Cyclic Queuing and Forwarding (CQF, a.k.a IEEE 802.1Qch) has been proposed to satisfy the demands of predictable deterministic latency only related to the number of hops. However, IEEE 802.1Qch only introduces the principle and working mechanism of CQF. There are still many open problems worth exploring to make CQF practical in real dynamic application scenarios. To achieve deterministic end-to-end latency by efficiently allocating the real flows with limited bandwidth resources, this paper proposes a reinforcement learning based joint routing and scheduling algorithm. We transform the CQF into formulated model and orchestrate path in consideration of hardware resources and latency requirements to accommodate more scheduled flows. Experimental results show that the proposed algorithm can achieve good performance in transmission stability and latency certainty than RIP.