Uptime monitoring from outside your own infrastructure that alerts you the minute your site goes down — upordown.online
Enabling HTTP/3 is often a single toggle on a CDN, or a few lines in an Nginx or Caddy config. And most of the time, nothing visible happens: pages load, a bit faster on bad mobile networks, and everyone moves on.
That's the problem. Browsers are built to fall back. If QUIC over UDP fails, Chrome and Firefox quietly retry over TCP with HTTP/2, often after a delay. So a broken HTTP/3 setup doesn't look like an outage. It looks like "the site is a bit slow sometimes, for some people", which is the hardest kind of bug to get reported.
Here's the checklist I'd use before and after flipping the switch.
Before you enable it
1. Know which layer terminates HTTP/3. If you're behind a CDN, the CDN speaks HTTP/3 to browsers and almost certainly still talks HTTP/1.1 or HTTP/2 to your origin. Your origin doesn't need any changes. If you terminate TLS yourself, your web server needs a QUIC-capable build: check nginx -V for --with-http_v3_module, or use a server like Caddy that has it built in.
2. Open UDP 443, not just TCP 443. This is the most common reason HTTP/3 "doesn't work". Cloud security groups, host firewalls (ufw, firewalld, nftables) and hosting provider firewalls usually only allow TCP 443 because that's all anyone ever needed. Add a UDP rule explicitly, and check every layer: a cloud firewall that allows UDP doesn't help if the host firewall drops it.
3. Check your load balancer supports UDP passthrough or QUIC. Many classic L7 load balancers don't forward UDP at all. If yours doesn't, HTTP/3 has to terminate at a layer that does, or it simply won't work behind it.
4. Plan the Alt-Svc header. Browsers discover HTTP/3 through the Alt-Svc response header on an HTTP/2 or HTTP/1.1 response, for example Alt-Svc: h3=":443"; ma=86400. The ma value is how long the browser remembers the advertisement. Start with a short ma (an hour or so). If something goes wrong, browsers forget about HTTP/3 quickly once you remove the header.
5. Watch out for the HTTPS DNS record. Some DNS providers and CDNs publish an HTTPS (type 65) record that advertises alpn=h3. Browsers that use it can try HTTP/3 on the very first connection, before seeing Alt-Svc. That's good for speed, but it means removing the header alone won't fully turn HTTP/3 off. Know whether your provider publishes it.
After you enable it
6. Confirm it actually negotiates. In Chrome DevTools, the Network tab's Protocol column (right-click the header to add it) should show h3. Usually the first load shows h2 and a reload shows h3, because of how Alt-Svc works. From the command line, a curl built with HTTP/3 support can test it directly:
curl -sI --http3-only https://www.example.com/ | head -1
If --http3-only fails while a normal curl -I works, HTTP/3 is advertised but broken, which is the worst combination.
7. Test from a network that isn't yours. Your office or home network may treat UDP differently from your users' networks. Test from a phone on mobile data and from a cloud VM in another region.
8. Compare latency, not just success. A browser that tries HTTP/3, waits for it to fail, and falls back can add hundreds of milliseconds to the first request. Your monitoring might show "up" while real visitors get a slower site than before. Compare time-to-first-byte before and after enabling it.
HTTP/3 failures hide behind a successful fallback, so the site looks up while it's getting slower. upordown.online checks your site from outside your own network, the way visitors reach it, so a protocol change that breaks the page for real users gets an alert instead of a shrug.
9. Check your rate limiting and DDoS rules. Rules written for TCP may not apply to QUIC traffic, or may apply too aggressively. A UDP flood rule at your provider might throttle legitimate QUIC traffic during a busy hour. Look at the provider's logs for dropped UDP after you enable it.
10. Make sure your logs still show the protocol. Add the protocol to your access log format (in Nginx, $server_protocol, which logs HTTP/3.0 for QUIC requests). Without it, you can't answer "what share of traffic uses HTTP/3, and do those requests have more errors?"
11. Know about connection migration and your sessions. QUIC connections can survive a client changing networks (Wi-Fi to mobile). That's a feature, but anything that ties a session to a client IP, like some fraud checks or sticky load balancing by IP, may now see the same connection come from two addresses. Rare, but confusing when it happens.
12. Have a one-line rollback. Write down exactly how to turn it off: the CDN toggle, or the config lines to remove, plus the DNS change if an HTTPS record is involved. Then remember that browsers may keep trying HTTP/3 until the ma time expires, which is why item 4 said to keep it short at first.
The quick version
- Open UDP 443 on every layer.
- Keep
Alt-Svcmashort until you trust it. - Verify with
--http3-onlyfrom outside your network. - Watch response times as well as success.
- Log the protocol so you can compare errors by version.
None of these take long. Skipping them is how you end up with a site that works perfectly in your own browser and is mysteriously slow on a hotel's Wi-Fi.
Whatever you enable, keep at least one outside check on the real URL. A monitor from upordown.online won't care which protocol the browser negotiated; it tells you when the page stops loading for people outside your network.
Related reading
- How to fix ERR_CONNECTION_RESET in Chrome (and how to tell when it's your server, not the visitor's Wi-Fi)
- How to fix ERR_HTTP2_PROTOCOL_ERROR in Chrome (what actually causes it, and how to find it on your server)
- Putting a CDN in front of your site? A 12-point checklist so the CDN isn't your next outage
- Six TLS certificate myths that cause outages (and one that just wastes money)
Have you turned on HTTP/3 in production yet? Did you see a measurable speedup, or did it mostly add a new thing to debug?
Top comments (0)