On November 18, 2025, Cloudflare’s network experienced a significant, multi-hour outage that resulted in widespread error pages for users attempting to access customer websites. The company confirmed the incident, which began at 11:20 UTC, was not caused by a cyberattack or any malicious activity.
The disruption was triggered by an internal configuration change. A modification to permissions in one of Cloudflare’s database systems caused a query for its Bot Management service to generate duplicate entries. This, in turn, doubled the size of a critical “feature file” that is regularly distributed across the company’s global network.
The core software that routes traffic across Cloudflare’s network has a predefined size limit for this file to optimize performance. When the oversized file was propagated, it exceeded this limit, causing the software to crash. This failure in the core proxy system resulted in HTTP 5xx error codes being served to users.
Initial diagnostic efforts were complicated by the intermittent nature of the problem. As the underlying database change was being rolled out gradually, the system would alternate every five minutes between generating a correct file and the oversized, faulty one. This caused the network to repeatedly fail and recover, a pattern that initially led engineers to suspect a large-scale DDoS attack. A coincidental, unrelated outage of Cloudflare’s independently hosted status page further compounded these initial suspicions.
After identifying the true root cause, the team stopped the faulty file’s propagation. By 14:30 UTC, they had deployed a previously known-good version of the file, and core traffic flow was largely restored. Over the subsequent hours, engineers worked to mitigate increased load as services came back online, with all systems returning to normal operation by 17:06 UTC.
The outage impacted a range of services beyond the core CDN. Cloudflare Access saw widespread authentication failures, and Workers KV returned a high rate of errors. The failure of Turnstile and Workers KV also prevented most users from logging into the Cloudflare Dashboard. While email processing was unaffected, Email Security experienced a temporary reduction in spam-detection accuracy.
In a detailed analysis of the event, Cloudflare apologized for the disruption to its customers and the wider internet, calling the outage “unacceptable.” The company has begun work to harden its systems against similar failures in the future. Planned improvements include treating internal configuration files with the same stringent validation as user-generated input, enabling more effective global kill switches for features, and conducting a comprehensive review of failure modes across all core proxy modules.
Source link
