Writing an an unbounded collection to GCS

Question

I have seen many questions on the same topic. But, I am still having problem with writing to GCS. I am reading the topic from pubsub and trying to push this to GCS. I have referred to this link. But, couldn't find the IOChannelUtils in the latest beam packages.

PCollection<String> details = pipeline
            .apply(PubsubIO.readStrings().fromTopic("/topics/<project>/sampleTopic"));

PCollection<KV<String, String>> keyedStream = details.apply(WithKeys.of(new SerializableFunction<String, String>() {
        public String apply(String s) {
            return "constant";
        }
    }));

    PCollection<KV<String, Iterable<String>>> keyedWindows = keyedStream.apply(Window.<KV<String, String>>into(FixedWindows.of(ONE_MIN)).withAllowedLateness(ONE_DAY)
            .triggering(AfterWatermark.pastEndOfWindow().withEarlyFirings(AfterPane.elementCountAtLeast(10))
                    .withLateFirings(AfterFirst.of(AfterPane.elementCountAtLeast(10),
                            AfterProcessingTime.pastFirstElementInPane().plusDelayOf(TEN_SECONDS))))
            .discardingFiredPanes()).apply(GroupByKey.create());

    PCollection<Iterable<String>> windows = keyedWindows.apply(Values.create());

This I have taken from many other similar topics in stack overflow. Now, I understand that, TextIO do support unbounded PCollection write option with withWindowedWrites and withNumShards.

ref : Writing to Google Cloud Storage from PubSub using Cloud Dataflow using DoFn

But, I did not understand, how I should do this.

I am trying to write to GCS as follows.

FilenamePolicy policy = DefaultFilenamePolicy.constructUsingStandardParameters(
            StaticValueProvider.of(outputDirectory), DefaultFilenamePolicy.DEFAULT_SHARD_TEMPLATE, "");

    details.apply(TextIO.write().to("gs://<bucket>/topicfile").withWindowedWrites()
            .withFilenamePolicy(policy).withNumShards(4));

I do not have sufficient points to add comments to those topics in Stack Overflow, hence I am raising it as a different question.

score 5 · Accepted Answer · answered Aug 08 '17 at 04:41

I could solve this issue by modifying the Windowing as given below

PCollection<String> streamedDataWindows = streamedData.apply(Window.<String>into(new GlobalWindows())
            .triggering(Repeatedly
                    .forever(AfterProcessingTime
                            .pastFirstElementInPane()
                            .plusDelayOf(Duration.standardSeconds(30))
                        )).withAllowedLateness(Duration.standardDays(1)).discardingFiredPanes());

 streamedDataWindows.apply(TextIO.write().to(CLOUD_STORAGE).withWindowedWrites().withNumShards(1).withFilenamePolicy(new PerWindowFiles()));


public static class PerWindowFiles extends FileBasedSink.FilenamePolicy {

public ResourceId windowedFilename(ResourceId outputDirectory, WindowedContext context, String extension) {

// OVERRIDE THE FILE NAME CREATION
}

}

Though I could solve it like this, I am still not sure about the windowing concept here. I will add more details as an when I find it. If anyone has more understanding, please add more details. Thanks

score 3 · Answer 2 · answered Aug 06 '17 at 05:33

3

Check out this Pub/Sub to GCS Pipeline which provides a full example of writing windowed files to GCS.

answered Aug 06 '17 at 05:33

Ryan McDowell

783
5
8

Hey.. thanks for the answer. I just could complete it a few minutes before. I will update this answer with the approach that I took. Thanks a lot again! – Balu Aug 06 '17 at 05:41

Writing an an unbounded collection to GCS

2 Answers2