This document discusses YARN high availability (HA) and describes: 1) The need for HA in YARN due to the ResourceManager being a single point of failure. 2) The HA architecture which uses an active/standby ResourceManager pair with shared state stored in Zookeeper. 3) How failover works automatically through Zookeeper election and how clients are redirected to the active ResourceManager. 4) The configuration needed to set up YARN HA.