The cloud-native computing is a promising solution for optimizing mobile networks, as it can improve the networks’ flexibility and scalability. However, it also brings new challenges to resource allocation, such as real-time performance and a smaller portion of the allocation. In this article, we develop a cloud-native network architecture which enables dynamic and flexible resource allocations. To address the challenges of real-time and fragmented allocation, we propose a resource allocation algorithm based on deep reinforcement learning towards a 6G wireless network. The proposed algorithm monitors the state of the whole network and trains the allocation policy, which can optimize the utility while meeting service level agreements of the network slices. We further validate the algorithm’s performance in the simulation environment and the experimental cloud-native network testbed. Furthermore, we highlight a suite of open research challenges in resource allocation in the cloud-native network towards 6G.