Cloudflare is developing Meerkat, a distributed consensus service designed to manage control-plane state across more than 330 data centers. The system replaces traditional Raft-based implementations, which Cloudflare found unreliable due to their reliance on leaders and timeouts that cause unavailability during network degradation or crashes.

Meerkat implements Que Paxa, an algorithm from EPFL researchers that allows all replicas to perform writes simultaneously. This architecture eliminates the single-point-of-failure inherent in leader-based systems and removes the need for hard-to-configure timeout values in unpredictable wide-area networks. The service provides linearizability, ensuring that all reads following a write reflect that write, effectively treating global state like local memory.

Cloudflare uses this consensus log to build a transactional key-value store and a leasing system. These tools manage critical metadata, such as AI model instance placement and database leadership. By tolerating f faults in a system of 2f + 1 machines, Meerkat maintains availability as long as a majority of replicas can communicate, regardless of which specific node a client contacts.