spark-shell启动报错:Yarn application has already ended! It might have been killed or unable to launch application master
spark-shell不支持yarn cluster,以yarn client方式启动
spark-shell --master=yarn --deploy-mode=client
启动日志,错误信息如下
其中“Neither spark.yarn.jars nor spark.yarn.archive is set, falling back to uploading libraries under SPARK_HOME”,只是一个警告,官方的解释如下:
大概是说:如果 spark.yarn.jars 和 spark.yarn.archive都没配置,会把$SPAR_HOME/jars下面所有jar打包成zip文件,上传到每个工作分区,所以打包分发是自动完成的,没配置这俩参数没关系。
"Yarn application has already ended! It might have been killed or unable to launch application master",这个可是一个异常,打开mr管理页面,我的是 http://192.168.128.130/8088 ,
重点在红框处,2.2g的虚拟内存实际值,超过了2.1g的上限。也就是说虚拟内存超限,所以contrainer被干掉了,活都是在容器干的,容器被干掉了,还玩个屁。
解决方案
yarn-site.xml 增加配置:
2个配置2选一即可
1 <!--以下为解决spark-shell 以yarn client模式运行报错问题而增加的配置,估计spark-summit也会有这个问题。2个配置只用配置一个即可解决问题,当然都配置也没问题--> 2 <!--虚拟内存设置是否生效,若实际虚拟内存大于设置值 ,spark 以client模式运行可能会报错,"Yarn application has already ended! It might have been killed or unable to l"--> 3 <property> 4 <name>yarn.nodemanager.vmem-check-enabled</name> 5 <value>false</value> 6 <description>Whether virtual memory limits will be enforced for containers</description> 7 </property> 8 <!--配置虚拟内存/物理内存的值,默认为2.1,物理内存默认应该是1g,所以虚拟内存是2.1g--> 9 <property> 10 <name>yarn.nodemanager.vmem-pmem-ratio</name> 11 <value>4</value> 12 <description>Ratio between virtual memory to physical memory when setting memory limits for containers</description> 13 </property>
修改后,启动hadoop,spark-shell.