Solana came within 4.5 percentage points of a network-wide halt, yet its uptime still shows 100%
🚨 行情变了群里说
At 3:43 a.m. on August 12, a bad route at a data center in Miami knocked about 90 validators offline at the same time. These validators controlled roughly 29% of all staked SOL. Block finalization, normally 13 seconds, stretched to as long as 30 minutes. But the chain didn’t stop. Blocks kept being produced, so every uptime chart showed it as an ordinary day 🦖
Metrika’s postmortem revealed the truly alarming numbers: 28.83% of staked SOL was offline. The network-wide finalization halt threshold was 33.34%. That left a gap of just 4.5 percentage points. Another 4 to 5 points offline, and Solana would have needed a coordinated restart. The outage spread from Miami to London, Amsterdam, Frankfurt, Singapore, and Tokyo. Most affected validators’ backup systems never kicked in.
During that half hour, the share of empty blocks surged from less than 1% to over 32%. Non-voting transactions plunged from 1,100–1,300 per second to fewer than 300. No block finalized for half an hour. The backlog was later cleared all at once, with an average of 1,600 seconds from production to confirmation. Your money was still in transit, and no one could tell you when the transaction would become irreversible ⚠️
Two months later, this concentration had barely changed. Fidelity’s FCAT report on October 16 showed that the largest hosting network still accounted for about 29% of stake. Meanwhile, since May 1, the Solana Foundation has required validators to use hosting networks with no more than 25% of stake and data centers with no more than 15%. The problem is that Foundation delegations cover only about 51% of validators, while accounting for just around 4% of active stake. The rules reach many validators, but have little influence.
More subtly, Alpenglow went live on October 21, aiming to cut finalization time from about 12.8 seconds to around 150 milliseconds. That’s fast—but it doesn’t change a thing about where validators run or which networks connect them. At the same time, about 63% of stake runs on Jito software, which still shares the underlying Agave code. A single bug could hit a huge swath of the network at once.
My take: what this incident really exposed isn’t whether Solana can go down. It’s that our definition of an outage is outdated. The chain still producing blocks doesn’t mean your transaction can settle. For people handling payments, clearing, or collateral pricing, 30 minutes of uncertainty is no different from a chain halt 📉
Next time you see a blockchain bragging about 100% uptime, will you ask about uptime first—or finalization? 💬
Every day, I bring you the latest in crypto infrastructure—not just what’s happening, but the logic and opportunities behind it 👀🚀
Tap the profile picture to watch the livestream
🚨 行情变了群里说
At 3:43 a.m. on August 12, a bad route at a data center in Miami knocked about 90 validators offline at the same time. These validators controlled roughly 29% of all staked SOL. Block finalization, normally 13 seconds, stretched to as long as 30 minutes. But the chain didn’t stop. Blocks kept being produced, so every uptime chart showed it as an ordinary day 🦖
Metrika’s postmortem revealed the truly alarming numbers: 28.83% of staked SOL was offline. The network-wide finalization halt threshold was 33.34%. That left a gap of just 4.5 percentage points. Another 4 to 5 points offline, and Solana would have needed a coordinated restart. The outage spread from Miami to London, Amsterdam, Frankfurt, Singapore, and Tokyo. Most affected validators’ backup systems never kicked in.
During that half hour, the share of empty blocks surged from less than 1% to over 32%. Non-voting transactions plunged from 1,100–1,300 per second to fewer than 300. No block finalized for half an hour. The backlog was later cleared all at once, with an average of 1,600 seconds from production to confirmation. Your money was still in transit, and no one could tell you when the transaction would become irreversible ⚠️
Two months later, this concentration had barely changed. Fidelity’s FCAT report on October 16 showed that the largest hosting network still accounted for about 29% of stake. Meanwhile, since May 1, the Solana Foundation has required validators to use hosting networks with no more than 25% of stake and data centers with no more than 15%. The problem is that Foundation delegations cover only about 51% of validators, while accounting for just around 4% of active stake. The rules reach many validators, but have little influence.
More subtly, Alpenglow went live on October 21, aiming to cut finalization time from about 12.8 seconds to around 150 milliseconds. That’s fast—but it doesn’t change a thing about where validators run or which networks connect them. At the same time, about 63% of stake runs on Jito software, which still shares the underlying Agave code. A single bug could hit a huge swath of the network at once.
My take: what this incident really exposed isn’t whether Solana can go down. It’s that our definition of an outage is outdated. The chain still producing blocks doesn’t mean your transaction can settle. For people handling payments, clearing, or collateral pricing, 30 minutes of uncertainty is no different from a chain halt 📉
Next time you see a blockchain bragging about 100% uptime, will you ask about uptime first—or finalization? 💬
Every day, I bring you the latest in crypto infrastructure—not just what’s happening, but the logic and opportunities behind it 👀🚀
Tap the profile picture to watch the livestream