Cloudflare全球故障，搞瘫了半个互联网！

Core Points - A significant outage occurred at Cloudflare on November 18, 2025, affecting major internet services globally, including ChatGPT, X (Twitter), and Spotify [1][13]. - The incident is described as a notable event in the history of internet disasters, warranting detailed documentation [2]. Incident Timeline - At 19:05, Cloudflare engineers deployed a change related to ClickHouse database access control [5]. - The change took effect at 19:28, initiating the outage [6]. - By 22:24, the team stopped generating new error configurations and rolled back to the previous stable version [7]. - The core outage lasted approximately 3 hours, with full recovery taking about 6 hours [8]. Impact and Scope - The outage had a global impact, affecting nearly half of internet services, including social media, AI platforms, online tools, and gaming services [13]. - Users experienced various errors, such as 500 errors and "Internal Server Error" messages, particularly noticeable during peak usage hours in China [15]. Technical Details - The root cause was identified as an internal database permission change that triggered a latent bug, leading to abnormal growth in bot management configuration files and subsequent software crashes across global nodes [8][14]. - The Cloudflare team began investigating the issue between 19:32 and 21:05, with the core problem identified by 21:37 [8]. Service Level Agreement (SLA) and Compensation - Cloudflare has not yet announced a compensation plan, but it offers SLA credit for Business and Enterprise plan customers if availability falls below 99.9%, which could result in a partial refund for the outage duration [19].