Cloudflare Research has announced Meerkat, an experimental global consensus service designed to provide strongly consistent, fault-tolerant coordination across geographically distributed systems. Meerkat is built on a novel consensus algorithm called QuePaxa, which aims to overcome the latency limitations of traditional consensus protocols like Paxos and Raft in wide-area networks.
- New consensus algorithm: QuePaxa is the core innovation behind Meerkat. Unlike classical algorithms that require multiple rounds of communication across all nodes, QuePaxa leverages a fast path that commits operations in as few as one round trip under favorable conditions, significantly reducing latency.
- Global scope: Meerkat operates across multiple Cloudflare data centers worldwide, simulating a true global consensus deployment. This contrasts with most consensus systems that are limited to a single region.
- Strong consistency model: Meerkat provides linearizable operations, meaning all nodes see the same order of operations—critical for building distributed databases, locks, and coordination services.
- Fault tolerance: The service tolerates up to <n/3> failures (with
nnodes), assuming a Byzantine fault model for safety and crash fault model for liveness—a pragmatic hybrid approach. - Application focus: Cloudflare plans to use Meerkat as the foundation for a distributed key-value store (
meerkat-kv) and other strongly consistent services, initially for internal use.
For developers, Meerkat represents a step toward globally scalable consensus that could enable truly distributed databases, asset locks for CDNs, or cross-region coordination with low latency. The introduction of QuePaxa may influence future consensus protocol design, especially for edge computing scenarios where minimizing wide-area communication is crucial. While still experimental, Meerkat's approach could simplify building fault-tolerant, globally distributed applications without sacrificing correctness.