Troubleshooting - DNS routing and tunnel identity
Incident: 22 November 2026. Application: the public documentation entry point. Internal addresses and tunnel identifiers are deliberately omitted.
Symptom
The public documentation hostname timed out from a mobile connection. A tunnel connector was running, but that alone did not establish that DNS sent traffic to the tunnel.
Investigation and evidence
| Observation | Interpretation |
|---|---|
| Public NS queries identified Cloudflare as authoritative DNS | DNS changes needed to be made in Cloudflare, not the old provider's zone |
| The application still had DNS-only A and AAAA records | The name selected the previous direct-address path instead of the tunnel |
| The backend returned HTTP 200 when requested from the connector with the correct Host header | The selected internal backend path worked at the time of the test |
| A replacement CNAME initially produced Error 1016 | The first DNS correction still did not select the correct tunnel target |
| The CNAME used a Replica ID instead of the Tunnel ID | Two different identifiers had been confused |
Earlier direct-access investigation had also established a private WAN address on an intermediate router and no administrative access to the upstream router. A local IPv6 HTTPS success did not establish external access. The tested mobile connection lacked IPv6, so it could not validate that alternative public path. The project therefore selected an outbound tunnel rather than continuing to change router settings without upstream access.
Correction
- Preserve the previous DNS values for a possible reversal.
- Replace only this hostname's old A and AAAA records with a proxied CNAME.
- Point the CNAME to the actual Tunnel ID followed by
.cfargotunnel.com. - Keep the matching tunnel application route directed at the separate public backend, with the intended HTTP Host header.
- Test from a mobile connection outside Wi-Fi and the VPN.
The first CNAME used a Replica ID and failed. Correcting it to the Tunnel ID was followed by a successful public page load.
Result and limits
The project owner confirmed access to the placeholder from the mobile network. This establishes that publication path, not the security or availability of all other services. The same DNS pattern was later used for Forgejo, followed by a separate successful mobile access test.
What I learned
- Authoritative DNS, cached resolver answers and application reachability are different questions.
- A healthy connector does not prove that a hostname routes to it.
- Tunnel ID and Replica ID are not interchangeable.
- Local IPv6 success cannot prove public IPv4 availability.
- Change one hostname at a time and preserve unrelated mail and private-service records.