Single-master is the mainstream architecture of cloud-native databases.For Single-master databases,reads can be spread across the cluster,while updates must be handled by one single master node. To meet the demands of large-scale transaction processing, some of the cloud-native databases attempt to further enable transaction execution across multiple nodes. One possible approach is to introduce shared cache.It enables cross-node data access and leads to a multi-master database based on shared caching. For shared-cache databases,the overhead of remote access is significantly higher than local access.Therefore,coherence protocol is the crucial factor that affects system performance and scalability. In this paper,we present PG-RAC,a multi-master shared-cache database based on PostgreSQL.This paper proposes two novel improvements to the coherence protocol.First,we propose chained routing strategy,which distributes the routing information acorss nodes in cluster. Compared to the routing strategy that utilizes centralized directory ,it reduces the average transaction latency by approximately 20%.Second,PG-RAC enhances the invalidation mechanism. PG-RAC separates invalidation operations from the transaction path, reducing the latency of the critical path in transaction processing. Additionally, based on the features of MVCC,PG-RAC proposes delaying the invalidation point,which effectively improves cache utilization.TPC-C benchmark results show that,for a cluster with 4 compute nodes, the throughput is 2× that of a single-node PostgreSQL and 1.5× that of the distributed database Citus.