The user’s quest for more extreme experience is promoting the aggregation of virtual network functions (VNFs) deployed on virtual machines or containers for providing services in the form of service function chain (SFC). The application of SFC relies on the SFC resource allocation (SFC-RA) scheme. In SFC-RA problem, multiple VNFs of the same type can share a VNF instance (VNFI), which also brings challenges to the online placement of SFCs of different lengths. However, existing studies either ignored the synergistic effects of resource constrains inherent in network environments, delay and service availability on deployment performance; or failed to perform well when faced with complex and highly dynamic problems; or were limited by the performance of deep reinforcement learning (DRL) algorithm. Therefore, how to achieve the online deployment of SFC with VNFI sharing is a challenge that needs to be solved. In this paper, we first formulate the above problem as a Markov decision process (MDP). We then propose an SFC-RA scheme of sharing VNFI based on Rainbow Deep Q Network (SV-RDQN) to optimize the deployment costs and delay, including VNFI sharing mechanism and reward buffer mechanism. Finally, we define the average value as the optimization goal to evaluate the quality of deployment policy for SFCs of different lengths. The experimental results demonstrate that the proposed method performs excellently in multiple evaluation metrics.