Apache Beam Not Saving Unbounded Data To Text File

Question

I've created a Pipeline to save Google Cloud Pubsub messages into text files using Apache Beam and Java. Whenever I run the pipeline within Google Dataflow with --runner=DataflowRunner the messages are saved correctly.

However, when I run the same pipeline with --runner=DirerctRunner the messages are not saved.

I can watch the events coming through the pipeline, but nothing happens.

The pipeline is the code below:

public static void main(String[] args) {
    ExerciseOptions options = PipelineOptionsFactory.fromArgs(args).withValidation().as(ExerciseOptions.class);

    Pipeline pipeline = Pipeline.create(options);

    pipeline
      .apply("Read Messages from Pubsub",
        PubsubIO
          .readStrings()
          .fromTopic(options.getTopicName()))

      .apply("Set event timestamp", ParDo.of(new DoFn<String, String>() {
        @ProcessElement
        public void processElement(ProcessContext context) {
          context.outputWithTimestamp(context.element(), Instant.now());
        }
      }))

      .apply("Windowing", Window.into(FixedWindows.of(Duration.standardMinutes(5))))

      .apply("Write to File",
        TextIO
          .write()
          .withWindowedWrites()
          .withNumShards(1)
          .to(options.getOutputPrefix()));

    pipeline.run();
  }

What I'm doing wrong? Is it possible to run this pipeline locally?

That looks like a bug in the DirectRunner, I filed issues.apache.org/jira/browse/BEAM-5488 — Lukasz Cwik

ineersa ineersa · Accepted Answer · 2018-10-05T09:36:35

I was facing same problem as yours, while testing pipeline. PubSubIO not working correctly with DirectRunner and TextIO.

I found some kind of workaround for this issue with triggering.

.apply(
                    "2 minutes window",
                    Window
                            .configure()
                            .triggering(
                                    Repeatedly.forever(
                                            AfterFirst.of(
                                                AfterPane.elementCountAtLeast(10),
                                                AfterProcessingTime
                                                        .pastFirstElementInPane()
                                                        .plusDelayOf(Duration.standardMinutes(2))
                                            )
                                    )
                            )
                            .into(
                                FixedWindows.of(
                                        Duration.standardMinutes(2)
                                )
                            )
            )

This way files are written as it should. Hope this will help someone.

Apache Beam Not Saving Unbounded Data To Text File

1 Answers