Spark Tips 1: RDD的collect action 不适用于单个element size过大的情况

collect是Spark RDD一个非常易用的action，通过collect可以轻易获得一个RDD当中所有的elements。当这些elements是String类型的时候，可以轻易将整个RDD转化成一个List<String>,简直不要太好用。

不过等一等，这么好用的action有一个弱点，它不适合size比较的element。举个例子来说吧。请看下面这段代码：

... ...

JavaPairInputDStream<String, String> messages = KafkaUtils.createDirectStream(

jssc,

String.class,

String.class,

StringDecoder.class,

StringDecoder.class,

kafkaParams,

topicsSet

);

JavaDStream<String> lines = messages.map(new Function<Tuple2<String, String>, String>() {

@Override

public String call(Tuple2<String, String> tuple2) {

return tuple2._2();

}

});

lines.foreachRDD(new Function<JavaRDD<String>, Void>(){

@Override

public Void call(JavaRDD<String> strJavaRDD) throws Exception {

List<String> messages = strJavaRDD.collect();

List<String> sizeStrs = new ArrayList<String>();

for (String message: messages) {

if (message== null)

continue;

String logStr = "message size is " + message.length();

strs.add(logStr);

}

saveToLog(outputLogPath, strs);

return null;

}

});

... ...

上述这段代码当Kafka中单个message（也就是）的size很小（比如200Bytes）的时候，运行得很好。可是当单个message size变大到一定程度（例如10MB），就会抛出以下异常：

sparkDriver-akka.actor.default-dispatcher-18 2015-10-15 21:52:28,606 ERROR JobSc

heduler - Error running job streaming job 1444971120000 ms.0

org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 238.0 failed 4 times, most recent failure: Lost task 0.3 in stage 238.0 (TID421, 127.0.0.1): ExecutorLostFailure (executor 123 lost)

Driver stacktrace:

at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1215)

at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1204)

at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1203)

at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)

at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)

at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1203)

at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:693)

at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:693)

at scala.Option.foreach(Option.scala:236)

at org.apache.spark.scheduler.DAGScheduler.handleTaskSetFailed(DAGScheduler.scala:693)

at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1404)

at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1365)

at org.apache.spark.util.EventLoop$$anon$1.run(EventLoop.scala:48)

原因很简单，collect（）无法handle“大数据”。对于10MB size这样的单条message。我们可以用下面这段代码替代上面最后一部分：

lines.foreachRDD(new Function<JavaRDD<String>, Void>() {

@Override

public Void call(JavaRDD<String> strJavaRDD) throws Exception {

JavaRDD<String> sizeRDD = strJavaRDD.map(new Function<String, String>() {

@Override

public String call(String message) throws Exception {

if (message == null)

return null;

String logStr = "Message size is " + message.length();

return logStr;

}

});

List<String> sizeStrs = sizeRDD.collect();

saveToLog(outputLogPat, sizeStrs);

return null;

}

});