Technical Error Correction Collective

Claim: Fastly broke the Internet

, by:

On a boring Tuesday in June a tech company called Fastly had a hiccup. Hiccups are nothing new nor surprising, especially in the tech industry, and usually go mostly unnoticed… unless they happen at what effectively is critical infrastructure for many thousands of websites, including some of the most popular ones.

In short, enough websites stopped working for long enough, that from the perspective of a regular user the Internet was, in a very real sense of the word, “broken.”

Fastly is a Content Delivery Network, or “CDN.” The whole reason for existence of CDNs, and their business proposition, is that they help websites handle immense traffic and deliver all bits and pieces a particular website requires (images, scripts, and whatever else) to a user’s browser as fast as possible. They do so by maintaining enormous infrastructure, spread around the world. Website administrators point their domains to the CDN’s endpoints or include content directly from them, and let CDNs handle the traffic from the location closest to the particular user, balancing the load across as many servers as needed.

The flip-side is that if a CDN goes down, all websites using it are affected.

Obviously, CDNs are supposed to be engineered in such a way that failures do not happen. But as surprising as it may be, they do, and quite regularly: Amazon AWS (which is often used as a CDN) had a large-scale failure in November 2020, taking down many thousands of websites including the Washington Post. CloudFlare, the biggest of the CDNs, experienced a major half-hour outage in July 2020. Both had outages in 2019 (Cloudflare in July, Amazon AWS in August), too.

And mere nine days after Fastly, Akamai (another huge company offering CDN services) had a fumble of their own, bringing down websites of banks and airlines.

To make matters worse, most websites are not engineered to withstand any CDN failures (as these are not supposed to happen), and a single website can use multiple different CDNs for it’s content and code. Thus, an outage of any one of them can cause a website to go down, just like removing one card makes the whole house of cards fall. The kicker is: many of the websites using a CDN, and affected by such outages, often don’t really need a CDN in the first place. (You’ll notice is not using any CDNs).

So, did Fastly break the Internet? Not single-handedly, no. But it is probably reasonable to say they knowingly benefited from (and promoted) industry practices that led to centralization and introduction of single points of failure to a network that was supposed to be decentralized and fault-tolerant.

“Companies don’t keep their own infrastructure or any redundant fallbacks, but expect a third party to handle that for them. Since there are very few such third parties, when one of them suffers a failure, a lot of websites fail as well, often starting a cascade of broken infrastructure,” notes experienced full-stack web developer Paweł Ngei. There are alternatives to centralized CDNs, but it’s up to the industry as a whole to decide to invest in them.