The increasing complexity and dynamics of 5G mobile networks have brought revolutionary changes in its modeling and control, where efficient routing and resource allocation strategies become beneficial. Software-Defined Network (SDN) makes it possible to achieve the automatic management of network resources. Relying on the powerful decision-making capability of SDNs, network association can be flexibly implemented for adapting to the dynamic of the real-time network status. In this paper, we first construct a jitter graph-based network model as well as a Poisson process-based traffic model in the context of 5G mobile networks. Second, we solve the problem of QoS routing with resource allocation based on queueing theory using a low computational complexity greedy algorithm, which takes finding a feasible path set as the main task and resource allocation as the auxiliary task. Finally, we design a QoS-oriented adaptive routing scheme based on Deep Reinforcement Learning (DRL) SPACE, which is a DRL architecture with parameterized action space, in order to find an optimal path from the source to the destination. To validate the feasibility of the greedy QoS routing strategy with resource allocation, we make a numerical packet-level simulation to model a M/M/C/N queuing system. Moreover, extensive simulation results demonstrate that our proposed routing strategy is able to improve the traffic’s QoS metrics, such as the packet loss ratio and queueing delay.