Traditional workload-based task scheduling has been well studied in edge computing. However, computing tasks with different types may have different sensitivities of processing latency to memory and CPU resources, which makes it challenging to design an efficient task scheduling strategy for edge servers with different configurations to guarantee the quality of service (QoS). To address this challenge, we first conduct the extensive testing on the processing latency of memory and CPU resources in the multi-container environment, and formulate the task scheduling problem as a long-term QoS optimization problem. Secondly, we transform the original problem into a task scheduling sub-problem for each time slot based on Lyapunov optimization theory, and propose an online task scheduling strategy using Genetic Simulated Annealing Algorithm, which can obtain an approximate optimal solution that minimizes makespan while ensuring the queue stability. Thirdly, based on matching theory, we use the backlog tasks information to update the approximate solution and further minimize makespan in real-time. Finally, the simulation results show that the proposed scheme can minimize the makespan comparing to other schemes.