hadoop - 如何在我的Spark Streaming应用程序中移动文件

Question

在这里，我从流目录中流传输数据并将其写入输出位置。我也在尝试实现将hdfs文件从输入文件夹移动到流目录的过程。此操作在流上下文开始之前发生一次。但是我希望此 Action 在每批Dstream中每次都执行。那有可能吗？

    val streamed_rdd = ssc.fileStream[LongWritable, Text, TextInputFormat](streaming_directory, (t:Path)=> true , true).map { case (x, y) => (y.toString) }
    streamed_rdd.foreachRDD( rdd => {
      rdd.map(x =>x.split("\t")).map(x => x(3)).foreachPartition { partitionOfRecords =>
        val connection: Connection = connectionFactory.createConnection()
        connection.setClientID("Email_send_module_client_id")
        println("connection started with active mq")
        val session: Session = connection.createSession(false, Session.AUTO_ACKNOWLEDGE)
        println("created session")
        val dest = session.createQueue("dwEmailsQueue2")
        println("destination queue name = dwEmailsQueue2")
        val prod_queue = session.createProducer(dest)
        connection.start()
        partitionOfRecords.foreach { record =>
          val rec_to_send: TextMessage = session.createTextMessage(record)
          println("started creating a text message")
          prod_queue.send(rec_to_send)
          println("sent the record")
        }
        connection.close()
      }
    }
    )
    **val LIST = scala.collection.mutable.MutableList[String]()
    val files_to_move = scala.collection.mutable.MutableList[String]()
    val cmd = "hdfs dfs -ls -d "+load_directory+"/*"
    println(cmd)
    val system_time = System.currentTimeMillis
    println(system_time)
    val output = cmd.!!
    output.split("\n").foreach(x => x.split(" ").foreach(x => if (x.startsWith("/user/hdpprod/")) LIST += x))
    LIST.foreach(x => if (x.toString.split("/").last.split("_").last.toLong < system_time) files_to_move += x)
    println("files to move" +files_to_move)
    var mv_cmd :String = "hdfs dfs -mv "
    for (file <- files_to_move){
      mv_cmd += file+" "
    }
    mv_cmd += streaming_directory
    println(mv_cmd)
    val mv_output = mv_cmd.!!
    println("moved the data to the folder")**
    if (streamed_rdd.count().toString == "0") {
      println("no data in the streamed list")
    } else {
      println("saving the Dstream at "+System.currentTimeMillis())
      streamed_rdd.transform(rdd => {rdd.map(x => (check_time_to_send+"\t"+check_time_to_send_utc+"\t"+x))}).saveAsTextFiles("/user/hdpprod/temp/spark_streaming_output_sent/sent")
    }
    ssc.start()
    ssc.awaitTermination()
  }
}

Answer 1

我尝试在Java实现中做如下相同的事情。您可以从rdd上的foreachPartion调用此方法

 public static void moveFiles(final String moveFilePath,
            final JavaRDD rdd) {
            for (final Partition partition : rdd.partitions()) {
                final UnionPartition unionPartition = (UnionPartition) partition;
                final NewHadoopPartition newHadoopPartition = (NewHadoopPartition)
                    unionPartition.parentPartition();
                final String fPath = newHadoopPartition.serializableHadoopSplit()
                    .value().toString();
                final String[] filespaths = fPath.split(":");

                if ((filespaths != null) && (filespaths.length > 0)) {
                    for (final String filepath : filespaths) {
                        if ((filepath != null) && filepath.contains("/")) {
                            final File file = new File(filepath);

                            if (file.exists() && file.isFile()) {
                                try {
                                    File destFile = new File(moveFilePath + "/" +
                                            file.getName());

                                    if (destFile.exists()) {
                                        destFile = new File(moveFilePath + "/" +
                                                file.getName() + "_");
                                    }

                                    java.nio.file.Files.move((file
                                            .toPath()), destFile.toPath(),
                                        StandardCopyOption.REPLACE_EXISTING);


                                } catch (Exception e) {
                                    logger.error(
                                        "Exception while moving file",
                                        e);
                                }
                            }
                        }
                    }
                }
            }

        }