最近在項(xiàng)目中遇到二次排序的需求，和平常開發(fā)spark的application一樣，開始查看API，編碼，調(diào)試，驗(yàn)證結(jié)果。由于之前對(duì)spark的API使用過，知道API中的sortByKey()可以自定義排序規(guī)則，通過實(shí)現(xiàn)自定義的排序規(guī)則來實(shí)現(xiàn)二次排序。
這里為了說明問題，舉了一個(gè)簡(jiǎn)單的例子,key是由兩部分組成的，我們這里按key的第一部分的降序排，key的第二部分升序排，具體如下：

JavaSparkContext javaSparkContext = new JavaSparkContext(sparkConf);

List<Integer> data = Arrays.asList(5, 1, 1, 4, 4, 2, 2);

JavaRDD<Integer> javaRDD = javaSparkContext.parallelize(data);

final Random random = new Random(100);

JavaPairRDD javaPairRDD = javaRDD.mapToPair(new PairFunction<Integer, String, Integer>() {    
        @Override    
        public Tuple2<String, Integer> call(Integer integer) throws Exception {        
          return new Tuple2<String, Integer>(Integer.toString(integer) + " " + random.nextInt(10),random.nextInt(10));   
     }
});

JavaPairRDD<String,Integer> sortByKeyRDD = javaPairRDD.sortByKey(new Comparator<String>() {    
    @Override    
    public int compare(String o1, String o2) {        
        String []o1s = o1.split(" ");        
        String []o2s = o2.split(" ");       
        if(o1s[0].compareTo(o2s[0]) == 0)            
              return o1s[1].compareTo(o2s[1]);        
        else            
              return -o1s[0].compareTo(o2s[0]);    
  }
});
System.out.println("~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~" + sortByKeyRDD.collect());

上面編碼從語法上沒有什么問題，可是運(yùn)行下報(bào)了如下錯(cuò)誤：

java.lang.reflect.InvocationTargetException
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:606)
    at org.apache.spark.serializer.SerializationDebugger$ObjectStreamClassMethods$.getObjFieldValues$extension(SerializationDebugger.scala:248)
    at org.apache.spark.serializer.SerializationDebugger$SerializationDebugger.visitSerializable(SerializationDebugger.scala:158)
    at org.apache.spark.serializer.SerializationDebugger$SerializationDebugger.visit(SerializationDebugger.scala:107)
    at org.apache.spark.serializer.SerializationDebugger$SerializationDebugger.visitSerializable(SerializationDebugger.scala:166)
    at org.apache.spark.serializer.SerializationDebugger$SerializationDebugger.visit(SerializationDebugger.scala:107)
    at org.apache.spark.serializer.SerializationDebugger$SerializationDebugger.visitSerializable(SerializationDebugger.scala:166)
    at org.apache.spark.serializer.SerializationDebugger$SerializationDebugger.visit(SerializationDebugger.scala:107)
    at org.apache.spark.serializer.SerializationDebugger$.find(SerializationDebugger.scala:66)
    at org.apache.spark.serializer.SerializationDebugger$.improveException(SerializationDebugger.scala:41)
    at org.apache.spark.serializer.JavaSerializationStream.writeObject(JavaSerializer.scala:47)
    at org.apache.spark.serializer.JavaSerializerInstance.serialize(JavaSerializer.scala:81)
    at org.apache.spark.util.ClosureCleaner$.ensureSerializable(ClosureCleaner.scala:312)
    at org.apache.spark.util.ClosureCleaner$.org$apache$spark$util$ClosureCleaner$$clean(ClosureCleaner.scala:305)
    at org.apache.spark.util.ClosureCleaner$.clean(ClosureCleaner.scala:132)
    at org.apache.spark.SparkContext.clean(SparkContext.scala:1891)
    at org.apache.spark.SparkContext.runJob(SparkContext.scala:1764)
    at org.apache.spark.SparkContext.runJob(SparkContext.scala:1779)
    at org.apache.spark.rdd.RDD$$anonfun$collect$1.apply(RDD.scala:885)
    at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:148)
    at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:109)
    at org.apache.spark.rdd.RDD.withScope(RDD.scala:286)
    at org.apache.spark.rdd.RDD.collect(RDD.scala:884)
    at org.apache.spark.api.java.JavaRDDLike$class.collect(JavaRDDLike.scala:335)
    at org.apache.spark.api.java.AbstractJavaRDDLike.collect(JavaRDDLike.scala:47)

因此，我再次去查看相應(yīng)的spark Java API文檔，但是我沒有發(fā)現(xiàn)任何指明錯(cuò)誤的地方。好吧，那只能扒下源碼吧，在javaPairRDD中

def sortByKey(comp: Comparator[K], ascending: Boolean): JavaPairRDD[K, V] = {  
    implicit val ordering = comp // Allow implicit conversion of Comparator to Ordering.  
    fromRDD(new OrderedRDDFunctions[K, V, (K, V)](rdd).sortByKey(ascending))
}

其實(shí)在OrderedRDDFunctions類中有個(gè)變量ordering它是隱形的：private val ordering = implicitly[Ordering[K]]。他就是默認(rèn)的排序規(guī)則，我們自己重寫的comp就修改了默認(rèn)的排序規(guī)則。到這里還是沒有發(fā)現(xiàn)問題，但是發(fā)現(xiàn)類OrderedRDDFunctions extends Logging with Serializable，又回到上面的報(bào)錯(cuò)信息，掃描到“serializable”！！！因此，返回上述代碼，查看Comparator interface實(shí)現(xiàn)，發(fā)現(xiàn)原來是它沒有extend Serializable，故只需創(chuàng)建一個(gè) serializable的comparator就可以：public interface SerializableComparator<T> extends Comparator<T>, Serializable { }
具體如下：

private static class Comp implements Comparator<String>,Serializable{    
    @Override    
    public int compare(String o1, String o2) {            
          String []o1s = o1.split(" ");            
          String []o2s = o2.split(" ");            
          if(o1s[0].compareTo(o2s[0]) == 0)                
              return o1s[1].compareTo(o2s[1]);
           else
                return -o1s[0].compareTo(o2s[0]);    
  }
}
JavaPairRDD<String,Integer> sortByKeyRDD = javaPairRDD.sortByKey(new Comp());

總結(jié)下，在spark的Java API中，如果需要使用Comparator接口，須注意是否需要序列化，如sortByKey(),repartitionAndSortWithinPartitions()等都是需要序列化的。