Abstract: Addressing the issue that existing container orchestration technologies are constrained by the Pod architecture, which makes it difficult to achieve fine-grained and unified scheduling of network and computing resources, thus failing to meet the high requirements of Industrial Internet of Things. This paper first proposes a container level network resource management mechanism, which achieves fine-grained network resource management of Kubernetes cluster systems, and designs a dynamic programming based multi-time slot task aggregation algorithm. By packet capacity and delay constraints, the optimal packet aggregation scheme can be obtained, thereby minimizing the number of packets and further improving network resource utilization efficiency. Subsequently, in order to reduce the average latency of tasks in Kubernetes cluster systems and improve task success rates, this paper transforms the task scheduling and network-computing resource allocation problem into a Markov decision process, and proposes a deep reinforcement learning based network-computing resource collaborative scheduling algorithm based on this. The algorithm integrates Deep Deterministic Policy Gradient (DDPG) and Dynamic Programming (DP), enabling effective handling of high-dimensional continuous action and state spaces. Simulation results demonstrate that the proposed DDPG+DP algorithm can significantly reduce the average delay cost within the system and achieves a higher task completion rate compared to existing task scheduling and resource allocation methods.