When a consensus leader dies inside a wide-area network, the whole system can go dark for writes until a new one is elected. Cloudflare, which runs services across more than 330 data centers strung along the unpredictable backbone of the public internet, says that failure mode has already caused multiple real incidents. Its answer, introduced July 8, 2026 in a research post, is a new internal consensus service called Meerkat that is built on an algorithm named QuePaxos — one where every replica can accept writes at all times and no timeout ever freezes progress [Cloudflare Research: Introducing Meerkat].
This is not a consumer feature you can sign up for. Meerkat is explicitly described as experimental and “internal-only for the immediate future,” initially tasked with managing small, critical pieces of control-plane state such as which machine currently owns the right to write to a replicated database [Cloudflare Research: Introducing Meerkat]. But the design choice behind it speaks to a class of distributed-systems problem that any engineer running state across regions runs into, and Cloudflare claims this is the first industrial deployment of QuePaxos at global scale [Cloudflare Research: Introducing Meerkat].
Why Raft fights back against global networks
Cloudflare’s own description of the pain is unusually concrete. Many of its internal services need to read and modify the same control-plane state from all over the world, and they need two guarantees: that different readers never see inconsistent state, and that the system stays available for writes even when some data centers or links fail. The internet makes that hard on purpose — servers and data centers go down, queues fill up, and undersea links get cut [Cloudflare Research: Introducing Meerkat].
The widely deployed alternative, Raft, puts a single leader in charge of all writes. If that leader crashes or the network degrades around it, the system stops accepting writes until another replica times out and a new leader is elected. Worse, Cloudflare notes the timeout values that trigger that failover are genuinely difficult to tune on a network whose latencies swing unpredictably — and it has lived through several outages caused precisely by unavailable leaders [Cloudflare Research: Introducing Meerkat].
What QuePaxos changes
The difference is structural. QuePaxos, published in 2023 by Tennage and Basescu et al., lets all replicas perform writes at all times, and progress is never halted by a timeout [Cloudflare Research: Introducing Meerkat]. For a network that spans the entire planet, that property matters more than leader efficiency does. Cloudflare layers ordinary applications — a transactional key-value store and a leasing system — on top of Meerkat’s consensus log [Cloudflare Research: Introducing Meerkat].
The consistency target is linearizability: operations appear to execute in the exact real-time order they occurred, which lets programmers reason about the system as if it were a single box rather than a fragile mesh. Cloudflare is explicit that many of its services want exactly this strongest level, because weaker models force developers to reason about every weird reordering a replica might show [Cloudflare Research: Introducing Meerkat].
The honest caveats
This is a research preview, not a generally available product. Cloudflare says Meerkat is still in development, scoped at first to tiny control-plane records rather than user data, and the post is framed as groundwork for a series of follow-ups [Cloudflare Research: Introducing Meerkat]. Anyone hoping to point their own stack at it should treat the July 8 writeup as a design disclosure, not a download.
The more durable takeaway is about the trade. Leader-based consensus buys you simplicity and a clear source of truth; it costs you availability the moment the leader blinks on a flaky link. Always-writable consensus keeps the writes flowing but shifts the hard work into conflict resolution and ordering. For a company whose “data center down” event is not a hypothetical but a Tuesday, that trade clearly points toward the latter [Cloudflare Research: Introducing Meerkat].
Meerkat is worth watching not because it will show up in a dashboard near you soon, but because it is a real, named attempt to run a harder consensus algorithm at the scale most of us only read about — and the incident-driven motivation Cloudflare published is the kind of field report the rest of the distributed-systems world reads closely.
