Amazon EMR：Pyspark 存在奇怪的依赖性问题

Question

我在将 pyspark 作业发送到 EMR 集群上运行时遇到了问题，所以我登录到主节点并直接运行 spark-submit

我有一个提交给 pyspark 的 python 文件，在这个文件中我有：

import subprocess
from pyspark import SparkContext, SparkConf
import boto3
from boto3.s3.transfer import S3Transfer
import os, re
import tarfile
import time
...

当我在集群模式下尝试运行时，我得到：（来自纱线日志，为简洁起见进行了修剪）

16/01/31 21:45:57 INFO spark.CacheManager: Partition rdd_2_0 not found, computing it
16/01/31 21:45:57 INFO spark.CacheManager: Partition rdd_1_0 not found, computing it
16/01/31 21:45:57 ERROR executor.Executor: Exception in task 0.0 in stage 0.0 (TID 0)
org.apache.spark.api.python.PythonException: Traceback (most recent call last):
  File "/mnt1/yarn/usercache/hadoop/appcache/application_1454273602144_0005/container_1454273602144_0005_01_000002/pyspark.zip/pyspark/worker.py", line 98, in main
    command = pickleSer._read_with_length(infile)
  File "/mnt1/yarn/usercache/hadoop/appcache/application_1454273602144_0005/container_1454273602144_0005_01_000002/pyspark.zip/pyspark/serializers.py", line 164, in _read_with_length
    return self.loads(obj)
  File "/mnt1/yarn/usercache/hadoop/appcache/application_1454273602144_0005/container_1454273602144_0005_01_000002/pyspark.zip/pyspark/serializers.py", line 422, in loads
    return pickle.loads(obj)
ImportError: No module named boto3.s3.transfer

        at org.apache.spark.api.python.PythonRunner$$anon.read(PythonRDD.scala:166)
        at org.apache.spark.api.python.PythonRunner$$anon.<init>(PythonRDD.scala:207)
        at org.apache.spark.api.python.PythonRunner.compute(PythonRDD.scala:125)
        at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:70)
        at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
        at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:69)
        at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
        at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:70)
        at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
        at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:69)
        at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
        at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:70)
        at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
        at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
        at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
        at org.apache.spark.scheduler.Task.run(Task.scala:89)
        at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
        at java.lang.Thread.run(Thread.java:745)

稍后我收到关于无法导入 boto3 的错误。

如果我运行在客户端模式下，我只会得到关于 boto3 的 ImportError。s3.transfer。

Job aborted due to stage failure: Task 0 in stage 0.0 failed 4 times, most recent failure: Lost task 0.3 in stage 0.0 (TID 3, ip-172-31-39-79.us-west-2.compute.internal): org.apache.spark.api.python.PythonException: Traceback (most recent call last):
  File "/mnt1/yarn/usercache/hadoop/appcache/application_1454273602144_0005/container_1454273602144_0005_01_000002/pyspark.zip/pyspark/worker.py", line 98, in main
    command = pickleSer._read_with_length(infile)
  File "/mnt1/yarn/usercache/hadoop/appcache/application_1454273602144_0005/container_1454273602144_0005_01_000002/pyspark.zip/pyspark/serializers.py", line 164, in _read_with_length
    return self.loads(obj)
  File "/mnt1/yarn/usercache/hadoop/appcache/application_1454273602144_0005/container_1454273602144_0005_01_000002/pyspark.zip/pyspark/serializers.py", line 422, in loads
    return pickle.loads(obj)
ImportError: No module named boto3.s3.transfer

    at org.apache.spark.api.python.PythonRunner$$anon.read(PythonRDD.scala:166)
    at org.apache.spark.api.python.PythonRunner$$anon.<init>(PythonRDD.scala:207)
    at org.apache.spark.api.python.PythonRunner.compute(PythonRDD.scala:125)
    at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:70)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
    at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:69)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
    at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:70)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
    at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:69)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
    at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:70)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
    at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
    at org.apache.spark.scheduler.Task.run(Task.scala:89)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
    at java.lang.Thread.run(Thread.java:745)

但是，如果我检查 pip freeze:

boto3==1.2.3
botocore==1.3.23

如果我在 master 上打开 Spark Shell 并执行以下操作：

import boto3
client = boto3.client("s3")

它工作正常。

这里有某种虚拟环境吗？我完全糊涂了。

编辑忘了说我正在使用最新的 EMR 版本和 Spark 1.6.0。

此外，这在我自己的机器上以本地模式运行良好。

Answer 1

嗯，derp，我发现了问题。

原来我不得不 pip install boto3，默认情况下，EMR 节点不会安装它。

这是一个错误描述性很强的案例。

Amazon EMR：Pyspark 存在奇怪的依赖性问题

Amazon EMR: Pyspark having strange dependency issues

python

amazon-web-services

amazon-emr

emr

pyspark