Cloudflare has released Meerkat, a distributed consensus service built by its Research team after roughly a year of development, aimed at keeping data consistent across more than 330 data centers worldwide.
The core problem Meerkat addresses is one of coordination under failure. Cloudflare's internal services share control-plane state across every data center on its network, which spans the full public Internet. Adverse conditions, including server outages, saturated queues, and severed cables, make it hard to guarantee that no two readers ever see conflicting state while still allowing writes when parts of the network go down.
Consensus algorithms solve this class of problem by getting multiple machines to reach agreement on an identical sequence of values. They remain correct as long as a majority of replicas stay alive and reachable. Raft, one of the most widely deployed such algorithms, assigns all write operations to a single designated leader and uses timeout mechanisms to detect failures. When that leader crashes or its network connection degrades, the entire system stalls until a replacement is elected, and Cloudflare said timeout values are difficult to tune correctly on networks where latency is unpredictable.
Cloudflare said it has experienced multiple outages tied directly to leader unavailability in consensus-driven systems. Meerkat runs on a consensus algorithm that differs from Raft, removing the single-leader bottleneck that caused those incidents.