ApacheSparkBook icon indicating copy to clipboard operation
ApacheSparkBook copied to clipboard

Shuffle Write和Shuffle Read阶段溢写的时机不同

Open MingRongXi opened this issue 2 years ago • 0 comments

利杰你好,我有个问题想请教你。我在看Spark源码时,发现Shuffle Write是先往Map里插入值,然后再判断是否需要溢写;而Shuffle Read是先判断是否需要溢写,然后再插入值。按照我个人理解,采用Shuffle Read的方式内存溢出的风险会更低,Shuffle Write可能会在扩容时导致溢出。你知道Spark为什么要这样设计吗 Shuffle Map image Shuffle Read image

MingRongXi avatar Jan 07 '23 10:01 MingRongXi