How to tell a node is unreachable: our probe methodology

A server can be up while your users can't reach it. Here's how we notice, and where the method falls short.

Updated: · Markdown version

The problem

Standard monitoring checks that a server responds. For a VPN node that isn't enough: the server can be running while connections from the networks your users are on fail. The user sees "VPN is down"; the owner sees green dashboards.

So a node has to be checked from where the users are.

How a check works

  • From where. Probe points inside the networks your users are on. These are volunteer vantage points, mostly in data centers and some on residential networks. We're starting with networks in Russia.
  • What we test. Whether a TCP connection opens and a TLS handshake with the expected SNI completes. It's a strong signal, but the probe doesn't replay a full Reality handshake.
  • How often. Once an hour.
  • How many probes. Three per node per check.

When a node counts as unreachable

One failed probe means nothing; networks are noisy. So:

  1. A check fails when at least two of three probes fail.
  2. A node is flagged after two failed checks in a row.
  3. The flag clears after the first successful check.

With an hourly interval, one to two hours pass between a problem starting and the flag. That's a deliberate tradeoff: more frequent checks mean more false alarms.

Two safeguards against false alarms

Control targets. Alongside your nodes we probe addresses that are known to be reachable. If those fail too, it's a monitoring failure and no nodes get flagged.

Mass failure. If half the nodes look unreachable at once, that's also treated as a monitoring failure rather than every node failing at the same moment.

What happens after a flag

  • We pull the node out of rotation and traffic moves to your other nodes.
  • We replace the node or its IP. If it runs on your server, you order the new IP from your provider and we move the node over.
  • You get a message: what happened and what we're doing.

Limits of the method

  • A probe isn't a user's phone. Reachability can depend on the carrier and region, and some networks may have no probe points.
  • If no probe points are available in a given hour, the result is "unknown," not "unreachable."
  • A passing handshake check doesn't prove real client traffic will get through.
  • We only probe someone else's nodes (for example, during an audit) with the owner's written consent.

Let us look at your network

The audit is free: we review nodes, panel and bot and show the weak spots. We reply on Telegram.

Free audit →