Benoit Ardietbenoitardiet@gmail.com

Business continuity, 18 November 2025

Back online in under half an hour during the Cloudflare outage

Cloudflare was down for nearly six hours and took a good part of the web with it. We were serving again in twenty-something minutes, from a decision made under time pressure that I would rather have made in advance.

The problem

On 18 November 2025 Cloudflare started returning 5xx network-wide. Our DNS was entirely proxied, so our traffic went down with it. Turnstile was failing too, which meant the Cloudflare dashboard login did not work, so there was no configuration change to make from their side even if one would have helped.

What I did

I cut traffic over to a route that reached our origin directly instead of passing through the proxy. It traded the protection the proxy gives us for being reachable at all, which was the right trade for the duration and the wrong one to leave in place afterwards.

Where it landed

We were serving again in under half an hour. Cloudflare's own postmortem puts the incident at 11:20 to 17:06 UTC. Any dependency you cannot log in to during its own outage needs a way around it that you have tested in advance, not one you invent at 11:30 on a Tuesday.

Evidence

Cloudflare published a full postmortem, linked below.

Cloudflare postmortem ↗

Recognise any of this?

Write me two paragraphs about what hurts. I will answer honestly, including when the answer is that you do not need me.

© 2026 Benoit Ardiet · Quito, Ecuador · UTC−5GitHubNo tracking, no cookies.