Four results that place the fault
Before testing, write down the time, OS version, Shadowrocket version, current network, selected node and the first error text. Do not publish the subscription URL, QR code, server address, password or token.
| Contrast result | Check first | What you must not conclude |
|---|---|---|
| Ordinary sites also fail with Shadowrocket off | Wi-Fi, cellular or the system network stack | That the node is broken |
| Another node on the same subscription works | The original node or its path | That the working node will stay stable |
| Safari works, one app does not | Rule matching, that app's networking, regional limits | That you should change DNS or reinstall |
| All nodes show connected, all sites fail | VPN capture, rules, DNS, subscription state, network conflict | That the VPN badge means proxied traffic succeeded |
Step 1: establish the baseline network
Disconnect inside Shadowrocket, then open an ordinary HTTPS site in Safari that normally loads without any proxy. If the system still reports a VPN state, check whether another VPN, a network filter or a managed configuration is active. If you cannot confirm the direct state, record this round as “baseline unconfirmed” and do not use it to judge the client or the node.
- Direct also fails: restore Wi-Fi or cellular first. Do not keep editing Shadowrocket settings on top of a broken baseline.
- Direct works: reconnect Shadowrocket and continue with single-variable contrasts.
- Company, school or managed device: do not remove management profiles on your own; ask the administrator. Apple documents that VPN software, network filters and other third-party security tools can affect internet connectivity, and that date, time, OS version, current network and the presence of other VPNs or filters should each be checked separately.
Step 2: confirm the node and subscription actually loaded
“The node name is visible” and “the node can carry a web request” are two different facts. Check in this order:
- Confirm the selected entry is the node you intend to test, not an old node or a direct rule.
- Switch to a different node on the same subscription — one variable only.
- Verify with an ordinary HTTPS page. An IP-check page is supporting evidence at best; if you cannot confirm which path that check used, do not treat a changed exit IP as proof of success or failure.
- If every node times out, record the time, the network and the error text. Free nodes may simply have expired; that is a subscription problem, not a client problem.
Step 3: reduce configuration variables, then compare
Configuration changes alter the request path. Go back to a baseline configuration and do not assume the current version has a particular button or feature — menus differ by release.
The baseline means one of two things depending on your history. If the client worked before, it is the configuration from the last time pages loaded normally. If you have never had a working connection, it is the original configuration the provider supplied for this specific client, with no extra modules or rewrites added. If the provider did not state which client or format the configuration targets, stop and ask before importing.
- Record the current configuration and anything recently added.
- Undo recent changes from unknown sources, one at a time.
- Test an ordinary page against the baseline.
- If exactly one site fails, record the site, the time and the log result before drawing a conclusion.
- Never overwrite a working configuration with a stranger's complete config file. If only one site fails, check that site's status and your account state first.
Step 4: decide whether DNS is involved
Only enter DNS-specific work when the error text points at name resolution. Compare several domains first; do not change DNS, node and configuration simultaneously.
- Several domains fail: record the exact error and the time. Do not infer the cause from the symptom alone.
- Recovers with the client off: keep the configuration and the error text, and change one condition per test.
- One domain fails: check the site's own status before treating it as a connection-wide fault.
Do not install an unknown root certificate to work around a certificate or DNS error, and do not disable TLS verification. Both convert a diagnosable problem into an unverifiable one.
Step 5: check capture and conflicts
If the system asks to add a VPN configuration, approve it only when you trust both the app and the source of the configuration. Apple's device-management documentation describes reviewing configuration profiles under Settings → General → VPN & Device Management; deleting a profile also deletes the settings attached to it, which is why that is not step one. Paths differ on other platforms.
- Is another VPN, proxy, ad filter or security tool running at the same time?
- After disconnecting Shadowrocket, does another VPN, filter or managed profile remain active?
- Is the device managed by an employer or school, and do the policies forbid user changes?
- Are date, time and time zone correct?
- Does the problem appear on both Wi-Fi and cellular, or only one?
Switching networks and recovering narrows the range of conditions; it does not by itself identify the carrier, the DNS server or a server-side limit.
What a useful log record contains
- Time with time zone.
- OS version and Shadowrocket version.
- Wi-Fi or cellular.
- Masked node name and protocol.
- Which sites or apps failed and which worked.
- A masked description of the current configuration.
- The first error text, plus request details if this version exposes them.
- The single-variable results: after switching node, after switching network.
Never upload the full subscription, a QR code, a UUID, a password, an account email or device-management details.
Restoring ordinary networking safely
- Turn the connection off in Shadowrocket and verify with an ordinary HTTPS page.
- If the system still reports a VPN state, look for other VPNs, filters or managed profiles. The icon alone is not the success criterion.
- Undo the configuration changes you made this round; keep the original subscription and the error record.
- Pause only the tools you are authorised to manage. For anything owned by an employer or school, ask the administrator.
- Retest. A restart followed by a minimal configuration is reasonable; a system-level network reset as the first move is not. Do not delete profiles on a managed device.
Where this page stops
If the provider has confirmed that the server is down, the subscription token is revoked, the plan is exhausted or the node was removed, that is theirs to resolve. A port that is unreachable still needs the current network, path and configuration to locate. This page does not judge whether a particular provider is stable, does not supply private subscriptions, and does not promise that free nodes stay available.