Major internet outages and what caused them

Big outages make headlines because so many people notice at once, but the causes are usually ordinary: a configuration change, a software bug, a certificate or a dependency nobody realised was shared. This guide summarises a few well-documented incidents and the lessons they share. The details come from the companies’ own published explanations and news reports, and exact times differ between sources, so treat durations as approximate.

Facebook, Instagram and WhatsApp, October 2021

On 4 October 2021 Facebook and its other services became unreachable for roughly six hours. The company explained that a routine maintenance command accidentally disconnected its data centres from each other and from the internet. Its DNS servers, which are designed to withdraw themselves when they cannot reach the data centres, then also stopped announcing themselves. The names of its sites effectively disappeared from the internet.

The recovery was slow because the same outage cut off many of the internal tools staff used to fix problems. Lesson: keep a way in to your systems that does not depend on the systems being repaired.

Fastly, June 2021

In June 2021 a large number of well-known websites returned errors for under an hour. A content delivery network called Fastly later said a valid customer configuration change triggered a bug in software that had been deployed earlier, causing a large part of its network to return errors. Fastly identified the cause quickly and most of the network recovered within the hour. Lesson: a latent bug can sit unnoticed until a particular input finds it, and when the provider is shared, the impact is wide. See what a CDN is and why outages spread.

The CrowdStrike update, July 2024

On 19 July 2024 a faulty content update for a widely used security product caused Windows computers running it to crash and fail to restart. Airlines, hospitals, banks and shops were affected. This was not a website outage at all, but it is worth including because it shows how one update delivered to millions of machines at the same time can cause a global problem. Lesson: staged roll-outs, where an update reaches a small share of machines first, limit the damage from a mistake.

Amazon Web Services, October 2025

On 20 October 2025 a problem in AWS’s US-EAST-1 region disrupted a very large number of services, including apps and websites that had nothing obviously to do with Amazon. Amazon’s explanation pointed to a fault involving DNS resolution of the endpoints for one of its database services, which then spread to other internal services that depended on it. Recovery took many hours, and the surge of retries from systems trying to reconnect made it harder. Lesson: a fault in one internal service can cascade, and a rush of automatic retries can prolong an outage.

Cloudflare, November 2025

On 18 November 2025 Cloudflare had a global outage that returned errors for many of the sites it protects. The company’s published account said a change to database permissions caused a configuration file used by its bot-management system to roughly double in size. The software that read the file had a size limit, and it failed when the file grew past it. The incident was caused by an internal change, not by an attack. Lesson: a harmless-looking internal change can break software that makes assumptions about its inputs, and shared infrastructure turns that into a worldwide event.

The patterns that repeat

  • It is rarely a hack. Most large outages come from a configuration change, a software bug or an expired certificate rather than an attack.
  • Shared layers amplify it. DNS, CDNs, cloud regions and identity services sit under many sites, so one fault looks like many.
  • Fixing it can be harder than breaking it. If the tools used to repair a system depend on that system, recovery slows down.
  • Retries make things worse. When a service comes back, a flood of automatic reconnection attempts can knock it over again.
  • Recovery is uneven. Caches, connections and queues clear at different speeds, so some people see the site return before others.

What this means for you

If many unrelated sites and apps are failing together, it is almost certainly a shared provider and nothing is wrong with your device. Check the provider’s status page, look at our outage map for clusters of reports, and wait. If it is only one site, work through our checklist.

You can also browse the outage history on this site to see which services have had problems on which days, based on our own checks and visitor reports.

What owners can learn

  • Roll changes out gradually and have a fast way to undo them.
  • Test what happens when a dependency fails, not only when everything works.
  • Back off and add randomness to automatic retries so recovery does not become a second outage.
  • Keep emergency access and communication tools independent of the systems they are meant to repair.
  • Tell customers early, plainly and often. Silence feels worse than bad news.

These summaries are simplified. For full technical detail, read each company’s own post-incident write-up.

Check a website now

Is it down for everyone, or just you? Paste an address and we test it from fifteen places.

Open the checker

Related guides

Last updated 2 November 2026. Written by the siteupdown.com team. Spotted a mistake?

← All help guides