Examples with SparkPortablePipelineTranslator org.apache.beam.runners.spark.translation.SparkPortablePipelineTranslator used on opensource projects

Example 1 with SparkPortablePipelineTranslator

use of org.apache.beam.runners.spark.translation.SparkPortablePipelineTranslator 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;
}