Predict Boldly, Recover Cautiously: Fast On-Router Route Anomaly Prediction and Recovery

Abstract

Route anomalies can quickly propagate into path detours, packet loss, and service degradation. However, existing pipelines are often centralized: routers export telemetry to a controller, the controller performs data cleaning and diagnosis, and mitigation actions are then pushed back to the network. This observe-then-react loop is too slow for transient or fast-propagating routing failures. We present RouterOPS, a router-native lightweight framework for fast route anomaly prediction and cautious recovery. RouterOPS enables each router to perform in-situ self-testing and neighbor-testing, and to maintain compact risk scores from local telemetry, active probes, and routing behavior. These scores are used to forecast suspicious self or neighbor states before visible service impact and are reported to the controller for cross-checking and audit. Following the principle of predict boldly and recover cautiously, RouterOPS triggers local recovery only when suspicion is persistent and safety guards are satisfied. Our production deployment shows that RouterOPS improves anomaly response timeliness while preserving routing safety.