At midday on 18 November 2025, a large share of websites started showing error pages at the same time: because of the Cloudflare outage, sites served through the provider’s network returned 5xx errors for hours. Cloudflare is one of the world’s largest CDN and proxy providers, so the fault affected not only large, well-known services but also a huge number of small business websites and online shops that use Cloudflare’s free or low-cost plans for protection and speed.
A month ago we wrote about the outage of the AWS us-east-1 region, where regional architecture was the main question. This time we look at a different dependency: what happens when the provider sitting between your website and your visitors goes down, and what you can prepare for that case.
What happened on 18 November?
That same evening, Cloudflare published a detailed summary. According to it, after a change to the permissions of an internal database (ClickHouse), a query returned duplicate rows, so the automatically generated configuration file used for bot detection (Bot Management) roughly doubled in size. This file is propagated across the whole network every few minutes. The proxy software could handle at most 200 entries from it; the oversized file made it crash, and it returned a 5xx error for every request passing through it.
The main timestamps in Hungarian time (CET):
| Time | Event |
|---|---|
| 12:20 | Errors begin on Cloudflare’s network |
| 15:30 | Most traffic flows normally again |
| 18:06 | All systems recovered |
The fault was not caused by an attack. What made the situation harder was that the faulty file was regenerated every five minutes, and it came out good or bad depending on which part of the database produced it – so for a while the network kept switching between working and failing. Cloudflare’s status page, which runs independently, happened to become unreachable at the same time, which at first raised suspicions of a coordinated attack.
Besides the CDN and the proxy, Turnstile (Cloudflare’s CAPTCHA alternative), Workers KV and Cloudflare Access were affected, and most users couldn’t even log in to the dashboard. The latter was particularly unpleasant: anyone who wanted to change their settings in the middle of it couldn’t.
What does your website depend on?
Cloudflare typically fills three roles at once, and it is worth knowing which ones you use:
| Role | What does it do? | What happens if it goes down? |
|---|---|---|
| Authoritative DNS | Says which IP address your domain can be reached at | New visitors can’t find the site |
| Proxy and CDN (“orange cloud”) | Traffic goes through Cloudflare’s servers: caching, firewall, DDoS protection | The site returns an error, even though your own server is working |
| Additional services | Turnstile, Access, Workers, email forwarding | Logins, forms and internal applications stop working |
On 18 November the proxy was the bottleneck: the servers themselves mostly ran without problems; the traffic just didn’t reach them. We explained the idea behind proxying – what is placed in front of the server, and why – in our article on reverse proxies, and about how DNS works in the knowledge base of our hosting brand, aWh.
What fallback options do you have?
1. Switching to “DNS only” mode
The most obvious emergency measure: you switch the records from proxied to “DNS only” mode (grey cloud), so Cloudflare only returns your server’s real IP address, and traffic goes directly to you. This works if you have prepared for it in advance:
- The server accepts direct traffic. If your firewall only lets in Cloudflare’s IP ranges (which is good practice anyway), have a rule ready so you can open it up quickly.
- The server has a certificate that is also valid in browsers. Browsers don’t accept Cloudflare Origin CA certificates; for direct traffic you need Let’s Encrypt or another public certificate.
- The server can handle the load without a cache. On a busy site, the CDN absorbs most of the requests.
- You know that this reveals the server’s IP address. After the outage, it is worth turning the proxy back on and considering whether the server can be attacked directly.
The TTL of proxied records is 300 seconds, so the switch takes effect within a few minutes.
2. Access without the dashboard
Since many people couldn’t log in to the dashboard, it is worth automating the switch in advance with a script and a narrowly scoped API token (DNS editing for that zone only). Turning off proxying for a record through the Cloudflare API:
curl -X PATCH \
"https://api.cloudflare.com/client/v4/zones/$ZONE_ID/dns_records/$RECORD_ID" \
-H "Authorization: Bearer $CF_API_TOKEN" \
-H "Content-Type: application/json" \
--data '{"proxied": false}'
Of course there is no guarantee that the API will be available during a major outage, but it has a better chance than a login that itself relies on Cloudflare services.
3. DNS with a separate provider
If your domain’s authoritative DNS is also with Cloudflare, all your fallback options depend on Cloudflare. If the DNS is with another provider and you connect Cloudflare only with a CNAME (Cloudflare offers this in its Business and Enterprise plans), you can bypass it by changing one record at your own DNS provider. For smaller sites this extra cost is rarely worth it, but for an online shop where an afternoon of downtime means a serious loss of revenue, it is worth doing the maths.
4. Multiple CDNs
Large providers use two CDNs in parallel and split traffic between them at DNS level. For small and medium-sized businesses this is usually overkill: the settings of two providers have to be kept in sync, and caching gets more complicated too.
5. Fault tolerance of the additional services
If your login page uses Turnstile, think through what should happen if Turnstile doesn’t respond: do you block logins, or let them through and protect them in another way (for example with stricter rate limiting)? This is a business decision, but it is better made in calm conditions than in the middle of an outage.
Checklist before the next outage
- Write down which Cloudflare service you use for which domain (DNS, proxy, Turnstile, Access, Workers, email forwarding).
- Have a public certificate valid in browsers on the servers, not just Origin CA.
- Prepare a firewall rule for direct traffic, in a disabled state.
- Create a narrowly scoped API token and a tested switch-over script, and store them somewhere independent of Cloudflare.
- Monitor the site from outside, both through Cloudflare and directly against the server – that way you see immediately where the fault is.
- Don’t run your own status page or customer notification channel on the same CDN as the website.
This wasn’t the first incident this year: in June, a Workers KV fault caused disruption in several Cloudflare services. The outage in itself is no reason to give up Cloudflare – DDoS protection and caching are worth a lot day to day. The lesson is rather that every provider that sits between your visitors and your server is also one more point that can fail. We have written before about how much an outage can cost.
We help with configuring web servers, DNS and CDNs, and with building fallback solutions; you can read more on our Infrastructure and operations page.