一：背景

Spark 任务出现了container内存负载出现OOM

二：问题

Application application_xxx_xxxx failed 2 times due to AM Container for appattempt_xxxx_xxxx_xxxx exited with exitCode: -104
Failing this attempt.Diagnostics: Container [pid=78835,containerID=container_e14_1611819047508_2623322_02_000003] is running beyond physical memory limits. Current usage: 6.6 GB of 6.6 GB physical memory used; 11.9 GB of 32.3 TB virtual memory used. Killing container.
Dump of the process-tree for container_e14_1611819047508_2623322_02_000003 :
|- PID PPID PGRPID SESSID CMD_NAME USER_MODE_TIME(MILLIS) SYSTEM_TIME(MILLIS) VMEM_USAGE(BYTES) RSSMEM_USAGE(PAGES) FULL_CMD_LINE

三、分析

1、Spark OOM主要出现方：Drive和Executor。

        Driver:主要是collect、show等

       Executor：主要是缓存RDD等

2、查看UI

4、内存简介

     参考Spark （一）：Executor内存_在前进的路上-CSDN博客一、背景 Spark是基于内存的分布式计算引擎，我们需对Executor内存管理的详细了解，方便我们遇到OOM解决问题、或者优化时更好调优，，Spark任务启动时有两个进程，分别为Driver、Executor进程，Driver进程（内存默认1G）可能在本地启动，也可能在集群中某个工作节点上启动（根据提交模式client、Cluster等）。Driver启动会申请资源（Executor），根据配置启动对应Executor数量，每个Executor进程都会对应一定数量的内存和CPU COR...https://blog.csdn.net/congcong68/article/details/122274441

3、调整对应参数

       executor-memory
       num-executors
        driver-memory