Hi all,
I have a hadoop cluster with 4 dataNodes+nodeManager and 1
namenode+resourceManager. I'm launching a MR job (identity mapper and
identity reducer) with the relevant memory settings set to appropriate
values :
mapreduce.[map|reduce].memory.mb, JAVA_CHILD_OPTS, map sort buffer,
reduce buffer, ...
Does the framework guarantee that I will not run into a "Out of memory"
situation for any input dataset size ? i.e. The only things that can
lead to a "Out of memory" on mappers or reducers are :
Bad memory settings (for example map sort buffer > mapreduce.map.memory.mb )
Bad Mapper/Reducer code (user code)
Best Regards,
