Question

我正在尝试按值对RDD进行排序，如果多个值相等，那么我需要按字典顺序按键这些值。

代码：

JavaPairRDD <String,Long> rddToSort = rddMovieReviewReducedByKey.mapToPair(new PairFunction < Tuple2 < String, MovieReview > , String, Long > () {

    @Override
    public Tuple2 < String, Long > call(Tuple2 < String, MovieReview > t) throws Exception {
        return new Tuple2 < String, Long > (t._1, t._2.count);
    }
});

我到目前为止所做的是使用takeOrdered并提供CustomComperator，但由于takeOrdered无法处理大量数据，因此在运行代码时退出（它占用了操作系统无法处理的大量内存）：

List < Tuple2 < String, Long >> rddSorted = rddMovieReviewReducedByKey.mapToPair(new PairFunction < Tuple2 < String, MovieReview > , String, Long > () {

    @Override
    public Tuple2 < String, Long > call(Tuple2 < String, MovieReview > t) throws Exception {
        return new Tuple2 < String, Long > (t._1, t._2.count);
    }
}).takeOrdered(newTopMovies, MapLongValueComparator.VALUE_COMP);

Comperator：

    static class MapLongValueComparator implements Comparator < Tuple2 < String, Long >> , Serializable {
        private static final long serialVersionUID = 1L;

        private static final MapLongValueComparator VALUE_COMP = new MapLongValueComparator();

        @Override
        public int compare(Tuple2 < String, Long > o1, Tuple2 < String, Long > o2) {
            if (o1._2.compareTo(o2._2) == 0) {
                return o1._1.compareTo(o2._1);
            }
            return -o1._2.compareTo(o2._2);
        }
}

ERROR：

16/06/30 21:09:23 INFO scheduler.DAGScheduler: Job 18 failed: takeOrdered at MovieAnalyzer.java:708, took 418.149182 s

你会如何排序这个RDD？您如何考虑TopKMovies考虑值，并且在按字典顺序排列密钥的情况下。

感谢。

Answer 1

使用sortByKey和比较器＆amp;解决了这个问题。将<String, Long> PairRDD与< Tuple2<String,Long> , Long> PairRDD

匹配后的分区

JavaPairRDD <Tuple2<String,Long>, Long> sortedRdd = rddMovieReviewReducedByKey.mapToPair(new PairFunction < Tuple2 < String, MovieReview > , Tuple2<String,Long>, Long > () {

    @Override
    public Tuple2 < Tuple2<String,Long>, Long > call(Tuple2 < String, MovieReview > t) throws Exception {
        return new Tuple2 < Tuple2<String,Long>, Long > (new Tuple2<String,Long>(t._1,t._2.count), t._2.count);
    }
}).sortByKey(new TupleMapLongComparator(), true, 100);


JavaPairRDD <String,Long> sortedRddToPairs = sortedRdd.mapToPair(new PairFunction<Tuple2<Tuple2<String,Long>,Long>, String, Long>() {

    @Override
    public Tuple2<String, Long> call(
            Tuple2<Tuple2<String, Long>, Long> t) throws Exception {
        return new Tuple2 < String, Long > (t._1._1, t._1._2);
    }

});

比较

private class TupleMapLongComparator implements Comparator<Tuple2<String,Long>>, Serializable {
    @Override
    public int compare(Tuple2<String,Long> tuple1, Tuple2<String,Long> tuple2) {

        if (tuple1._2.compareTo(tuple2._2) == 0) {
            return tuple1._1.compareTo(tuple2._1);
        }
        return -tuple1._2.compareTo(tuple2._2);
    }
}

Answer 2

你在Spark中尝试过二次排序吗？

Spark Secondary Sort

首先按值排序JavaPairRDD然后按键排序

2 个答案: