Hadoop平台下的内存溢出诊断方法

Method for Overflow Diagnosis of Memory in Hadoop Platform

针对Hadoop平台的内存溢出问题,结合框架结构和运行机理,分析了常见的内存溢出原因,提出一种诊断方法,通过分析Hadoop运行时数据,自动定位错误所在运行阶段,并结合内存对象分配情况和系统配置参数,诊断导致内存溢出的原因.利用电网数据处理的典型应用场景进行实验,实验结果验证了方法的有效性.

In this paper, it analyzes the mechanism of Hadoop and summarizes the common issue of memory leak, and proposes a method to diagnose this issue. The proposed approach could diagnose the phase of the overflow of memory occurs, the objects which consume most of the memory space, and the related configurations, to help the Hadoop user to find the root cause of error during out of memory. It also evaluates the effectiveness of the proposed approach under typical data processing applications for the power grid.