Abstract
<title>Abstract</title> <p> <italic> <bold>Cross-region cloud services may violate predefined recovery objectives when replication lag, insufficient standby capacity, permission propagation failures, or service dependency anomalies delay failover. This study develops a failover validation framework integrating fault injection, state monitoring, DeepAR-based probabilistic prediction, pre-switchover constraint checks, and dynamic action selection to jointly assess recovery point objective (RPO) and recovery time objective (RTO) risks. Experiments covered four cloud regions and 860 failover simulations, generating 180 TB of test data and 76 million monitoring points. DeepAR achieved a 7.1% error for 60-min replication-lag prediction, a 90.5% recall for RPO-exceedance detection, and 93.8% coverage for RTO prediction intervals. After prediction and dependency checks were incorporated into the switching process, the median failover time decreased from 31.4 to 12.7 min, data consistency reached 99.98%, and secondary failures associated with permissions, capacity, and service dependencies decreased by 48.6%. The framework provides a quantitative approach for validating recovery objectives and improving cross-region failover decisions for critical cloud services.</bold> </italic> </p>