Distributed as part of Elasticsearch for Apache Hadoop, Elasticsearch on YARN is a separate, stand-alone, self-contained CLI (command-line interface) that allows Elasticsearch to run (and thus be managed) within a YARN environment.
In other words, Elasticsearch can use YARN to allocate its resources and be started and stopped, on said resources, through YARN infrastructure.
IMPORTANT: Currently, due to the inherent limitations in YARN, Elasticsearch for Apache Hadoop is in beta; that is we encourage users to try it out and report feedback however it might not contain all the features needed for production usage (most of which can be addressed outside YARN).
Unfortunately, YARN does not (currently) provide comprehensive support for long-running services; features such as storage/network/host affinity, topology information and broadcasting (in particular in case of restarts) and ip/storage provisioning are missing; while there are solutions for them, these are outside YARN which is far from ideal.
However this topic is addressed within the YARN community - as it progresses, the new functionality will be incorporated into this project and once completed, the project will be promoted out of beta.