The CDH 5.4.0 plugin allows the building of a high-availability CDH cluster through HDFS cluster groups. This can be achieved by setting up a YARN HA ResourceManager in Active/Standby mode architecture. The Sahara plugins deploys automatically the CDH cluster, and if HA enabled, ResourceManagers can fail over automatically through its failover controller or manually using the admin command-line interface. As shown in the next diagram, any active ResourceManager in the cluster needs to write its current states into ZooKeeper clusters. When the failure occurs, the StandBy ResourceManager will be promoted to be active and starts to load the ResourceManager state:
Like HDP, there are few requirements in order to support HA functionality in Sahara. Make sure to add an odd number of nodes running JOURNALNODE
processes with at least a number of 3 per CDH cluster. Unlike HDP, it is not necessary to go through the cluster template and enable the HDFSHA option in the template...