For someone coming from the MapReduce v1 world of job trackers and task trackers, switching to YARN can be a little confusing. The number of “slots” that are available in YARN is more dynamic, depending on the needs of the application.
YARN is quite flexible in how memory requirements for applications are handled. Along with this comes a number of configuration settings that could be a little confusing. Here’s an overview of the settings that are available. This post is specifically related to MapReduce jobs, but it includes general YARN concepts.