, pipeline, SplittableParDoExpander.createSizedReplacement());
    // Don't let the fuser fuse any subcomponents of native transforms.
    Pipeline trimmedPipeline = TrivialNativeTransformExpander.forKnownUrns(pipelineWithSdfExpanded, translator.knownUrns());
    // Fused pipeline proto.
    // TODO: Consider supporting partially-fused graphs.
    RunnerApi.Pipeline fusedPipeline = trimmedPipeline.getComponents().getTransformsMap().values().stream().anyMatch(proto -> ExecutableStage.URN.equals(proto.getSpec().getUrn())) ? trimmedPipeline : GreedyPipelineFuser.fuse(trimmedPipeline).toPipeline();
    prepareFilesToStage(pipelineOptions);
    PortablePipelineResult result;
    final JavaSparkContext jsc = SparkContextFactory.getSparkContext(pipelineOptions);
    final long startTime = Instant.now().getMillis();
    EventLoggingListener eventLoggingListener = startEventLoggingListener(jsc, pipelineOptions, startTime);
    // Initialize accumulators.
    AggregatorsAccumulator.init(pipelineOptions, jsc);
    MetricsEnvironment.setMetricsSupported(true);
    MetricsAccumulator.init(pipelineOptions, jsc);
    final SparkTranslationContext context = translator.createTranslationContext(jsc, pipelineOptions, jobInfo);
    final ExecutorService executorService = Executors.newSingleThreadExecutor();
    LOG.info(String.format("Running job %s on Spark master %s", jobInfo.jobId(), jsc.master()));
    if (isStreaming) {
        final JavaStreamingContext jssc = ((SparkStreamingTranslationContext) context).getStreamingContext();
        jssc.addStreamingListener(new JavaStreamingListenerWrapper(new AggregatorsAccumulator.AccumulatorCheckpointingSparkListener()));
        jssc.addStreamingListener(new JavaStreamingListenerWrapper(new MetricsAccumulator.AccumulatorCheckpointingSparkListener()));
        // Register user-defined listeners.
        for (JavaStreamingListener listener : pipelineOptions.as(SparkContextOptions.class).getListeners()) {
            LOG.info("Registered listener {}." + listener.getClass().getSimpleName());
            jssc.addStreamingListener(new JavaStreamingListenerWrapper(listener));
        }
        // Register Watermarks listener to broadcast the advanced WMs.
        jssc.addStreamingListener(new JavaStreamingListenerWrapper(new GlobalWatermarkHolder.WatermarkAdvancingStreamingListener()));
        jssc.checkpoint(pipelineOptions.getCheckpointDir());
        // Obtain timeout from options.
        Long timeout = pipelineOptions.as(SparkPortableStreamingPipelineOptions.class).getStreamingTimeoutMs();
        final Future<?> submissionFuture = executorService.submit(() -> {
            translator.translate(fusedPipeline, context);
            LOG.info(String.format("Job %s: Pipeline translated successfully. Computing outputs", jobInfo.jobId()));
            context.computeOutputs();
            jssc.start();
            try {
                jssc.awaitTerminationOrTimeout(timeout);
            } catch (InterruptedException e) {
                LOG.warn("Streaming context interrupted, shutting down.", e);
            }
            jssc.stop();
            LOG.info(String.format("Job %s finished.", jobInfo.jobId()));
        });
        result = new SparkPipelineResult.PortableStreamingMode(submissionFuture, jssc);
    } else {
        final Future<?> submissionFuture = executorService.submit(() -> {
            translator.translate(fusedPipeline, context);
            LOG.info(String.format("Job %s: Pipeline translated successfully. Computing outputs", jobInfo.jobId()));
            context.computeOutputs();
            LOG.info(String.format("Job %s finished.", jobInfo.jobId()));
        });
        result = new SparkPipelineResult.PortableBatchMode(submissionFuture, jsc);
    }
    executorService.shutdown();
    result.waitUntilFinish();
    MetricsPusher metricsPusher = new MetricsPusher(MetricsAccumulator.getInstance().value(), pipelineOptions.as(MetricsOptions.class), result);
    metricsPusher.start();
    if (eventLoggingListener != null) {
        eventLoggingListener.onApplicationStart(SparkCompat.buildSparkListenerApplicationStart(jsc, pipelineOptions, startTime, result));
        eventLoggingListener.onApplicationEnd(new SparkListenerApplicationEnd(Instant.now().getMillis()));
        eventLoggingListener.stop();
    }
    return result;
}

Example 4 with JobInfo

use of org.apache.beam.runners.fnexecution.provisioning.JobInfo in project beam by apache.

the class FlinkPipelineRunner method runPipelineWithTranslator.

private <T extends FlinkPortablePipelineTranslator.TranslationContext> PortablePipelineResult runPipelineWithTranslator(final Pipeline pipeline, JobInfo jobInfo, FlinkPortablePipelineTranslator<T> translator) throws Exception {
    LOG.info("Translating pipeline to Flink program.");
    // Expand any splittable ParDos within the graph to enable sizing and splitting of bundles.
    Pipeline pipelineWithSdfExpanded = ProtoOverrides.updateTransform(PTransformTranslation.PAR_DO_TRANSFORM_URN, pipeline, SplittableParDoExpander.createSizedReplacement());
    // Don't let the fuser fuse any subcomponents of native transforms.
    Pipeline trimmedPipeline = TrivialNativeTransformExpander.forKnownUrns(pipelineWithSdfExpanded, translator.knownUrns());
    // Fused pipeline proto.
    // TODO: Consider supporting partially-fused graphs.
    RunnerApi.Pipeline fusedPipeline = trimmedPipeline.getComponents().getTransformsMap().values().stream().anyMatch(proto -> ExecutableStage.URN.equals(proto.getSpec().getUrn())) ? trimmedPipeline : GreedyPipelineFuser.fuse(trimmedPipeline).toPipeline();
    FlinkPortablePipelineTranslator.Executor executor = translator.translate(translator.createTranslationContext(jobInfo, pipelineOptions, confDir, filesToStage), fusedPipeline);
    final JobExecutionResult result = executor.execute(pipelineOptions.getJobName());
    return createPortablePipelineResult(result, pipelineOptions);
}

Example 5 with JobInfo

use of org.apache.beam.runners.fnexecution.provisioning.JobInfo in project beam by apache.

the class FlinkPipelineRunner method main.

/**
 * Main method to be called only as the entry point to an executable jar with structure as defined
 * in {@link PortablePipelineJarUtils}.
 */
public static void main(String[] args) throws Exception {
    // Register standard file systems.
    FileSystems.setDefaultPipelineOptions(PipelineOptionsFactory.create());
    FlinkPipelineRunnerConfiguration configuration = parseArgs(args);
    String baseJobName = configuration.baseJobName == null ? PortablePipelineJarUtils.getDefaultJobName() : configuration.baseJobName;
    Preconditions.checkArgument(baseJobName != null, "No default job name found. Job name must be set using --base-job-name.");
    Pipeline pipeline = PortablePipelineJarUtils.getPipelineFromClasspath(baseJobName);
    Struct originalOptions = PortablePipelineJarUtils.getPipelineOptionsFromClasspath(baseJobName);
    // The retrieval token is only required by the legacy artifact service, which the Flink runner
    // no longer uses.
    String retrievalToken = ArtifactApi.CommitManifestResponse.Constants.NO_ARTIFACTS_STAGED_TOKEN.getValueDescriptor().getOptions().getExtension(RunnerApi.beamConstant);
    FlinkPipelineOptions flinkOptions = PipelineOptionsTranslation.fromProto(originalOptions).as(FlinkPipelineOptions.class);
    String invocationId = String.format("%s_%s", flinkOptions.getJobName(), UUID.randomUUID().toString());
    FlinkPipelineRunner runner = new FlinkPipelineRunner(flinkOptions, configuration.flinkConfDir, detectClassPathResourcesToStage(FlinkPipelineRunner.class.getClassLoader(), flinkOptions));
    JobInfo jobInfo = JobInfo.create(invocationId, flinkOptions.getJobName(), retrievalToken, PipelineOptionsTranslation.toProto(flinkOptions));
    try {
        runner.run(pipeline, jobInfo);
    } catch (Exception e) {
        throw new RuntimeException(String.format("Job %s failed.", invocationId), e);
    }
    LOG.info("Job {} finished successfully.", invocationId);
}