Trending Articles

Blog Post

Apache Hadoop YARN – Comprehensive Guide

Apache Hadoop YARN – Comprehensive Guide


The Apache Hadoop YARN is the resource management and job scheduling technology in the distributed processing framework of the open-source Hadoop. The heart of this Apache technology is commonly known as just YARN, which stands for Yet Another Resource Negotiator.

Additionally, it plays a crucial role in efficiently allocating resources and scheduling tasks across a Hadoop cluster. Apache Hadoop existed to enhance the usage and solve challenging big data issues.

How Does Apache Hadoop YARN Work?

In brief, YARN acts like a leader managing a data processing group. Client applications submit jobs to the ResourceManager to run on the Hadoop cluster. Subsequently, the Manager allocates resources and launches an ApplicationMaster container on a cluster node.

The ApplicationMaster then negotiates appropriate resource containers from the ResourceManager. The NodeManagers run on each worker node and are responsible for launching and monitoring containers that execute the actual task the applications need to complete.

The ApplicationMaster assists the NodeManagers to launch and monitor the containers executing the tasks for the application. Upon completion, the ApplicationMaster returns the final output to the client after aggregating all the outputs of the containers.

Consequently, the ResourceManager de-allocates the resources as the ApplicationMaster returns the final output. Therefore, this dynamic process of allocating and separating responsibilities makes YARN flexible, scalable, and efficient for various big data performances.

Features and Functions of Apache Hadoop YARN:

  1. Centralized Resource Management: By introducing two key components, YARN separates resource management from job execution:
  • ResourceManager (RM): The central authority, arbitrating resources among applications and scheduling tasks.
  • NodeManager (NM): Runs on each cluster node, managing resources and monitoring container execution.
  1. ApplicationMaster (AM): Each application has its own AM, which is responsible for negotiating resources with the RM, launching tasks on NMs, and monitoring the progress.
  • Job/Task Scheduling: The AM coordinates with the ResourceManager to schedule tasks and containers.
  1. Resource Allocation: To offer further flexibility and better resource utilization compared to Hadoop 1.0’s static allocation, YARN allocates resources in smaller units called containers (memory + CPU).
  2. Flexibility and Scalability: Beyond MapReduce, YARN supports various data processing frameworks, including Spark, Tez, and Flink, which makes it a versatile platform for various big data workloads.
  3. Multi-tenancy: The negotiator supports the operations of numerous applications from different users or organizations safely and securely in a shared cluster.
  4. Reservation System: In ensuring anticipated execution, the system allows users to reserve resources in advance for critical jobs.

Benefits of YARN:

  1. YARN splits HDFS from MapReduce to make Hadoop more suitable for real-time processing and other non-MapReduce applications.
  2. MapReduce is currently just one of many processing engines that can run on YARN, and in addition, it doesn’t have any lock on Hadoop batch processing.
  3. Spark and other technologies like Flink and Storm can run stream processing on YARN.
  4. The technology opened up new uses for HBase, Hive, Drill, Impala, and other engines.
  5. It offers scalability, resource utilization, high availability, and performance improvements over MapReduce.

Applications and  Use Cases of Apache Hadoop YARN:

  • It permits functioning distributed batch processing jobs for ETL, data analysis, machine learning, etc.
  • Through frameworks like Storm, Samza, and Spark, streaming the process of large volumes of data is possible in real-time.
  • Interactive SQL engines like Hive, Impala, and Presto can run on YARN for faster queries on huge datasets.
  • Graph-parallel systems like Giraph and GraphX leverage YARN to acquire large graph computations.
  • The component can be used as a standard cluster manager to run containerized applications like Docker.
  • Sqoop and Flume are tools that use YARN to take in data from external sources into HDFS or HBase.


Apache Hadoop YARN is a dominant resource management framework that has revolutionized big data processing. By separating the resource governance and processing elements, YARN empowers a broad spectrum of processing engines to run on a single Hadoop cluster, providing a cost-effective solution for large-scale data analysis.

Overall, YARN’s ability to manage resources across the cluster makes it essential for Hadoop and a favored choice for big data analytics.

Related posts