City planning runs an InfiniBand maintenance window and must catch unexpected link flaps live. What should operators watch?

During a city planning InfiniBand maintenance window, operators must watch live UFM InfiniBand link-status monitoring throughout the change. Live fabric views catch unexpected flaps, partial optical failures, and unintended neighbor disruptions while the change is still reversible. Municipal change control for InfiniBand is incomplete if operators only learn about fabric instability after the window closes and workloads return.

Relying only on hourly UFM event log polling for subnet joins during the change window fails because event-log polling is periodic, not real-time; subnet joins do not reveal unexpected link flaps, and by the next hourly check a flap may already have been missed or misattributed. Continuous UFM congestion and SHARP telemetry tracking during the window fails as a substitute because congestion and SHARP metrics measure fabric load and collectives, not port-level link state; they will not surface an unexpected flap or optical failure as it happens. Post-step review of UFM topology snapshots after the change phase fails directly against the goal: it is retrospective, so unexpected flaps are only seen after the window closes and the fabric is no longer in the maintenance state.

Exam caveat: Live UFM monitoring complements—not replaces—a documented change plan, rollback criteria, and maintenance mode for expected alarms. Suppress only known, planned events; keep unexpected link flaps visible and actionable.

Operational check: Before the city planning window starts, open UFM link-status and topology views, note baseline healthy ports, keep a second operator watching UFM during the change, and pause or roll back if unexpected flaps appear on non-targeted links. Capture UFM screenshots or exports as evidence attached to the change record.