Application Master

Once a client submits a new job to Resource Manager, it will then contact any available Node Manager and initiate the Application Master service in the data node. Application Master will then contact NameNode to retrieve the metadata required to execute the job. It will then create one Map task per input split.

Application Master will also decide how many containers it requires to execute a job. If the task is small, Application Master executes them within its own JVM or container, which helps to reduce the overhead of requesting and allocating a separate container. These small tasks are commonly known as Uber Tasks.

Parameters of Uber Jobs can be configured by editing the mapred-site.xml file in /etc/hadoop/ directory of the Hadoop installation. These parameters are as follows:

Application Master will terminate itself after completing a job assigned to it. You may think of Application Master as a person hired for a specific task who leaves once they've finished.