java - 如何在spark RDD(JavaRDD)中获取记录的文件名

Question

我正在使用 JavaRDD 加载多个文件

JavaRDD<String> allLines = sc.textFile(hdfs://path/*.csv);

加载文件后，我修改了每条记录并想保存它们。但是，我还需要将原始文件名 (ID) 与记录一起保存，以备将来引用。无论如何，我可以从 RDD 中的单个记录中获取原始文件名吗？谢谢

Answer 1

您可以尝试执行以下代码段中的操作:

JavaPairRDD<LongWritable, Text> javaPairRDD = sc.newAPIHadoopFile(
    "hdfs://path/*.csv", 
    TextInputFormat.class, 
    LongWritable.class, 
    Text.class, 
    new Configuration()
);
JavaNewHadoopRDD<LongWritable, Text> hadoopRDD = (JavaNewHadoopRDD) javaPairRDD;

JavaRDD<Tuple2<String, String>> namedLinesRDD = hadoopRDD.mapPartitionsWithInputSplit((inputSplit, lines) -> {
    FileSplit fileSplit = (FileSplit) inputSplit;
    String fileName = fileSplit.getPath().getName();

    Stream<Tuple2<String, String>> stream =
        StreamSupport.stream(Spliterators.spliteratorUnknownSize(lines, Spliterator.ORDERED), false)
            .map(line -> {
                String lineText = line._2().toString();
                // emit file name as key and line as a value
                return new Tuple2(fileName, lineText);
            });
    return stream.iterator();
}, true);

更新(针对 java7)

JavaRDD<Tuple2<String, String>> namedLinesRDD = hadoopRDD.mapPartitionsWithInputSplit(
    new Function2<InputSplit, Iterator<Tuple2<LongWritable, Text>>, Iterator<Tuple2<String, String>>>() {
        @Override
        public Iterator<Tuple2<String, String>> call(InputSplit inputSplit, final Iterator<Tuple2<LongWritable, Text>> lines) throws Exception {
            FileSplit fileSplit = (FileSplit) inputSplit;
            final String fileName = fileSplit.getPath().getName();
            return new Iterator<Tuple2<String, String>>() {
                @Override
                public boolean hasNext() {
                    return lines.hasNext();
                }
                @Override
                public Tuple2<String, String> next() {
                    Tuple2<LongWritable, Text> entry = lines.next();
                    return new Tuple2<String, String>(fileName, entry._2().toString());
                }
            };
        }
    }, 
    true
);