When will the next Solana mainnet-beta cluster restart occur?

The broken base rate and exact event criteria. The historical base rate of roughly one restart-triggering incident per year (2022–2024) has clearly broken down. Solana mainnet-beta has maintained roughly 30 months of uninterrupted uptime since the last manual restart on 2024-02-06 2 sources, with 100% cluster uptime reported through the summer of 2026 2 sources. Importantly, the August 2026 TeraSwitch routing failure does not qualify as a restart event under the current criteria. Although a single ASN fault knocked ~90 validators offline, affected stake peaked at 28.83%—safely below the 33.34% finality-loss threshold—meaning the network continued to produce and finalize blocks without a consensus halt 3 sources.

Structural improvements driving down the baseline hazard. The forward baseline hazard is substantially lower today due to mature structural mitigations. The spam and resource-exhaustion failure modes that drove early halts have been largely neutralized by QUIC, stake-weighted QoS, and localized priority fee markets solanacompass.com. Furthermore, client diversity is now a functional reality: the independent Firedancer client currently commands roughly 11.64% of network stake solanacompass.com. While an Agave-only bug could still halt the chain given its ~88% stake dominance solanacompass.com, Firedancer introduces partial protection. Additionally, Anza's deployment of the wen-restart feature in Agave 2.3 automates cluster-restart coordination anza.xyz. If a future crash does occur, automated recovery could effectively prevent the event from meeting the strict "manual, network-wide cluster restart" definition.

The dominant near-term hazard: Alpenglow. The primary near-term vulnerability is the imminent Alpenglow consensus overhaul, replacing Tower BFT with Votor and Rotor. Scheduled for activation around September or October 2026 3 sources, this represents the largest live consensus change in Solana's history. Precedent warrants caution: a May 2026 community-cluster test migration failed due to a consensus bug and peer-banning issues, requiring a manual relaunch 2 sources. While Anza has implemented heavy mitigations—including a parallel Tower BFT/Votor transition period that defaults to a known-good state without supermajority confirmation techtimes.com—the sheer scale of the change introduces a distinct risk bulge over the next 6–9 months. This elevated transitional risk pulls the 10th percentile date into early spring 2027.

Long-term trajectory and resolution dynamics. Assuming the network safely navigates the Alpenglow burn-in period, the hazard rate should decay meaningfully through the late 2020s. Post-upgrade, Alpenglow's "20+20" model raises non-adversarial fault tolerance from ~33% to ~40% offline stake solanacompass.com, providing deeper insulation against future infrastructure-concentration events. Coupled with the expected growth in Firedancer adoption, the long-term annual hazard drops low enough to push the median estimate into early 2031. Because there is a substantial probability that ongoing protocol maturation and automated recovery will completely avert any further manually coordinated restarts, a significant point mass rests on the "no restart ever" branch, anchoring the 90th percentile firmly at the 2036-08-16 cap.