How can I use the results of a batch spark execution to a streaming one?

114 views Asked by At

I would like to use the results of the batch execution (which computes the classic word-count example) , which are saved in the "counts" variable, as a start for the streaming job which begins right after. Is this possible? For example if the batch has computed (Hello,2), (World, 1) and the first word from streaming is "Hello" I would like the first result to be (Hello,3), (World, 1). Thank you

public class UpdateWordCount {

private static final Pattern SPACE = Pattern.compile(" ");

public static void main(String args[]){

    // Create the context with a 1 second batch size
    SparkConf sparkConf = new SparkConf().setAppName("JavaNetworkWordCount").setMaster("local[2]");
    JavaSparkContext ctx = new JavaSparkContext(sparkConf);


    ArrayList<String> file=new ArrayList<String>(); file.add("Hello");file.add("Hello");file.add("World");
    JavaRDD<String> linesB = ctx.parallelize(file);

    JavaRDD<String> wordsB = linesB.flatMap(new FlatMapFunction<String, String>() {
      public Iterable<String> call(String s) {
        return Arrays.asList(SPACE.split(s));
      }
    });

    JavaPairRDD<String, Integer> ones = wordsB.mapToPair(new PairFunction<String, String, Integer>() {
      public Tuple2<String, Integer> call(String s) {
        return new Tuple2<String, Integer>(s, 1);
      }
    });

    final JavaPairRDD<String, Integer> counts = ones.reduceByKey(new Function2<Integer, Integer, Integer>() {
      public Integer call(Integer i1, Integer i2) {
        return i1 + i2;
      }
    });


    List<Tuple2<String, Integer>> output = counts.collect();

    for (Tuple2<?,?> tuple : output) {
        System.out.println(tuple._1() + ": " + tuple._2());
      }    

    ctx.stop();

    JavaStreamingContext ssc = new JavaStreamingContext(sparkConf, Durations.seconds(1));
    ssc.checkpoint("checkpointWordCount");


    JavaReceiverInputDStream<String> lines = ssc.socketTextStream(
            "localhost", 9999, StorageLevels.MEMORY_AND_DISK_SER);
    JavaDStream<String> words = lines.flatMap(new FlatMapFunction<String, String>() {
      public Iterable<String> call(String x) {
        return Lists.newArrayList(SPACE.split(x));
      }
    });


    Function2<List<Integer>, Optional<Integer>, Optional<Integer>> updateFunction =
              new Function2<List<Integer>, Optional<Integer>, Optional<Integer>>() {
                public Optional<Integer> call(List<Integer> values, Optional<Integer> state) {
                  return Optional.of(values.size()+state.or(0));
                }
              };
    //I would like the wordCounts to already contain the results of previous batch contained in counts.
    JavaPairDStream<String, Integer> wordCounts = words.mapToPair(
      new PairFunction<String, String, Integer>() {
        public Tuple2<String, Integer> call(String s) {
          return new Tuple2<String, Integer>(s, 1);
        }
      }).updateStateByKey(updateFunction);

    wordCounts.print();
    ssc.start();
    ssc.awaitTermination();
  }

}

1

There are 1 answers

0
urug On

how about you convert the results from wordcount JavaRDD into a Java Map(you could use countByValue()) and each time you encounter a word from streaming window just update/add to the map?