Major processes in Hadoop

Major processes in a Hadoop cluster
Master:namenode, ResourceManager,
Slaves:datanode, NodeManager, Runjar, Mrappmaster,yarnchild

Where Runjar, Mrappmaster,yarnchild is created with the creation of a job and terminates as the job completes. Their roles are:
Runjar: Completes the job initialization, including getting the Jobid, uploading the jar package to HDFs, and so on.
Mrappmaster: A process for each job, primarily tracking job operations, requesting resources from RM, and so on.
Yarnchild: Runs the specific map/reduce task.

Job Startup process:
Resourcemanage,nodemanager->runjar->mrappmaster->yarnchild

Job Exit Process:
Yarnchild->mrappmaster->runjar
That is, all the map/reduce are completed, Mrappmaster only exit, the final Runjar exit, job completion.