本文介绍如何通过Spark Streaming消费消息服务MNS（Message Notification Service）中的数据，并统计每个Batch内的单词个数。

Spark接入MNS

示例代码如下。 
val conf = new SparkConf().setAppName("Test MNS Streaming")
    val batchInterval = Seconds(10)
    val ssc = new StreamingContext(conf, batchInterval)
    val queuename = "queuename"
    val accessKeyId = "<accessKeyId>"
    val accessKeySecret = "<accessKeySecret>"
    val endpoint = "http://xxx.yyy.zzzz/abc"
    val mnsStream = MnsUtils.createPullingStreamAsRawBytes(ssc, queuename, accessKeyId, accessKeySecret, endpoint,
      StorageLevel.MEMORY_ONLY)
    mnsStream.foreachRDD( rdd => {
      rdd.map(bytes => new String(bytes)).flatMap(line => line.split(" "))
        .map(word => (word, 1))
        .reduceByKey(_ + _).collect().foreach(e => println(s"word: ${e._1}, cnt: ${e._2}"))
    })
    ssc.start()
    ssc.awaitTermination()

支持MetaService

上面的示例是显式地将AccessKey传入到MNS中。从E-MapReduce SDK 1.3.2版本开始，Spark Streaming可以基于MetaService实现免AccessKey处理MNS数据。具体可以参见E-MapReduce SDK中的MnsUtils类说明。 
MnsUtils.createPullingStreamAsBytes(ssc, queueName, endpoint, storageLevel)
MnsUtils.createPullingStreamAsRawBytes(ssc, queueName, endpoint, storageLevel)

附录

完整示例代码，请参见Spark接入MNS。