<p>The rapid adoption of IoT applications has led to the continuous generation of vast amounts of data, demanding efficient processing, storage, and real-time response delivery. Edge computing addresses these demands by processing data closer to its source, significantly reducing latency and rejection rate. However, limited and heterogeneous resources across edge nodes pose challenges for dynamic resource allocation, especially when handling critical tasks with strict Quality of Service (QoS) requirements and shifting priorities. This paper proposes a novel reinforcement learning-based resource management framework that enables priority-aware task scheduling and dynamic resource reallocation in edge environments. “Priority” refers to the categorization of tasks into critical, important, and general classes based on urgency and resource demands. “Reallocation” involves the preemptive reassignment of resources from lower-priority to higher-priority tasks to meet deadline constraints and improve system responsiveness. We design two Q-learning-based algorithms: Episode-Level Reallocation and Final Reallocation, which formulate the task allocation problem as a Markov Decision Process and learn optimal policies through interaction with the environment. These algorithms enable context-aware decisions that adapt to task priority and real-time system states. Simulation results demonstrate that our approach reduces rejection rates by up to 30% for critical tasks and achieves over 80% acceptance for high-priority workloads. The proposed method demonstrates effective trade-offs between resource efficiency and QoS under dynamic edge conditions, outperforming baseline Q-learning and non-prioritized strategies.</p>