window滑動窗口

Spark Streaming提供了滑動窗口操作的支持，從而讓我們可以對一個滑動窗口內(nèi)的數(shù)據(jù)執(zhí)行計算操作。每次掉落在窗口內(nèi)的RDD的數(shù)據(jù)，會被聚合起來執(zhí)行計算操作，然后生成的RDD，會作為window DStream的一個RDD。比如下圖中，就是對每三秒鐘的數(shù)據(jù)執(zhí)行一次滑動窗口計算，這3秒內(nèi)的3個RDD會被聚合起來進(jìn)行處理，然后過了兩秒鐘，又會對最近三秒內(nèi)的數(shù)據(jù)執(zhí)行滑動窗口計算。所以每個滑動窗口操作，都必須指定兩個參數(shù)，窗口長度以及滑動間隔，而且這兩個參數(shù)值都必須是batch間隔的整數(shù)倍。（Spark Streaming對滑動窗口的支持，是比Storm更加完善和強(qiáng)大的）

window滑動窗口.png

window滑動窗口操作

案例

案例：熱點搜索詞滑動統(tǒng)計，每隔10秒鐘，統(tǒng)計最近60秒鐘的搜索詞的搜索頻次，并打印出排名最靠前的3個搜索詞以及出現(xiàn)次數(shù)
Java版本

public class WindowHotWord {
    public static void main(String[] args) {
        SparkConf conf = new SparkConf().setAppName("WindowHotWordJava").setMaster("local[2]");
        JavaStreamingContext streamingContext = new JavaStreamingContext(conf, Durations.seconds(10));

        // 說明一下，這里的搜索日志的格式
        // leo hello
        // tom world
        JavaReceiverInputDStream<String> searchLogsDStream = streamingContext.socketTextStream("hadoop-100", 9999);
        // 將搜索日志給轉(zhuǎn)換成，只有一個搜索詞，即可
        JavaDStream<String> searchWordsDStream = searchLogsDStream.map(new Function<String, String>() {
            @Override
            public String call(String v1) throws Exception {
                return v1.split(" ")[1];
            }
        });

        // 將搜索詞映射為(searchWord, 1)的tuple格式
        JavaPairDStream<String, Integer> searchWordPairDStream = searchWordsDStream.mapToPair(new PairFunction<String, String, Integer>() {
            @Override
            public Tuple2<String, Integer> call(String s) throws Exception {
                return new Tuple2<>(s, 1);
            }
        });

        // 針對(searchWord, 1)的tuple格式的DStream，執(zhí)行reduceByKeyAndWindow，滑動窗口操作
        // 第二個參數(shù)，是窗口長度，這里是60秒
        // 第三個參數(shù)，是滑動間隔，這里是10秒
        // 也就是說，每隔10秒鐘，將最近60秒的數(shù)據(jù)，作為一個窗口，進(jìn)行內(nèi)部的RDD的聚合，然后統(tǒng)一對一個RDD進(jìn)行后續(xù)計算
        // 所以說，這里的意思，就是，之前的searchWordPairDStream為止，其實，都是不會立即進(jìn)行計算的
        // 而是只是放在那里
        // 然后，等待我們的滑動間隔到了以后，10秒鐘到了，會將之前60秒的RDD，因為一個batch間隔是，5秒，所以之前
        // 60秒，就有12個RDD，給聚合起來，然后，統(tǒng)一執(zhí)行redcueByKey操作
        // 所以這里的reduceByKeyAndWindow，是針對每個窗口執(zhí)行計算的，而不是針對某個DStream中的RDD
        JavaPairDStream<String, Integer> searchWordCountsDStream = searchWordPairDStream.reduceByKeyAndWindow(new Function2<Integer, Integer, Integer>() {
            @Override
            public Integer call(Integer v1, Integer v2) throws Exception {
                return v1 + v2;
            }
        }, Durations.seconds(60), Durations.seconds(10));

        JavaPairDStream<String, Integer> finalDStream = searchWordCountsDStream.transformToPair(new Function<JavaPairRDD<String, Integer>, JavaPairRDD<String, Integer>>() {
            @Override
            public JavaPairRDD<String, Integer> call(JavaPairRDD<String, Integer> v1) throws Exception {

                // 執(zhí)行搜索詞和出現(xiàn)頻率的反轉(zhuǎn)
                JavaPairRDD<Integer, String> countSearchWordsRDD = v1.mapToPair(new PairFunction<Tuple2<String, Integer>, Integer, String>() {
                    @Override
                    public Tuple2<Integer, String> call(Tuple2<String, Integer> stringIntegerTuple2) throws Exception {
                        return new Tuple2<>(stringIntegerTuple2._2, stringIntegerTuple2._1);
                    }
                });

                // 然后執(zhí)行降序排序
                JavaPairRDD<Integer, String> sortedCountSearchWordsRDD = countSearchWordsRDD.sortByKey(false);

                // 然后再次執(zhí)行反轉(zhuǎn)，變成(searchWord, count)的這種格式
                JavaPairRDD<String, Integer> sortedSearchWordCountsRDD = sortedCountSearchWordsRDD.mapToPair(new PairFunction<Tuple2<Integer, String>, String, Integer>() {
                    @Override
                    public Tuple2<String, Integer> call(Tuple2<Integer, String> integerStringTuple2) throws Exception {
                        return new Tuple2<>(integerStringTuple2._2, integerStringTuple2._1);
                    }
                });
                // 然后用take()，獲取排名前3的熱點搜索詞
                List<Tuple2<String, Integer>> hogSearchWordCounts = sortedSearchWordCountsRDD.take(3);
                for (Tuple2<String, Integer> wordCount : hogSearchWordCounts) {
                    System.out.println(wordCount._1 + ": " + wordCount._2);
                }

                return v1;
            }
        });

        // 這個無關(guān)緊要，只是為了觸發(fā)job的執(zhí)行，所以必須有output操作
        finalDStream.print();
        
        streamingContext.start();
        streamingContext.awaitTermination();
        streamingContext.close();
    }
}

Scala版本

object WindowHotWord {
  def main(args: Array[String]): Unit = {
    val conf = new SparkConf().setAppName("WindowHotWordScala").setMaster("local[2]")
    val streamingContext = new StreamingContext(conf,Seconds(10))

    val searchLogsDStream = streamingContext.socketTextStream("hadoop-100", 9999)

    val searchWordPairDStream = searchLogsDStream.map(word => (word.split(" ")(1), 1))

    val searchWordCountsDSteram = searchWordPairDStream.reduceByKeyAndWindow((v1:Int, v2:Int) => v1 + v2, Seconds(60), Seconds(10))

    val finalDStream = searchWordCountsDSteram.transform(searchWordCountsRDD => {
      val countSearchWordsRDD = searchWordCountsRDD.map(wordCount => (wordCount._2, wordCount._1))
      val sortedCountSearchWordsRDD = countSearchWordsRDD.sortByKey(false)
      val sortedSearchWordCountsRDD = sortedCountSearchWordsRDD.map(sorted => (sorted._2, sorted._1))
      val tuples = sortedSearchWordCountsRDD.take(3)
      for(tuple <- tuples) {
        println(tuple._1 + ": " + tuple._2)
      }
      searchWordCountsRDD
    })
    finalDStream.print()

    streamingContext.start()
    streamingContext.awaitTermination()
  }
}