On the morning of June 8, 2021, a startling number of well-known websites returned errors at almost the same instant. Reddit, Amazon, the BBC, The Guardian, Twitch, the UK government’s gov.uk, and many others showed error pages instead of content. The common thread was Fastly, a content-delivery network that sits in front of those sites to cache and serve their pages from edge servers around the world. When Fastly faltered, everything behind it appeared to go down together.
Fastly’s own summary of the outage explains that the trigger was not a malicious attack or a routine hardware failure, but a latent software bug. A software deployment on May 12, 2021 had introduced a bug that, in Fastly’s words, “could be triggered by a specific customer configuration under specific circumstances.” For almost four weeks the bug lay dormant, harmless until exactly the right conditions appeared.
Those conditions arrived on June 8, when a customer pushed a valid configuration change that happened to match the precise pattern needed to wake the bug. According to Fastly, the change caused 85 percent of the network to return errors. The customer had done nothing wrong; their configuration was legitimate. The fault was entirely in Fastly’s code, which had shipped weeks earlier and was only now exercised by a real-world input no one had anticipated.
Fastly’s published timeline shows how quickly a global CDN can both fail and recover. The disruption began at 09:47 UTC, the company detected it within a minute, identified the affected configuration within roughly 40 minutes, and saw most services recovering by 11:00 UTC, about an hour and 15 minutes after the start. A permanent fix for the underlying bug was deployed later the same day.
The incident is a clean illustration of two ideas at once. First, latent bugs: defective code can sit in production indefinitely, passing every test, until some specific input finally triggers it, which is why a four-week gap separated cause from effect here. Second, concentration risk in the modern web: because so many independent sites share a handful of CDNs, a single edge provider’s bug can make a wide swath of the internet appear to fail simultaneously, even though the origin servers behind it are perfectly healthy.
Fastly committed to finding why the bug had not been caught in testing and to deploying the fix and detection improvements so that a similar latent fault would not slip through again, the kind of follow-up that turns an outage into a more resilient system.