Cloudflare picks leader-optional QuePaxa algorithm to keep its global Meerkat control plane moving
Cloudflare says its experimental Meerkat consensus service runs on QuePaxa, a 2023 academic algorithm that does not depend on a single leader, to avoid stalls on volatile wide-area networks. The performance figures come from Cloudflare's own testing and have not been independently verified.
Cloudflare says Meerkat, an experimental distributed consensus service for its internal global control plane, is built on QuePaxa. The company attributes the algorithm to researchers Tennage & Băsescu et al., who published it in 2023. Cloudflare described the choice in a blog post on 11 November. [1]
Most consensus systems use Raft, Multi-Paxos or Viewstamped Replication. Cloudflare argues these depend on one leader being reachable and responsive. When that leader fails or slows down, the cluster stalls until a replacement is elected. The company says this is a poor fit for Meerkat, whose replicas may sit on different continents and be linked by high, variable latency. [1]
Cloudflare cites its own history as motivation. In 2020, it says, a partially failed core network switch degraded a consensus-backed system. Its dashboard then ran about 80 times slower than normal for six hours and could not accept writes, although no data was lost. [1]
According to the post, QuePaxa does not rely on fault detection and treats the leader as optional. Any proposer can submit proposals. These carry random priorities, and proposers adopt the highest-priority proposal they see instead of blocking one another. The leader always holds top priority and is the only proposer that can decide in a single round trip. [1]
Cloudflare reports that in its own evaluations, with servers on nearly every continent, Meerkat never fully stalled. It says more than 95% of decisions were made by the leader in the first phase. When that fast path failed, throughput fell to as little as one third of the normal rate. [1]
The company acknowledges the trade-offs. QuePaxa has seen far less real-world use than Raft or Paxos, and Cloudflare is extending the algorithm's original crash-stop design to support crash recovery. It says it is formally verifying its Rust implementation and running simulation and fault-injection tests. It also says benchmarks will follow in later posts. [1]
Sources
- Cloudflare: Choosing QuePaxa, an unusual consensus algorithmSource published · TechDots found it
Statements labelled as a company's claim have not been independently confirmed by TechDots.
Get Cybersecurity stories by email
Related coverage
AWS reworks Well-Architected Framework, adding secure coding and backup resilience guidance
AWS says its updated Well-Architected Framework splits environment-wide guardrails from per-workload reviews, adds secure coding and ransomware-resilient backup practices, and trims workload best practices from 307 to 216 to ease automated reviews.
AWS previews Infoblox threat intelligence rules inside Route 53 Resolver DNS Firewall
AWS says Infoblox Threat Defense is now in preview in Route 53 Resolver DNS Firewall, letting customers subscribe via AWS Marketplace to partner-managed DNS threat feeds. The Marketplace fee is waived during preview, but the rules require the paid DNS Firewall Advanced option.
Cloudflare warns Quad9's exit from France could leave unaware DNS users offline
In a blog post, Cloudflare says Quad9 is pulling its free public DNS resolver out of France after a court imposed daily penalties for not blocking piracy-linked domains. Cloudflare, Google and Cisco are appealing a similar Belgian order with a January 2027 deadline.