spark作业执行失败分析

2019独角兽企业重金招聘Python工程师标准>>>

spark作业执行失败，重新执行的时候，查看sparkui，发现存在大量失败的task，执行结束后，通过yarn-ui看到报错日志如下：

User class threw exception: org.apache.spark.SparkException: Job aborted due to stage failure: 
ShuffleMapStage 1 (javaRDD at SumDeliveryIndexFactory.java:628) has failed the maximum allowable 
number of times: 4. Most recent failure reason: org.apache.spark.shuffle.FetchFailedException: 
Failed to connect to xxxx/10.136.22.22:34192

由报错可见，Failed to connect to xxxx/10.136.22.22:34192，连接10.136.22.22失败。进入10.136.22.22主机，查看nodemanager日志，于是看到了以下的错误信息：running beyond physical memory limits.Killing container。可见，由于使用的物理内存超出了container的内存大小，被强制kill了。

解决办法：spark-submit 添加参数，调大spark.yarn.executor.memoryOverhead=4G

错误日志：

2017-11-14 11:33:07,273 INFO 
org.apache.hadoop.yarn.server.nodemanager.containermanager.monitor.ContainersMonitorImpl: Memory 
usage of ProcessTree 236569 for container-id container_e31_1510205192678_147416_02_000024: 
40.7 GB of 40 GB physical memory used; 41.9 GB of 84 GB virtual memory used
2017-11-14 11:33:07,273 WARN 
org.apache.hadoop.yarn.server.nodemanager.containermanager.monitor.ContainersMonitorImpl: Process 
tree for container: container_e31_1510205192678_147416_02_000024 has processes older than 1 
iteration runningover the configured limit. Limit=42949672960, current usage = 43653300224
2017-11-14 11:33:07,274 WARN 
org.apache.hadoop.yarn.server.nodemanager.containermanager.monitor.ContainersMonitorImpl: 
Container [pid=236569,containerID=container_e31_1510205192678_147416_02_000024] is running beyond 
physical memory limits. Current usage: 40.7 GB of 40 GB physical memory used; 41.9 GB of 84 GB 
virtual memory used. Killing container.    

参考博客：

1、Spark Executor在YARN上的内存分配
	http://blog.csdn.net/hammertank/article/details/48346285

2、yarn is running beyond physical memory limits 问题解决
	http://blog.csdn.net/oaimm/article/details/25298691

3、Yarn简单介绍及内存配置 
    http://blog.chinaunix.net/uid-28311809-id-4383551.html      

 

转载于:https://my.oschina.net/sniperLi/blog/1574280