Near real-time anomaly detection and localization are critical requirements for operational Internet Service Provider (ISP) networks, where short-lived performance degradations can significantly affect service quality. However, widely deployed monitoring systems still rely on static thresholds or offline analysis, which are poorly suited to non-stationary traffic patterns and heterogeneous network metrics. This paper presents a near real-time anomaly detection and localization framework for Tier-2 ISP networks based on Round-Trip Time (RTT) and interface utilization telemetry. The proposed method applies a rolling-window exhaustive change-point detection strategy derived from the penalized optimal partitioning formulation associated with the PELT family of methods. Instead of relying on global pruning and fixed penalty selection, all candidate split points are exhaustively evaluated within bounded temporal windows to improve detection stability under dynamic traffic conditions. Detected change points are temporally aggregated across multiple metrics and further analyzed using an ensemble of complementary statistical and density-based techniques, including Z-score deviation analysis, Shape-Based Distance, Multidimensional Scaling, Local Outlier Factor, and Isolation Forest, to enable interface-level anomaly localization. The framework is evaluated using both a public labeled network anomaly dataset and real production telemetry collected from a Tier-2 ISP backbone network. Experimental results on the labeled dataset demonstrate consistent detection performance across multiple Key Performance Indicators (KPIs), achieving F1-scores of up to 0.84. Evaluation on production telemetry further shows that the proposed framework successfully identifies multiple operationally confirmed anomalies while maintaining detection latency suitable for near real-time network operations. Overall, the results indicate that rolling-window exhaustive changepoint analysis combined with ensemble-based multivariate localization provides a practical and deployable solution for anomaly detection in large-scale ISP networks. © 2026 The Authors.