Either one will work, but on a long-running cluster where you rarely restart Yarn, set it to the FQDN of rm1, and also set rm1 to be the active RM. That's because the discovery of active RM is done in sequence rm1 --> rm2.
Either one will work, but on a long-running cluster where you rarely restart Yarn, set it to the FQDN of rm1, and also set rm1 to be the active RM. That's because the discovery of active RM is done in sequence rm1 --> rm2.