hsprober

Monitors availability of Tor Hidden Services using simple TCP probes, and exports the results to Prometheus.

The tool periodically runs a set of probes towards one or more target Hidden Services. It can start a new Tor circuit on every cycle if desired (using the --new-circuits command-line option), in which case it will warm-up the circuit with a test probe before starting measurements for the target. A probe is simply a TCP connection attempt: the prober won't exchange any data on the connection, a probe is considered successful if the connection is established.

By default, it will run one probe per service every probing interval. If you are interested in measuring connection latency, it can be useful to increase this number with the --probes command-line option, to run multiple connection probes for each target one after another in sequence. This gives better statistical accuracy for latency measurements at a specific point in time. Note that, if using --new-circuits, the latency of the first connection is ignored (as it presumably includes the time to fetch the HS descriptor).

Installation

You will need a working Go environment. Then, simply running:

$ go get git.autistici.org/ale/hsprober

should build and install the hsprober binary in $GOPATH/bin.

Usage

The hsprober binary is a standalone daemon. Run it with --help to get the full list of supported options. The most important one is --targets, which should specify a list of host:port Hidden Service targets to be probed, separated by commas, e.g.:

$ hsprober --targets=autinv5q6en4gpf4.onion:443

will monitor the autinv5q6en4gpf4.onion Hidden Service on port 443.

The tool needs access to the Tor control socket in order to set up circuits (it will also use it to wait until Tor reports that it can successfully access the network on startup). Consider using a dedicated Tor instance for hsprober, so that its circuit switching does not affect other traffic.

Metrics

Metrics are collected and exported in Prometheus-compatible format (under the default /metrics path) by a small embedded HTTP server.

The following metrics are collected, each with a target label:

• tor_connection_latency is a histogram (with exp(sqrt(2)) buckets) of connection latencies
• tor_connections is a counter of the total number of attempted connections
• tor_connection_errors counts connection errors: it has an additional label error which identifies the error type (either TIMEOUT for timeouts, or ERROR for any other error).