Cloudflare Introduces Meerkat: A Leaderless Consensus Service for Global Control-Plane State
Migrate control‑plane state to Meerkat to avoid leader unavailability and achieve strong consistency across all data centers.
Migrate control‑plane state to Meerkat to avoid leader unavailability and achieve strong consistency across all data centers.
Summary
Cloudflare has launched Meerkat, an experimental consensus service built on the QuePaxa algorithm, to manage control‑plane data across its 330+ global data centers.
Unlike Raft, QuePaxa allows all replicas to write at all times and eliminates leader‑based timeouts, ensuring that writes remain available even when some data centers or links fail. Meerkat guarantees strong consistency (linearizability) and fault tolerance for any majority of alive replicas, and it supports applications such as a transactional key‑value store and leasing system.
The service is currently internal‑only and designed for small pieces of control‑plane state, such as leadership information for replicated databases. Developers can request a cluster of replicas, specify allowed data centers, and interact with the cluster via get/put requests.
Meerkat represents the first industrial deployment of QuePaxa at global scale, providing a robust foundation for future control‑plane services that require consistent state across a wide‑area network.
Key changes
- Meerkat uses QuePaxa, allowing all replicas to write without a leader
- No timeouts; progress never halted, improving availability
- Guarantees linearizability and fault tolerance for any majority of alive replicas
- Supports transactional KV store and leasing applications
- Designed for small control‑plane state like leadership info
- Developers request replica clusters and can specify data center placement
- Internal‑only experimental deployment at global scale
- First industrial deployment of QuePaxa worldwide