我有一个程序： 1.读取一些数据 2.执行一些操作 3.保存csv文件 4.将该文件传输到FTP
我正在使用Amazon EMR集群和PySpark来完成此任务。
对于步骤4，我需要将CSV保存在本地存储上而不是HDFS上。为此，我将Spark Dataframe转换为Pandas数据帧。
一个片段可能是：
from pyspark import SparkContext
from pyspark.conf import SparkConf
from pyspark.sql import HiveContext
from pyspark.sql.types import StructType, StructField, LongType, StringType
from pyspark.mllib.evaluation import *
from pyspark.sql.functions import *
from pyspark.sql import Row
from time import time
import timeit
from datetime import datetime, timedelta
import numpy as np
import random as rand
import pandas as pd
from itertools import combinations, permutations
from collections import defaultdict
from ftplib import FTP
from pyspark.sql import SQLContext
conf = SparkConf().setAppName("Recommendation").set('spark.driver.memory', '8G').set('spark.executor.memory', '4G')
sc = SparkContext(conf = conf)
sqlContext = SQLContext(sc)
readRdd = sqlContext.read.format('com.databricks.spark.csv').load('s3n://my-bucket/myfile' + path)
df = readRdd.toPandas() # <---------- PROBLEM
print('toPandas() completed')
df.to_csv('./myFile')
问题是：
当我从同一群集上的Jpyter iPython笔记本运行此代码时，它就像一个魅力。但是当我使用Spark Submit运行此代码，或将其作为EMR的一个步骤添加时，代码在以下行中失败：
df = readRdd.toPandas()
'toPandas（）已完成'永不打印
在spark作业监视器中，我可以看到toPandas（）方法被执行但在此之后我得到了错误。
16/10/10 13:17:47 INFO YarnAllocator: Driver requested a total number of 1 executor(s).
16/10/10 13:17:47 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:17:47 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:17:47 INFO TaskSetManager: Finished task 1462.0 in stage 17.0 (TID 10624) in 2089 ms on ip-172-31-38-70.eu-west-1.compute.internal (1515/1516)
16/10/10 13:17:47 INFO TaskSetManager: Finished task 1485.0 in stage 17.0 (TID 10647) in 2059 ms on ip-172-31-38-70.eu-west-1.compute.internal (1516/1516)
16/10/10 13:17:47 INFO YarnClusterScheduler: Removed TaskSet 17.0, whose tasks have all completed, from pool
16/10/10 13:17:47 INFO DAGScheduler: ResultStage 17 (toPandas at 20161007_RecPipeline.py:182) finished in 12.609 s
16/10/10 13:17:47 INFO DAGScheduler: Job 4 finished: toPandas at 20161007_RecPipeline.py:182, took 14.646644 s
16/10/10 13:17:47 INFO YarnAllocator: Driver requested a total number of 0 executor(s).
16/10/10 13:17:47 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:17:47 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:17:50 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:17:50 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:17:53 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:17:53 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:17:56 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:17:56 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:17:59 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:17:59 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:02 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:02 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:05 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:05 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:08 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:08 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:11 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:11 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:14 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:14 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:17 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:17 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:20 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:20 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:23 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:23 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:26 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:26 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:29 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:29 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:32 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:32 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:35 INFO YarnAllocator: Canceling requests for 0 executor containers
16/10/10 13:18:35 WARN YarnAllocator: Expected to find pending requests, but found none.
16/10/10 13:18:36 ERROR ApplicationMaster: RECEIVED SIGNAL 15: SIGTERM
16/10/10 13:18:36 INFO SparkContext: Invoking stop() from shutdown hook
16/10/10 13:18:36 INFO SparkUI: Stopped Spark web UI at http://172.31.37.28:45777
16/10/10 13:18:36 INFO YarnClusterSchedulerBackend: Shutting down all executors
16/10/10 13:18:36 INFO YarnClusterSchedulerBackend: Asking each executor to shut down
16/10/10 13:18:36 INFO MapOutputTrackerMasterEndpoint: MapOutputTrackerMasterEndpoint stopped!
16/10/10 13:18:36 ERROR PythonRDD: Error while sending iterator
java.net.SocketException: Connection reset
at java.net.SocketOutputStream.socketWrite(SocketOutputStream.java:118)
at java.net.SocketOutputStream.write(SocketOutputStream.java:159)
at java.io.BufferedOutputStream.write(BufferedOutputStream.java:122)
at java.io.DataOutputStream.write(DataOutputStream.java:107)
at java.io.FilterOutputStream.write(FilterOutputStream.java:97)
at org.apache.spark.api.python.PythonRDD$.org$apache$spark$api$python$PythonRDD$$write$1(PythonRDD.scala:440)
at org.apache.spark.api.python.PythonRDD$$anonfun$writeIteratorToStream$1.apply(PythonRDD.scala:452)
at org.apache.spark.api.python.PythonRDD$$anonfun$writeIteratorToStream$1.apply(PythonRDD.scala:452)
at scala.collection.Iterator$class.foreach(Iterator.scala:727)
at scala.collection.AbstractIterator.foreach(Iterator.scala:1157)
at org.apache.spark.api.python.PythonRDD$.writeIteratorToStream(PythonRDD.scala:452)
at org.apache.spark.api.python.PythonRDD$$anon$2$$anonfun$run$1.apply$mcV$sp(PythonRDD.scala:648)
at org.apache.spark.api.python.PythonRDD$$anon$2$$anonfun$run$1.apply(PythonRDD.scala:648)
at org.apache.spark.api.python.PythonRDD$$anon$2$$anonfun$run$1.apply(PythonRDD.scala:648)
at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1250)
at org.apache.spark.api.python.PythonRDD$$anon$2.run(PythonRDD.scala:649)
16/10/10 13:18:36 ERROR ApplicationMaster: User application exited with status 143
16/10/10 13:18:36 INFO ApplicationMaster: Final app status: FAILED, exitCode: 143, (reason: User application exited with status 143)
16/10/10 13:18:36 INFO MemoryStore: MemoryStore cleared
16/10/10 13:18:36 INFO BlockManager: BlockManager stopped
16/10/10 13:18:36 INFO BlockManagerMaster: BlockManagerMaster stopped
16/10/10 13:18:36 INFO OutputCommitCoordinator$OutputCommitCoordinatorEndpoint: OutputCommitCoordinator stopped!
16/10/10 13:18:36 INFO RemoteActorRefProvider$RemotingTerminator: Shutting down remote daemon.
16/10/10 13:18:36 INFO SparkContext: Successfully stopped SparkContext
16/10/10 13:18:36 INFO ShutdownHookManager: Shutdown hook called
16/10/10 13:18:36 INFO RemoteActorRefProvider$RemotingTerminator: Remote daemon shut down; proceeding with flushing remote transports.
16/10/10 13:18:36 INFO ShutdownHookManager: Deleting directory /mnt3/yarn/usercache/hadoop/appcache/application_1476100925559_0002/spark-eab43d4e-7201-4bcb-8ee7-0e7b546e8fd8
16/10/10 13:18:36 INFO ShutdownHookManager: Deleting directory /mnt/yarn/usercache/hadoop/appcache/application_1476100925559_0002/spark-1d88398f-ecd5-4d94-a42a-a406b3d566af/pyspark-34bec23c-a686-475d-85c9-9e9228b23239
16/10/10 13:18:36 INFO ShutdownHookManager: Deleting directory /mnt/yarn/usercache/hadoop/appcache/application_1476100925559_0002/spark-1d88398f-ecd5-4d94-a42a-a406b3d566af
16/10/10 13:18:36 INFO ShutdownHookManager: Deleting directory /mnt3/yarn/usercache/hadoop/appcache/application_1476100925559_0002/container_1476100925559_0002_01_000001/tmp/spark-96cdee47-e3f3-45f4-8bc7-0df5928ef53c
16/10/10 13:18:36 INFO ShutdownHookManager: Deleting directory /mnt2/yarn/usercache/hadoop/appcache/application_1476100925559_0002/spark-f6821ea1-6f37-4cc6-8bba-049ac0215786
16/10/10 13:18:36 INFO ShutdownHookManager: Deleting directory /mnt1/yarn/usercache/hadoop/appcache/application_1476100925559_0002/spark-1827cae8-8a60-4b29-a4e5-368a8e1856fd
我的群集配置如下：
spark-defaults spark.driver.maxResultSize 8G
spark-defaults spark.driver.memory 8G
spark-defaults spark.executor.memory 4G
Spark Submit命令如下所示：
spark-submit --deploy-mode cluster s3://my-bucket/myPython.py
答案 0 :(得分：0)
问题在于：
spark-submit --deploy-mode cluster s3://my-bucket/myPython.py
在上面的命令中，部署模式设置为cluster，这意味着将从核心节点中选择一个节点来运行驱动程序。由于允许的驱动程序内存为8G且核心节点是较小的物理实例，因此它们总是会耗尽所需的内存。
解决方案是在客户端模式下部署，其中驱动程序将始终在主节点上运行（在我的情况下，具有更多资源的更大的物理实例）将不会耗尽整个过程所需的内存。
由于它是一个专用集群，因此该解决方案适用于我的情况。
如果共享群集的部署模式必须是群集，则使用更大的实例应该可以正常工作。