Solana Mainnet Came Close to Finality Breakdown After Teraswitch Routing Incident

Solana's mainnet briefly approached a critical threshold this week after a routing failure at infrastructure provider Teraswitch knocked roughly 28.83% of staked SOL offline, CoinDesk reported. The chain kept producing blocks and did not halt, but the episode again underscored how heavily validators depend on a small set of infrastructure providers. Under Solana's operating rules, transaction finality stops if more than 33.34% of stake goes offline. Marinade Finance estimated the incident reached about 86% of that limit, leaving roughly 20 million SOL of additional offline stake that could have pushed the network into a full finality loss. About 90 validators were affected for around 33 minutes. Marinade estimated missed rewards at about 333 SOL, expected to be made whole via validator staking at the end of the current epoch. The disruption stemmed from Teraswitch edge routers and impacted multiple nodes across Europe and Asia. Certain sites in those regions incorrectly treated a misrouted path as the local default route and prioritized it over the correct path. Core infrastructure rejected the route as invalid, leaving 12 sites without valid forwarding paths. Affected locations included London, Amsterdam, Dublin, Frankfurt, Singapore, and Tokyo. No comparable outage was reported in North America. Teraswitch said it identified the issue roughly 10 minutes after it began and restored traffic around 04:16:15 UTC, though the Miami node was temporarily disconnected from its backbone during troubleshooting. Marinade's analysis pointed to staking concentration as a key amplifier. AS20326 was estimated to hold 118.89 million SOL, more than a quarter of all staked supply. Around the time of the incident, about 94% of that stake was simultaneously offline, accelerating the risk buildup. Marinade also observed roughly 14.1 million SOL offline in environments tied to Latitude.sh, Limestone, Butterfly Research, and Allnodes, though it said available data cannot confirm whether those outages shared a common dependency or were simply concurrent. Marinade reported that 59 validators controlling 80.2 million SOL returned within a similar recovery window across Amsterdam, Frankfurt, and Tokyo. The group did not appear to fail over automatically and instead came back only after routing reconverged. Among 74 validators Marinade could measure, only three recovered via redundant setups: Laine, Cogent Crypto, and Lion3d. Marinade also noted centralization in its own delegation program, where four ASNs account for two-thirds of delegated stake and AS395201 alone represents 36.94%. The mainnet remained operational throughout, while test environments saw interruptions. Jacob Creech, Solana Foundation Vice President of Technology, said that out of 699 staked validators, 597 continued voting, roughly six-sevenths of the set. He added that affected validators recovered within 40 minutes and that validators in the Foundation's planned delegation program were not impacted. Anza said Solana Devnet halted during the same incident but recovered automatically after routing was restored. Developers noted that Devnet's behavior may signal improved resilience, since it runs the same software as mainnet. Solana's last full mainnet outage occurred in February 2024 and required nearly five hours to restart. While this week's event did not progress into a full outage, it reinforced that concentrated infrastructure can allow localized routing failures to escalate quickly into broader network risk. Pyth was also affected. Teraswitch hosted two of Pyth's four NATS nodes, and the loss of those nodes temporarily caused NATS to lose quorum, taking the dependent Router service offline. Pyth said it will accelerate onboarding additional infrastructure providers and spread critical services across independent failure domains.