Global Database for Serverless and Edge

Upstash has launched a Global Database designed to optimize low-latency read operations for serverless and edge computing environments by deploying replicas across multiple global regions. This database utilizes a single leader replication model where a leader replica processes write requests and propagates them to backup replicas, with a gossip-based communication protocol managing cluster membership and failure detection. Although this setup provides weak consistency and may result in stale reads, it enhances read scalability by allowing any replica to handle read requests. The system employs a last-write-wins algorithm to resolve conflicts arising from network partitions, and learner replicas, which are read-only and unsuitable for leadership, are used to expand edge locations without causing write conflicts. Currently, the focus is on minimizing read latency rather than handling write-heavy loads, with plans to improve write latency and expand the number of regions in response to user feedback.