Mobile Edge Computing is an emerging paradigm that offloads tasks to edge servers located near end users, effectively reducing backhaul network congestion and enhancing Quality of Service. However, due to the limited caching capacity and computational resources of edge servers, MEC systems continue to face significant challenges in service caching and task offloading. Existing approaches fail to simultaneously ensure user QoS and maintain computational reliability under high failure rates. To address these challenges in failure-prone MEC environments, we propose a fault-tolerant approach for service caching and task offloading. Our method incorporates a primary-backup replication mechanism to enhance fault tolerance and employs a Deep Q-Network-based framework to optimize service caching distribution. Experimental results demonstrate that the proposed method outperforms benchmarks.