Examples with StreamingWindowPipelineDetector com.cloudera.dataflow.spark.streaming.StreamingWindowPipelineDetector used on opensource projects

Example 1 with StreamingWindowPipelineDetector

use of com.cloudera.dataflow.spark.streaming.StreamingWindowPipelineDetector in project spark-dataflow by cloudera.

the class SparkPipelineRunner method run.

@Override
public EvaluationResult run(Pipeline pipeline) {
    try {
        // validate streaming configuration
        if (mOptions.isStreaming() && !(mOptions instanceof SparkStreamingPipelineOptions)) {
            throw new RuntimeException("A streaming job must be configured with " + SparkStreamingPipelineOptions.class.getSimpleName() + ", found " + mOptions.getClass().getSimpleName());
        }
        LOG.info("Executing pipeline using the SparkPipelineRunner.");
        JavaSparkContext jsc = SparkContextFactory.getSparkContext(mOptions.getSparkMaster(), mOptions.getAppName());
        if (mOptions.isStreaming()) {
            SparkPipelineTranslator translator = new StreamingTransformTranslator.Translator(new TransformTranslator.Translator());
            // if streaming - fixed window should be defined on all UNBOUNDED inputs
            StreamingWindowPipelineDetector streamingWindowPipelineDetector = new StreamingWindowPipelineDetector(translator);
            pipeline.traverseTopologically(streamingWindowPipelineDetector);
            if (!streamingWindowPipelineDetector.isWindowing()) {
                throw new IllegalStateException("Spark streaming pipeline must be windowed!");
            }
            Duration batchInterval = streamingWindowPipelineDetector.getBatchDuration();
            LOG.info("Setting Spark streaming batchInterval to {} msec", batchInterval.milliseconds());
            EvaluationContext ctxt = createStreamingEvaluationContext(jsc, pipeline, batchInterval);
            pipeline.traverseTopologically(new SparkPipelineEvaluator(ctxt, translator));
            ctxt.computeOutputs();
            LOG.info("Streaming pipeline construction complete. Starting execution..");
            ((StreamingEvaluationContext) ctxt).getStreamingContext().start();
            return ctxt;
        } else {
            EvaluationContext ctxt = new EvaluationContext(jsc, pipeline);
            SparkPipelineTranslator translator = new TransformTranslator.Translator();
            pipeline.traverseTopologically(new SparkPipelineEvaluator(ctxt, translator));
            ctxt.computeOutputs();
            LOG.info("Pipeline execution complete.");
            return ctxt;
        }
    } catch (Exception e) {
        // SparkProcessException), or just use the SparkException cause.
        if (e instanceof SparkException && e.getCause() != null) {
            if (e.getCause() instanceof SparkProcessContext.SparkProcessException && e.getCause().getCause() != null) {
                throw new RuntimeException(e.getCause().getCause());
            } else {
                throw new RuntimeException(e.getCause());
            }
        }
        // otherwise just wrap in a RuntimeException
        throw new RuntimeException(e);
    }
}