Glad we have an internal load balancing/failover LLM inference system, so this outage won't take prod down for us.