A parks computer-vision InfiniBand cluster loses its primary subnet manager during a drill. What should operators validate?

During a parks computer-vision InfiniBand HA drill, operators should intentionally fail the primary subnet manager and validate that the standby promotes and continues managing the fabric. HA is proven by failover behavior—role promotion, continued SM service, and preserved fabric management—not by chasing SM priority reassignment on surviving switches after a failover event. Reassigning SM priority on surviving switches after a failover event fails by concept: that is a configuration change, not HA validation, and parks CV workloads lose management focus if operators chase priority tweaks instead of confirming the standby's active SM role. Exporting the subnet manager SA database to compute nodes for local election fails because compute nodes lack SM capability and the SA database is a centralized fabric state, not a distributed election dataset; validating export distracts from confirming the standby's promotion. Initiating a LID reassignment sweep manually on the host channel adapters fails because SM failover automatically manages LIDs, and a manual sweep disrupts active connections without proving the standby's SM promotion; it's a disruptive action, not an HA validation. Exam caveat: NCP-AIN HA questions focus on SM failover/promotion; SM priority reassignment, SA database export to compute nodes, and manual LID sweeps are distractors. Operational check: fail the parks primary SM in a maintenance window, confirm standby becomes master, verify management of the IB subnet continues, restore the original node as standby or per runbook, and record promotion time for civic HA evidence.