With the rapid development of mobile edge computing, more and more tasks require substantial communication and computing resources, which makes resource management in Industrial Internet of Things (IIoT) face the challenge of resource skew. Resource skew can affect the efficiency of task execution, as overloading of edge servers can significantly degrade the system performance. To address this issue, an Online-Learning based Task Scheduling framework is proposed to optimize task scheduling. The framework can accurately sense the task and edge server states in the network and select the optimal allocation scheme, thereby reducing the total energy consumption of task execution. By modeling the optimization problem as a Markov Decision Process (MDP), an Online-Learning based Task Scheduling algorithm, combining with the Upper Confidence Bound (UCB) strategy to enable efficient decision-making in task scheduling, is designed to provide a effective solution. Experimental results show that the proposed method is superior to the compared methods in terms of delay and energy consumption.