ResourceManager
L the ResourceManager responsible for resource management and allocation of the entire cluster, is a global resource management systems.
L the NodeManager in a heartbeat way to the ResourceManager report resource usage (currently mainly CPU and memory usage). RM only accept NM resource information in return for specific resources to deal with the NM own process.
L YARN Scheduler according to application assigned to the requested resource, is not responsible for application job of monitoring, tracking, operational status feedback, to start work.
NodeManager
L the NodeManager resource and task manager on each node, which is managing the machine agent in charge of the node program run, and manage and monitor the node resources . YARN each cluster node runs a the NodeManager .
L the NodeManager timing of the ResourceManager report node resources ( the CPU , memory) and the usage Container operating state. When ResourceManager when down NodeManager automatically connects RM standby node.
L the NodeManager receives and processes from ApplicationMaster the Container start, stop, and other requests.
ApplicationMaster
l submitted by the user for each application contains a ApplicationMaster , it can run ResourceManager on the outside of the machine.
l responsible RM consultation scheduler for resources (with Container represented).
l The resulting further tasks assigned to the internal tasks ( second allocation of resources ) .
l and NM communication to start / stop the task.
l monitoring all the tasks running, and when a job fails to re-apply for funding for the mission to restart the task.
l Current YARN comes with two ApplicationMaster achieved, for a demonstration of AM Examples programming method DistributedShell , it can apply a number of Container to run a parallel Shell command or Shell scripts; the other is running MapReduce application AM - MRAppMaster .
Note: RM is only responsible for monitoring AM , and AM start when it fails. RM is not responsible AM fault-tolerant internal tasks, fault-tolerant task by AM complete.
to sum up:
- ResourceManager (RM): main entrance role Uniform Resource Scheduling
- Nodemanager (NM): Each machine is responsible for specific resource management
- ApplicationMaster (AM): supervision and management of resources within the program leader responsible for execution of the application program