python - Spark运行时错误: uninitialized classmethod object

Question

我用 Python 编写了一个非常简单的 Spark 代码:

import collections
Person = collections.namedtuple('Person', ['name', 'age', 'gender'])

a = sc.parallelize([['Barack Obama', 54, 'M'], ['Joe Biden', 74, 'M']])
a = a.map(lambda row: Person(*row))

print a.collect()

def func(row):
    tmp = row._replace(name='Jack Rabbit')
    return tmp

print a.map(func).collect()

我得到以下输出和错误:

[Person(name='Barack Obama', age=29, gender='M'), Person(name='Joe Biden', age=57, gender='M')]

Py4JJavaError: An error occurred while calling z:org.apache.spark.api.python.PythonRDD.collectAndServe.
: org.apache.spark.SparkException: Job aborted due to stage failure: Task 49 in stage 11.0 failed 4 times, most recent failure: Lost task 49.3 in stage 11.0 (TID 618, 172.19.75.121): org.apache.spark.api.python.PythonException: Traceback (most recent call last):
  File "/etc/spark-1.4.0-bin-cdh4/python/lib/pyspark.zip/pyspark/worker.py", line 111, in main
    process()
  File "/etc/spark-1.4.0-bin-cdh4/python/lib/pyspark.zip/pyspark/worker.py", line 106, in process
    serializer.dump_stream(func(split_index, iterator), outfile)
  File "/etc/spark-1.4.0-bin-cdh4/python/lib/pyspark.zip/pyspark/serializers.py", line 263, in dump_stream
    vs = list(itertools.islice(iterator, batch))
  File "<ipython-input-19-f0b4885784cb>", line 2, in func
  File "<string>", line 32, in _replace
RuntimeError: uninitialized classmethod object

    at org.apache.spark.api.python.PythonRDD$$anon$1.read(PythonRDD.scala:138)
    at org.apache.spark.api.python.PythonRDD$$anon$1.<init>(PythonRDD.scala:179)
    at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:97)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:277)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:244)
    at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:63)
    at org.apache.spark.scheduler.Task.run(Task.scala:70)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
    at java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:886)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:908)
    at java.lang.Thread.run(Thread.java:662)

但是，如果我运行以下代码，则不会收到任何错误:

for row in a.collect():
  func(row)

什么给出了？

Answer 1

编辑:

SPARK-10542 引入了对序列化命名元组的支持.

原始答案

为什么不起作用？因为 namedtuple 调用创建了一个类，并且 Spark 中的类不会序列化为闭包的一部分。这意味着您必须创建一个单独的模块*并确保它对工作人员可用:

txt = "\n".join(["import collections",
    "Person = collections.namedtuple('Person', ['name', 'age', 'gender'])"])

with open("persons.py", "w") as fw:
    fw.write(txt)

sc.addPyFile("persons.py")  # Ship module to the worker nodes

接下来您可以简单地导入，一切都应该按预期工作:

import persons

a.map(func).collect()

旁注中前导下划线的存在是有原因的。

---

* 它可以像这样动态完成 a.map(lambda row: collections.namedtuple('Person', ['name', 'age', 'gender'])(*row)) 或在 mapPartitions 中定义 Person，但它既不漂亮也不高效。