用Spark往Kafka里面写对象设计与实现

Spark和 Kafka都是比较常用的两个大数据框架， Spark里面提供了对 Kafka读写的支持。默认情况下我们 Kafka只能写Byte数组到Topic里面，如果我们想往Topic里面读写String类型的消息，可以分别使用Kafka里面内置的StringEncoder编码类和StringDecoder解码类。那如果我们想往Kafka里面写对象怎么办？
　　别担心，Kafka中的kafka.serializer里面有Decoder和Encoder两个trait，这两个trait就是Kafka Topic消息相关的解码类和编码类，内置的StringDecoder和StringEncoder类分别都是继承那两个trait的。直接将String对象用给定的编码转换成Byte数组。来看下Decoder和Encoder两个trait的实现：

 
   01/** 
 
   02 * A decoder is a method of turning byte arrays into objects. 
 
   03 * An implementation is required to provide a constructor that 
 
   04 * takes a VerifiableProperties instance. 
 
   05 */
 
   06trait  Decoder[T] { 
 
   07  def fromBytes(bytes: Array[Byte]): T 
 
   08} 
 
   09  
 
   10/** 
 
   11 * An encoder is a method of turning objects into byte arrays. 
 
   12 * An implementation is required to provide a constructor that 
 
   13 * takes a VerifiableProperties instance. 
 
   14 */
 
   15trait  Encoder[T] { 
 
   16  def toBytes(t:  T): Array[Byte] 
 
   17}

　　也就是说，我们自定义的编码和解码类只需要分别实现toBytes和fromBytes函数即可。那我们如何将对象转换成Byte数组，并且如何将Byte数组转换回对象呢？记得Java中写对象的类没？我们可以用ByteArrayOutputStream并结合ObjectOutputStream类将对象转换成Byte数组；并用ByteArrayInputStream结合ObjectInputStream类将Byte数组转换回对象。这不就实现了吗？？废话不多说，来看看怎么实现：

 
   01class  IteblogDecoder[T](props:  VerifiableProperties =  null) extends Decoder[T] { 
 
   02  
 
   03    def fromBytes(bytes: Array[Byte]): T =  { 
 
   04      var t:  T = null.asInstanceOf[T] 
 
   05      var bi:  ByteArrayInputStream =  null
 
   06      var oi:  ObjectInputStream =  null
 
   07      try { 
 
   08        bi = new ByteArrayInputStream(bytes) 
 
   09        oi = new ObjectInputStream(bi) 
 
   10        t = oi.readObject().asInstanceOf[T] 
 
   11      } 
 
   12      catch { 
 
   13        case e:  Exception => { 
 
   14          e.printStackTrace(); null
 
   15        } 
 
   16      } finally {  
 
   17        bi.close() 
 
   18        oi.close() 
 
   19      } 
 
   20      t 
 
   21    }  
 
   22} 
 
   23  
 
   24class  IteblogEncoder[T](props:  VerifiableProperties =  null) extends Encoder[T] { 
 
   25  
 
   26    override def toBytes(t: T):  Array[Byte] =  { 
 
   27      if (t ==  null) 
 
   28        null
 
   29      else { 
 
   30        var bo:  ByteArrayOutputStream =  null
 
   31        var oo:  ObjectOutputStream =  null
 
   32        var byte:  Array[Byte] =  null
 
   33        try { 
 
   34          bo = new ByteArrayOutputStream() 
 
   35          oo = new ObjectOutputStream(bo) 
 
   36          oo.writeObject(t) 
 
   37          byte = bo.toByteArray 
 
   38        } catch {  
 
   39          case ex:  Exception =>  return byte 
 
   40        } finally {  
 
   41          bo.close() 
 
   42          oo.close() 
 
   43        } 
 
   44        byte 
 
   45      } 
 
   46    }  
 
   47}

　　这样我们就定义了自己的编码和解码器。那如何使用呢？？假设我们有一个Person类。如下：

`1`	`case` `class` `Person(var` `name:` `String,` `var` `age:` `Int)`

　　我们可以在发送数据这么使用：

 
   01def  getProducerConfig(brokerAddr:  String): Properties = { 
 
   02    val props =  new Properties() 
 
   03    props.put("metadata.broker.list", brokerAddr) 
 
   04    props.put("serializer.class", classOf[IteblogEncoder[Person]].getName) 
 
   05    props.put("key.serializer.class", classOf[StringEncoder].getName) 
 
   06    props 
 
   07  }  
 
   08  
 
   09def  sendMessages(topic:  String, messages:  List[Person], brokerAddr:  String) { 
 
   10    val producer =  new Producer[String, Person]( 
 
   11　　　　new ProducerConfig(getProducerConfig(brokerAddr))) 
 
   12    producer.send(messages.map { 
 
   13      new KeyedMessage[String, Person](topic, "Iteblog", _) 
 
   14    }: _*) 
 
   15    producer.close() 
 
   16} 
 
   17  
 
   18def  main(args: Array[String]) { 
 
   19    val sparkConf =  new S parkConf().setAppName(this.getClass.getSimpleName) 
 
   20    val ssc =  new StreamingContext(sparkConf, Milliseconds(500)) 
 
   21    val topic =  args(0)  
 
   22    val brokerAddr =  "http://www.iteblog.com:9092"
 
   23  
 
   24    val data =  List(Person("wyp", 23), Person("spark", 34), Person("kafka", 23), 
 
   25         Person("iteblog", 23)) 
 
   26    sendMessages(topic, data, brokerAddr) 
 
   27}

　　在接收端可以这么使用

 
   01val  sparkConf =  new S parkConf().setAppName(this.getClass.getSimpleName) 
 
   02val  ssc = new StreamingContext(sparkConf, Milliseconds(500)) 
 
   03val  (topic, groupId) =  (args(0), args(1)) 
 
   04  
 
   05val  kafkaParams =  Map("zookeeper.connect"  -> "http://www.iteblog.com:2181", 
 
   06      "group.id" -> groupId, 
 
   07      "auto.offset.reset" -> "smallest") 
 
   08  
 
   09val  stream = KafkaUtils.createStream[String, Person, StringDecoder, 
 
   10　　 IteblogDecoder[Person]](ssc, kafkaParams, Map(topic -> 1),  
 
   11　　StorageLevel.MEMORY_ONLY) 
 
   12    stream.foreachRDD(rdd => { 
 
   13      if (rdd.count() != 0) { 
 
   14        rdd.foreach(item => if (item !=  null) println(item)) 
 
   15      } else {  
 
   16        println("Empty rdd!!") 
 
   17      } 
 
   18    })  
 
   19    ssc.start() 
 
   20    ssc.awaitTermination() 
 
   21    ssc.stop()

　　这样可以发送任意可序列化的对象了。下面是效果：

 
   1Empty rdd!! 
 
   2(Iteblog,Person(wyp,23)) 
 
   3(Iteblog,Person(spark,34)) 
 
   4(Iteblog,Person(kafka,23)) 
 
   5(Iteblog,Person(iteblog,23)) 
 
   6Empty rdd!! 
 
   7Empty rdd!! 
 
   8Empty rdd!! 
 
   9Empty rdd!!