Spark 2.0:从RDD迁移到Dataset

Question

我想调整我的Java Spark应用程序(实际上使用RDD进行某些计算)来Datasets代替RDDs.我是数据集的新手,不知道如何将哪个事务映射到相应的数据集操作.

目前我将它们映射为:

JavaSparkContext.textFile(...)                       -> SQLContext.read().textFile(...)
JavaRDD.filter(Function)                             -> Dataset.filter(FilterFunction)
JavaRDD.map(Function)                                -> Dataset.map(MapFunction)
JavaRDD.mapToPair(PairFunction)                      -> Dataset.groupByKey(MapFunction) ???
JavaPairRDD.aggregateByKey(U, Function2, Function2)  -> KeyValueGroupedDataset.???

相应的问题是:

• 等于JavaRDD.mapToPair该Dataset.groupByKey方法？
• 是否JavaPairRDD映射到KeyValueGroupedDataset？
• 哪种方法等于JavaPairRDD.aggregateByKey方法？

但是,我想将以下RDD代码移植到数据集中:

JavaRDD<Article> goodRdd = ...

JavaPairRDD<String, Article> ArticlePairRdd = goodRdd.mapToPair(new PairFunction<Article, String, Article>() {              // Build PairRDD<<Date|Store|Transaction><Article>>
    public Tuple2<String, Article> call(Article article) throws Exception {
        String key = article.getKeyDate() + "|" + article.getKeyStore() + "|" + article.getKeyTransaction() + "|" + article.getCounter();
        return new Tuple2<String, Article>(key, article);
    }
});

JavaPairRDD<String, String> transactionRdd = ArticlePairRdd.aggregateByKey("",                                              // Aggregate distributed data -> PairRDD<String, String>
    new Function2<String, Article, String>() {
        public String call(String oldString, Article newArticle) throws Exception {
            String articleString = newArticle.getOwg() + "_" + newArticle.getTextOwg();                                     // <<Date|Store|Transaction><owg_textOwg###owg_textOwg>>
            return oldString + "###" + articleString;
        }
    }, 
    new Function2<String, String, String>() {
        public String call(String a, String b) throws Exception {
            String c = a.concat(b);
            ...
            return c;
        }
    }
);

我的代码看起来还是这样:

Dataset<Article> goodDS = ...

KeyValueGroupedDataset<String, Article> ArticlePairDS = goodDS.groupByKey(new MapFunction<Article, String>() {
    public String call(Article article) throws Exception {
        String key = article.getKeyDate() + "|" + article.getKeyStore() + "|" + article.getKeyTransaction() + "|" + article.getCounter();
        return key;
    }
}, Encoders.STRING());

// here I need something similar to aggregateByKey! Not reduceByKey as I need to return another data type (String) than I have before (Article)