来自套接字的Spark Streaming不适用于reduce操作

时间:2013-06-28 10:08:26

标签: java apache-spark streaming bigdata

我正在尝试在本地计算机上运行一个简单的Spark-Streaming示例 我有一个线程将As / Bs / Cs写入套接字:

serverSocket = new ServerSocket(Constants.PORT);
s1 = serverSocket.accept();
while(true) {
    Thread.sleep(random.nextInt(100));
    String character = alphabet.get(random.nextInt(alphabet.size())) ;
    PrintWriter out = new PrintWriter(s1.getOutputStream());
    out.println(character);
    out.flush();
}

我的主程序,我尝试计算As / Bs / Cs的数量如下(没有reduce步骤):

public static void main(String[] args) {
    // start socket writer thread
    System.setProperty("spark.cleaner.ttl", "10000");
    JavaSparkContext sc = new JavaSparkContext(
            "local", 
            "Test",
            Constants.SPARK_HOME, 
            new String[]{"target/spark-standalone-0.0.1-SNAPSHOT.jar"});
    Duration batchDuration = new Duration(TIME_WINDOW_MS);
    JavaStreamingContext streamingContext = new JavaStreamingContext(sc, batchDuration);
    JavaDStream<String> stream = streamingContext.socketTextStream("localhost", Constants.PORT);
    stream.print();
    JavaPairDStream<String, Long> texts = stream.map(new PairFunction<String, String, Long>() {

            @Override
            public Tuple2<String, Long> call(String t) throws Exception {
                return new Tuple2<String, Long>("batchCount" + t, 1l);
            }

        });
     texts.print();
     streamingContext.checkpoint("checkPointDir");
     streamingContext.start();

在这种情况下,一切正常(批量的样本输出):

Time: 1372413296000 ms
-------------------------------------------
B
A
B
C
C
C
A
B
C
C
...

-------------------------------------------
Time: 1372413296000 ms
-------------------------------------------
(batchCountB,1)
(batchCountA,1)
(batchCountB,1)
(batchCountC,1)
(batchCountC,1)
(batchCountC,1)
(batchCountA,1)
(batchCountB,1)
(batchCountC,1)
(batchCountC,1)
...

但如果我在地图之后添加减少步骤则不再有效。此代码位于texts.print()

之后
JavaPairDStream<String, Long> reduced = texts.reduceByKeyAndWindow(new Function2<Long, Long, Long>() {

    @Override
    public Long call(Long t1, Long t2) throws Exception {
        return t1 + t2;
    }
    }, new Duration(TIME_WINDOW_MS));
reduced.print();

在这种情况下,我只得到第一个“stream”变量和“texts”变量的输出,而没有用于reduce的输出。第一批处理后也没有任何反应。我还将火花日志级别设置为DEBUG,但没有遇到任何异常或其他奇怪的事情。

这里发生了什么?为什么我会被锁定?

1 个答案:

答案 0 :(得分:2)

仅供记录:我在Spark用户组中得到了答案 错误是必须使用

"local[2]"

而不是

"local"

作为实例化Spark上下文的参数,以便启用并发处理。