Cloudflare 跨 330 数据中心强一致共识
推荐指数 59.0 NO. 016 · 2026.07.09
发布2026/07/08Score158Comments36
为什么值得看
Cloudflare 开源 Meerkat，一个能在全球广域网故障环境下保证强一致性的分布式共识系统。对需要在多地域部署控制面服务的 AI 基础设施团队有直接参考价值。
编辑判断
之前业界处理跨地域强一致主要靠 Spanner 的 TrueTime 或 CockroachDB 的混合逻辑时钟，但两者都假设时钟同步相对可靠。Meerkat 的核心差异是彻底放弃对时钟的依赖，用纯异步消息和版本向量解决分区问题，这在海底电缆频繁中断的场景下更务实。
论文里提到写入延迟中位数 180ms、P99 约 1.2 秒，对于控制面配置下发是可接受的，但显然不适合数据面高频写入。如果你在做多区域模型服务路由或联邦学习调度层，Meerkat 的故障模型比 etcd 或 ZooKeeper 的 leader-election 假设更贴近真实公网环境，值得读一下它的安全证明部分。
代码未完全开源，但论文足够详细，预计半年内会有社区 Go/Rust 重写版本出现。
社区反馈
意见分歧 35 条评论
核心争论：Meerkat 是 Paxos 的部署实践还是 Raft 的替代品，文章是否营销过度且过早发布
if youve ever fought a raft cluster on a bad network with leaders flapping, elections storming and latency spiking this genuinely doesnt seem that bad. i believe this will be very useful to those dealing with messy networks
This article is a bit hard for me to grasp the main ideas of because, given Cloudflare's requirements (e.g. no strong leaders), it immediately seems like they should be comparing to leaderless protocols like Paxos-class algorithms. Comparing to Raft and saying it's better because Meerkat is leaderle
Worth noting if it wasn't obvious from the article that Cloudflare did not develop QuePaxa. It's from an SOSP paper back in 2023[0]. The article is discussing what is the first known large-scale public deployment of the protocol. [0] https://dl.acm.org/doi/10.1145/3600006.36