spark出现task不能序列化错误的解决方法

应用场景：使用JavaHiveContext执行SQL之后，希望能得到其字段名及相应的值，但却出现"Caused by: java.io.NotSerializableException: org.apache.spark.sql.api.java.StructField"的错误，代码如下：

JavaSparkContext sc = new JavaSparkContext(conf);
JavaHiveContext sqlContext = new JavaHiveContext(sc);
JavaSchemaRDD schema = sqlContext.sql("select * from default.dual");
final StructField[] fields = schema.schema().getFields();
JavaRDD<String> result = schema.map(new Function<Row, String>() {
	private static final long serialVersionUID = 1L;
	@Override
	public String call(Row row) throws Exception {
		StringBuffer out = new StringBuffer();
		for (int i = 0; i < row.length(); i++) {
			out.append(fields[i].getName() + "->" + row.get(i) + ";");
		}
		return out.toString();
	}
});
System.out.println(result.collect());

在spark官网上查找序列化方面的内容，看到可以通过注册的方式自定义类的序列化方式，故在conf上添加以下设置：

conf.registerKryoClasses(new Class[] { org.apache.spark.sql.api.java.StructField.class });

测试执行后，还是报相同的错误：java.io.NotSerializableException: org.apache.spark.sql.api.java.StructField，不知道为什么，如果有朋友知道，可在下面留言。

上述方法测不通后，又再网上寻求方法，此时看到了下面的这篇文章，内容摘录见下：http://www.cnblogs.com/zwCHAN/p/4305156.html

按照第一种方法，将依赖的变量StructField[]放到map内部定义，代码见下：

JavaSparkContext sc = new JavaSparkContext(conf);
JavaHiveContext sqlContext = new JavaHiveContext(sc);
JavaSchemaRDD schema = sqlContext.sql("select * from default.dual");
JavaRDD<String> result = schema.map(new Function<Row, String>() {
	private static final long serialVersionUID = 1L;
	@Override
	public String call(Row row) throws Exception {
                StructField[] fields = schema.schema().getFields();
		StringBuffer out = new StringBuffer();
		for (int i = 0; i < row.length(); i++) {
			out.append(fields[i].getName() + "->" + row.get(i) + ";");
		}
		return out.toString();
	}
});
System.out.println(result.collect());

测试通过，但考虑到每次map都需要从JavaSchemaRDD中获取一次schema信息，比较耗时，而在map中有只需要String类型的字段名就可以了，故在原有基础上对代码进行优化，见下：

JavaSparkContext sc = new JavaSparkContext(conf);
JavaHiveContext sqlContext = new JavaHiveContext(sc);
JavaSchemaRDD schema = sqlContext.sql("select * from default.dual");
StructField[] fields = schema.schema().getFields();
final String[] fieldsName = new String[fields.length];
for (int i = 0; i < fields.length; i++) {
	fieldsName[i] = fields[i].getName();
}
JavaRDD<String> result = schema.map(new Function<Row, String>() {
	private static final long serialVersionUID = 1L;
	@Override
	public String call(Row row) throws Exception {
		StringBuffer out = new StringBuffer();
		for (int i = 0; i < row.length(); i++) {
			out.append(fieldsName[i] + "->" + row.get(i) + ";");
		}
		return out.toString();
	}
});
System.out.println(result.collect());

以下内容摘录自：http://www.cnblogs.com/zwCHAN/p/4305156.html

出现“org.apache.spark.SparkException: Task not serializable"这个错误，一般是因为在map、filter等的参数使用了外部的变量，但是这个变量不能序列化。特别是当引用了某个类（经常是当前类）的成员函数或变量时，会导致这个类的所有成员（整个类）都需要支持序列化。解决这个问题最常用的方法有：

1. 如果可以，将依赖的变量放到map、filter等的参数内部定义。这样就可以使用不支持序列化的类；
2. 如果可以，将依赖的变量独立放到一个小的class中，让这个class支持序列化；这样做可以减少网络传输量，提高效率；
3. 如果可以，将被依赖的类中不能序列化的部分使用transient关键字修饰，告诉编译器它不需要序列化。
4. 将引用的类做成可序列化的。
5. 以下这两个没试过。。

• Make the NotSerializable object as a static and create it once per machine.

• Call rdd.forEachPartition and create the NotSerializable object in there like this:

==================

ref[1]：<http://databricks.gitbooks.io/databricks-spark-knowledge-base/content/troubleshooting/javaionotserializableexception.html>

If you see this error:

org.apache.spark.SparkException: Job aborted due to stage failure: Task not serializable: java.io.NotSerializableException: ...

The above error can be triggered when you intialize a variable on the driver (master), but then try to use it on one of the workers. In that case, Spark Streaming will try to serialize the object to send it over to the worker, and fail if the object is not serializable. Consider the following code snippet:

NotSerializable notSerializable = new NotSerializable();
JavaRDD<String> rdd = sc.textFile("/tmp/myfile");

rdd.map(s -> notSerializable.doSomething(s)).collect();

This will trigger that error. Here are some ideas to fix this error:

• Serializable the class
• Declare the instance only within the lambda function passed in map.
• Make the NotSerializable object as a static and create it once per machine.
• Call rdd.forEachPartition and create the NotSerializable object in there like this:

rdd.forEachPartition(iter -> {
  NotSerializable notSerializable = new NotSerializable();

  // ...Now process iter
});

Pasted from: <http://databricks.gitbooks.io/databricks-spark-knowledge-base/content/troubleshooting/javaionotserializableexception.html>

另外， stackoverflow上http://stackoverflow.com/questions/25914057/task-not-serializable-exception-while-running-apache-spark-job 这个答的也很简明易懂。