# Cloudflare Adopts QuePaxa for Global Distributed Consensus Moving beyond traditional leader-based algorithms to improve stability in volatile wide-area network environments. - **Source:** [Cloudflare Blog](https://blog.cloudflare.com) - **Published:** 2026-11-11 - **Topic:** [technology](https://brifing.si/c/technology) - **Original article:** - **On Brifing:** https://brifing.si/s/9223c496-0704-42cc-a04c-256632d5adf2 ## The Meerkat Control Plane Cloudflare introduced Meerkat, an experimental distributed consensus service designed to manage their internal global control plane. This system ensures replicas agree on an ordered log of events, maintaining a consistent key-value state across continents. Unlike local systems, Meerkat operates over the public internet, facing significant challenges regarding network latency and connectivity volatility. ## Limitations of Traditional Consensus Most consensus systems rely on established algorithms like Raft or Paxos, which typically utilize a single leader process. These models perform well in stable, local networks but struggle in global deployments. If a leader experiences latency or disconnects, the entire system stalls, leading to unacceptable downtime for Cloudflare's globally distributed infrastructure requirements. ## The Destructive Interference Problem In leader-based protocols, destructive interference occurs when multiple proposers compete, blocking each other from reaching a decision. Even when a leader fails, the recovery process often triggers new elections where candidates compete for majority support. This can create metastable failures, where the system remains unavailable long after the initial network issue is resolved. ## Fault Detection Challenges Fault detection in wide-area networks is inherently error-prone because asynchronous networks lack guaranteed message arrival times. Systems cannot distinguish between a crashed server and a slow network link. Consequently, conservative timeouts are often set high to avoid false positives, which significantly slows down recovery times when actual failures occur within the cluster. ## Introducing the QuePaxa Algorithm To address these issues, Cloudflare adopted QuePaxa, a 2023 consensus algorithm designed for asynchronous environments. QuePaxa avoids the pitfalls of strict leader reliance and precise timing requirements. By design, it maintains stability despite latency spikes and connectivity issues, making it a superior choice for the unpredictable nature of global wide-area network deployments. ## Core Design of QuePaxa QuePaxa functions through two primary roles: proposers and recorders. Proposers submit events, while recorders act as intermediaries that track proposals. Unlike traditional systems where proposers compete, QuePaxa encourages cooperation. Proposers assign random priorities to their submissions, ensuring that the system collectively converges on the highest priority proposal known to the active participants. ## Constructive Proposer Cooperation In QuePaxa, if a proposer receives a proposal with a higher priority than its own, it adopts that proposal. This mechanism prevents proposers from locking each other out indefinitely. By shifting from competitive to constructive interference, the algorithm ensures that the system makes progress even when network conditions are hostile or highly variable. ## Optional Leader Advantage While QuePaxa supports an optional leader, it does not mandate one for system progress. The leader is simply assigned an exclusive maximum priority to optimize performance under favorable conditions. This hybrid approach allows for single round-trip consensus when the network is healthy, while gracefully degrading performance rather than failing during network instability. ## Mitigating Metastable Failures By removing the strict reliance on leader recovery, QuePaxa avoids the common trap of continuous, failing elections. Because the algorithm does not require a complex fault detection mechanism to trigger leadership changes, it bypasses the downtime associated with inaccurate timeout guesses. This results in a more resilient control plane for global operations. ## Global Scale Requirements Meerkat requires a system that can handle high and variable latency across continents. Traditional consensus algorithms are typically deployed in smaller, local environments with strict latency bounds. QuePaxa’s ability to function without assuming partial synchrony makes it uniquely suited for the global scale at which Cloudflare operates its internal control plane services. ## Performance Under Volatility QuePaxa’s randomized priority mechanism ensures that as more proposers submit proposals, the likelihood of convergence increases. This probabilistic approach is robust against the sudden bursts of latency common in wide-area networks. By prioritizing the highest known value, the system maintains a shared truth across all replicas without needing constant, perfectly timed communication. ## Future of Distributed Consensus The shift toward QuePaxa represents a move away from rigid, leader-centric architectures toward more flexible, randomized protocols. As global distributed systems become more common, algorithms that prioritize availability and stability over strict, synchronous timing will likely become the standard. Cloudflare’s implementation serves as a practical case study for this evolving architectural paradigm. ## Sources covering this story - https://blog.cloudflare.com/choosing-quepaxa/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._