Internet broke? Go for a walk

Tuesday evening. Surfing the web. Noticed 5xx errors. Lots of them.

My self hosted stuff is behind Cloudflare. Almost everything that I run. So when Cloudflare goes down, they all go down too.

Since I used to work there, I text a few people still inside. They’re in the middle of it, phones blowing up.

Nothing much I could do. Laughed at a few memes.

Go out for a walk. Touch grass.

Fresh air. No screens. No error messages. No frantic messages.

It’s actually nice.

What that evening looks like from outside

The errors were not on one site. They were everywhere at once, because the web has a load-bearing company now. Cloudflare fronts a slice of the internet large enough that when it fails, pages you have never thought about fail with it: the sites you read, their images, the scripts they load, the login flows they delegate. A site can avoid Cloudflare itself and still go dark when three of its dependencies did not make that choice. Concentration does that. Every operator who picked the same provider made a rational choice, and the sum of those choices is a single point of failure that none of them controls alone.

For a bystander there is nothing useful to do at 8pm on a Tuesday, and that is the actual lesson of the evening. The people inside own the incident: they have the runbooks, the dashboards, the pager. Your own move was made months earlier, in architecture. Static generation, this site is a folder of HTML built before anything fails, means an edge outage is bad luck instead of a data-loss event. Cache headers and serving from more than one place buy the same kind of distance. None of it keeps the internet up. All of it decides whether the blast radius includes you.

And when none of that is actionable, the walk is the correct engineering decision. Panic at a keyboard does not ship the fix. Fresh air does not either, but it costs nothing and you come back with a working head.