A Distributed State Management Approach for FaaS Platforms

Function-as-a-Service (FaaS) has become increasingly popular due to its ease-of-use operations, scalability, and low management overhead. However, the stateless nature makes it inappropriate for deploying applications that require maintaining state data across series execution. This motivates state management for FaaS platforms, which is currently satisfied by integrating additional storage services or a central coordinator. In this paper, we propose a new approach for FaaS platforms to manage function states in a distributed way. Specifically, we utilize function-level checkpoint to maintain function state and enable state to scale with the number of function instances. To synchronize state among multiple instances of the same function, we employ a synchronization mechanism supported by consensus protocols. Compared to existing works, our new approach distributes the burden of state management across all nodes and does not rely on additional services. In evaluation, we compare our approach to baseline system with a micro-benchmark and three representative applications. The results show that our approach can keep strong consistency of state between functions and reduce function execution time significantly.