Most large-scale Peer-to-Peer (P2P) live streaming systems are constructed as a mesh structure, which can provide robustness in the dynamic P2P environment. The pull scheduling algorithm is widely used in this mesh structure, which degrades the performance of the entire system. Recently, network coding was introduced in mesh P2P streaming systems to improve the performance, which makes the push strategy feasible. One of the most famous scheduling algorithms based on network coding is R2, with a random push strategy. Although R2 has achieved some success, the push scheduling strategy still lacks a theoretical model and optimal solution. In this paper, we propose a novel optimal pull-push scheduling algorithm based on network coding, which consists of two stages: the initial pull stage and the push stage. The main contributions of this paper are: 1) we put forward a theoretical analysis model that considers the scarcity and timeliness of segments; 2) we formulate the push scheduling problem to be a global optimization problem and decompose it into local optimization problems on individual peers; 3) we introduce some rules to transform the local optimization problem into a classical min-cost optimization problem for solving it; 4) We combine the pull strategy with the push strategy and systematically realize our scheduling algorithm. Simulation results demonstrate that decode delay, decode ratio and redundant fraction of the P2P streaming system with our algorithm can be significantly improved, without losing throughput and increasing overhead.