Three rules to start with
- “Connected” means the client started. It does not mean a name resolved.
- A reachable IP with an unreachable domain is the strongest DNS signal. If both fail, check the network or the node first.
- Change one condition per test and keep the exact error text. Changing DNS, TUN and rules together destroys your ability to locate the cause.
Symptoms that look like DNS
- The client reports a connection, but the browser says the domain cannot be resolved, the server cannot be found, or name resolution failed.
- On the same network, a known IP address connects while the corresponding domain keeps failing.
- Switching nodes changes nothing, but changing the system resolver or disabling the client's DNS capture restores service.
- Exactly one domain fails while others work — which may also be that domain's records, DNSSEC, regional policy or local cache.
If every site fails, every latency test times out, or the base network is down, run the full connection checklist first rather than attributing everything to DNS.
Step 1: build a baseline without the proxy
- Disconnect the proxy, TUN and system proxy, then confirm a common HTTPS site loads on plain networking.
- Record the network type: home broadband, corporate network, or mobile data. Corporate and campus networks may force a specific resolver.
- Test one public domain that carries no sensitive information. Do not experiment with banking, work consoles or private services.
- If direct resolution also fails, reconnect the network or router first. The proxy client is not the first suspect.
Step 2: compare the default resolver with an independent one
On Windows, nslookup example.com shows the default result, and nslookup example.com 1.1.1.1 queries a specific resolver for comparison. On macOS and Linux, dig example.com and dig example.com @1.1.1.1 do the same.
| Contrast result | More likely direction | Next step |
|---|---|---|
| Default fails, the named resolver succeeds | System, router or carrier resolver problem | Test a temporary resolver change, and check whether an old client left network settings modified |
| Both resolvers time out | Routing, firewall, or blocked UDP/TCP access | Confirm the base network; do not keep stacking public resolvers |
| NXDOMAIN returned | The name does not exist or the record has not propagated | Check spelling and the authoritative record; switching nodes will not help |
| SERVFAIL returned | Upstream resolution or DNSSEC validation failed | Compare with another resolver and keep the full response code |
| Command line resolves, the app fails | Browser secure DNS, app-level resolution, or a rule conflict | Inspect the application's own DoH setting and the client log |
Cloudflare's resolver troubleshooting documentation makes the same point: compare resolvers for one domain, and treat a timeout with an unreachable resolver address as a routing problem rather than a record problem.
Step 3: check who captures DNS
- Read the log first. Look for DNS timeout, NXDOMAIN, SERVFAIL, DoH connection failure or a rule miss. The “connected” badge on the main screen says nothing about any of these.
- Separate system proxy from TUN. A system proxy depends on applications honouring system settings; TUN captures more broadly and collides more easily with virtual adapters, other VPNs and security software.
- Remove duplicate capture. Keep one proxy client active. Browser secure DNS, OS encrypted DNS and the client's internal DNS can all be enabled at once — test them one at a time.
- Look at the rule result. A domain query may be routed direct while the connection goes through the proxy, or the reverse. Read the DNS and connection decisions together.
- Return to defaults. Disable custom overrides, scripts and third-party rule sets to build a minimal reproducible environment. Do not paste fake-IP, DNS hijack or complex routing fragments you do not understand: importing successfully is not the same as being correct for your version and network.
Step 4: restore resolver state per platform
Windows
- Quit the client, clear the system proxy, then check whether the adapter still holds a manual DNS entry.
- Reconnect the network and retest. Clearing the cache removes stale records only; it does not repair routing, a dead node or an upstream resolver.
- If TUN failed, check for another VPN, virtual adapter or security tool still capturing traffic.
macOS
- Check System Settings → Network → the active service → Details → DNS for servers and search domains.
- Record the original values before deleting a temporary test entry. A corporate DNS may be managed by a configuration profile.
iOS and Android
- Disconnect the client, confirm the system VPN indicator disappears, then switch once between Wi-Fi and mobile data.
- Check private DNS, secure DNS, on-demand rules and other VPN configurations for simultaneous activation.
- Do not install an unknown root certificate or profile to “repair DNS”.
Error layers, read from the log
| Log line or symptom | More likely layer | Next step |
|---|---|---|
set system dns failed: exit code 1 | The client failed to write system DNS, or capture conflicts | Record OS and client versions, check for parallel VPNs, then use the dedicated system-DNS-failure path |
dns resolve failed with every resolver timing out | Routing, firewall, TUN or an unreachable upstream | Restore the direct baseline, then check resolver reachability |
| NXDOMAIN | The name or its authoritative record | Verify spelling and records; do not use a node switch as the diagnosis |
| SERVFAIL | Upstream resolution, DNSSEC, or the path | Single-variable comparison with another resolver; keep the full response code |
| Command line works, one app fails | That app's secure DNS, cache or proxy rules | Inspect that app's resolution path |
Parallel VPNs and network extensions
Tailscale, WireGuard, corporate VPNs, filters, TUN clients and leftover proxy services can all participate in routing or DNS capture. Before starting, record each tool's purpose, current state and the way to restore it. Then pause one tool you are sure you can restore, repeat the same resolution test, and save the result.
Do not uninstall a work VPN, delete a network extension you do not recognise, or run an unverified “network reset” script. If the failure disappears after disabling one tool, that establishes a conflict relationship and nothing more — neither side is proven permanently broken.
One site only
A single failing site is not automatically DNS. Compare results across networks and resolvers, then check whether the site restricts exit region, account or IP type. If it returns NXDOMAIN or SERVFAIL, record the domain, time, resolver and response code. If it resolves but HTTPS fails, examine the certificate, system time, rule match and the site's own status. Do not switch to global proxying permanently because of one site, and do not clear all network settings repeatedly — that widens the fault without fixing the record.
What to report
- Time with time zone, OS, client name and version.
- The failing domain, without tokens or the full subscription address.
- Default and comparison resolver results, response codes, and whether each timed out.
- Mode (system proxy or TUN), node region, and the result of another node on the same subscription.
- A necessary log excerpt, with subscription URL, username, device identifier and real IP masked.
Scope
Sources: Cloudflare 1.1.1.1 resolver troubleshooting for cross-resolver comparison and timeout reasoning; Cloudflare's expanded DNS error codes for SERVFAIL and DNSSEC meanings; Apple support documentation for the macOS DNS settings path. This page gives a general sequence. It does not present any public resolver as a permanent fix for every network, and changing DNS does not repair a dead node, an account restriction or a blocked site.