# Cloudflare Research introduces Meerkat consensus experiment > Meerkat applies the leaderless QuePaxa algorithm to small, strongly consistent control-plane state across Cloudflare’s global network. Canonical URL: https://www.devobs.io/news/news-cloudflare-meerkat-quepaxa-consensus/ By: Amara Okafor Published: 2026-09-06T11:58:54.640Z Updated: 2026-09-06T11:58:54.640Z Event date: 2026-07-08 Section: Infrastructure Cloudflare Research introduced Meerkat on July 8, an experimental consensus service built around the QuePaxa algorithm. The [research post](https://blog.cloudflare.com/meerkat-introduction/) says the project targets small pieces of strongly consistent control-plane state, such as leadership records for replicated databases, across Cloudflare’s geographically distributed network. ## Writes do not depend on one elected leader Cloudflare contrasts QuePaxa with leader-based protocols whose write path can pause while a failed or unreachable leader is replaced after a timeout. In Meerkat, every replica can receive reads and writes, and each replica connects to every other member of its cluster. Applications such as a transactional key-value store or leasing system can be layered on the agreed log. Developers specify eligible data centers and Meerkat places replicas. The intended data is written infrequently but requires consistency and fault tolerance. Cloudflare reports proofs of concept with as many as 50 replicas around the world, including tests in which leader roles repeatedly failed without an increased error rate. The system was not deployed to production at publication and was planned to remain internal in the immediate future. ## An experiment needs adversarial evaluation Wide-area consensus should be tested against asymmetric partitions, delayed messages, replica restarts, and simultaneous membership changes. Availability claims need to specify which replicas can communicate and whether the tested operation was a read or write. The project’s use of formal verification and deterministic simulation, both described as continuing work, is well matched to failures that are difficult to reproduce in live networks. Operators following the research should focus on the protocol assumptions and measured failure cases instead of treating a successful prototype as production validation. Meerkat is a substantive exploration of leaderless global consensus, with its experimental status and narrow initial workload clearly defined. ## Source references -