Cloudflare opened a major incident at 15:47Z on October 23, describing degraded performance between Ashburn and Atlanta, with latency, packet loss and intermittent errors hitting Workers and other services across the Eastern US. Twenty minutes later, at 16:07Z, the company identified the cause as an outage at a connectivity provider upstream of its own network, and said it was rerouting traffic. By 17:36Z, the incident was still marked identified, not resolved, after the upstream failure recurred and traffic engineering continued to mitigate packet drops.
A second, separate incident compounds the picture. At 15:27Z, Cloudflare reported delayed, missed and duplicated Workers Cron Triggers, a distinct problem from the Ashburn-Atlanta routing issue. That incident remained in investigating status as of its last update at 16:12Z, with no identified cause disclosed. Two open incidents touching scheduled jobs and live traffic on the same day is the kind of overlap that matters less for its coincidence than for what it reveals: multiple failure paths through infrastructure that customers do not operate and, in this case, do not own.
The decision this changes is how much buyers can treat an edge or serverless platform as a single accountable unit. Cloudflare's status page pins the root cause on a third-party connectivity provider, not on Cloudflare's own hardware or code. That distinction matters for anyone negotiating uptime commitments or reading a postmortem later, because liability and remediation often hinge on whether the failure was internal or inherited from a vendor's vendor. Cloudflare did not disclose which provider failed or why the failure recurred after initial rerouting, so the durability of the fix is untested as of the last update.
For procurement teams, the immediate checklist is narrower than it looks. Contracts should be checked for how service credits are defined when the proximate cause sits outside the primary vendor's own network, since Cloudflare's language points squarely at an upstream provider rather than internal infrastructure. Operations teams running cron-dependent workloads on Workers should treat the missed and duplicated trigger reports as a separate risk category from the Ashburn-Atlanta latency problem, since idempotency and retry logic for scheduled jobs may need rechecking regardless of how the routing incident resolves. Neither incident, as of the last posted updates, had a confirmed resolution time.
None of this changes switching costs overnight. Moving off Cloudflare's edge network remains expensive and slow regardless of a single afternoon's packet loss, and the company's quick identification of an external cause will likely be read internally as evidence the core platform held up. But buyers evaluating renewal terms now have a concrete data point: a named two-city path failed because of someone else's infrastructure, and the fix took hours, not minutes, to hold.