DEV Community

dong liu
dong liu

Posted on

System proxy is on but there is still no internet — four checks, in this order

A green "configuration validated" banner in your proxy client proves one thing: the YAML parsed. It does not prove the proxy is carrying a single packet. Those are different statements, and confusing them is why people reinstall the client when the real fault is a port number.

Here is the order that actually narrows things down, cheapest check first.

Step 0: separate the client's mode from the OS proxy switch

Selecting a mode inside the client and the operating system handing web requests to the client are two separate events. Keep TUN off while you do this, otherwise you cannot tell which takeover method is in play.

Look at three places, because each answers a different question:

What you see What it proves What to do next
One browser with the local port typed in manually loads pages, everything else fails That port and that browser's path may work; says nothing about the OS proxy being on Compare the client switch against the OS proxy settings, retest the same URL
The switch is on, but the client logs no matching connection The request may never have reached the current core Check the address and port the system proxy points at, and whether the browser overrides it
The client logs a failed connection The request did reach the core — the fault is node, rules or DNS Keep that log line and go to step 3

Some applications ignore the OS proxy setting entirely; browser extensions can override it. Only ever compare the same URL across these three places, or you are comparing two different failures and calling it one.

Step 1: get plain networking back

Turn off the system proxy and TUN, quit the client completely — not minimise, not "close window" if it lives in the tray — and confirm ordinary browsing returns. If browsing does not return with the client gone, the problem was never the proxy, and every check below will waste your time.

Step 2: is the core actually running, and on which port

A client can show a healthy UI while the core process has exited. Two things matter:

  • the core process exists now, and
  • the port in the system proxy setting equals the port in the current config, not the previous one.

Port mismatches are the single most common "it broke after the update" cause. An update can change mixed-port or the external controller, and the OS proxy setting keeps pointing at the old number. You then get a proxy that resolves nothing, because nothing is listening there.

If the switch is on and there is no matching entry in the client's connection log, this is your prime suspect.

Step 3: does the node have current evidence

"Current" is the operative word. A node that worked yesterday is not evidence today — free and shared nodes rotate, and expiry is usually silent. Test the node in a way that separates node health from rule matching: a direct connectivity test to a plain endpoint from inside the client, and a comparison against a different node in the same profile.

If node A fails and node B works with identical rules, you have a node problem, not a configuration problem. Changing YAML at this point will not help.

Step 4: rules and DNS for this request

Only now is it worth looking at rules. And look at them for the specific request that fails, not in general:

  • Is the domain matched by a rule that sends it to DIRECT — a rule you added for something else?
  • Is a fake-IP range or a DNS policy causing a resolution path the client cannot complete?
  • Does the failing domain resolve at all inside the client's DNS, or does it return a synthetic address that nothing is listening for?

This last case is the confusing one: pages hang instead of failing, because the client handed the browser an address that will never answer.

The pattern behind most reports

Across the reports we collected for this class of failure, the distribution is lopsided:

  • port or takeover-mode mismatch — most common, and the one people check last
  • node expired or rotated — second
  • a rule matching by accident — third
  • actual DNS misconfiguration — least common, despite being the first thing people blame

Change one item per round, and keep the result of each round written down. Two changes at once and you learn nothing.


If you want the version with the verbatim client log lines and the exact order of the four checks, it is here: System proxy is on but there is no internet. And if the failure only starts when you enable TUN rather than the system proxy, that is a different layer entirely — start with the TUN page instead, because the two failure modes get mixed up constantly.

Top comments (0)