Epimetheus

An easy-to-use prometheus-compatible metrics library.

Epimetheus is probably the easiest way to get your Rust application serving metrics. Just set the metrics, and everything else happens automatically.

Watch your hashmaps grow!

metric!(my_data_len).set(my_data.len());
metric!(my_data_cap).set(my_data.capacity());

Monitor the latency of your functions!

let start = Instant::now();
my_function();
metric!(my_function_duration_sum).add(start.elapsed().as_secs_f64());
metric!(my_function_duration_count).add(1.0);

Track the status codes of your responses!

let resp = compute_response();
metric!(responses{code=resp.status()}).add(1.0);
Ok(resp)

Now, run your app with RUST_METRICS_PORT set:

$ RUST_METRICS_PORT=9898 cargo run

...and connect to see what's happening:

$ curl localhost:9898
my_data_cap 1024
my_data_len 764
my_function_duration_count 6032
my_function_duration_sum 8.32
responses{code="200 OK"} 5443
responses{code="404 Not Found"} 587
responses{code="500 Internal Server Error"} 2

Your program can also call epimetheus::query() to inspect its own metrics.

Features

• Multi-threaded OK! Everyone shares the same set of metrics.
• The format is prometheus-compatible. Point prometheus' watchful eye at your program and get nice graphs.
• Updating metrics is fast (...mostly. See below.)
• The port number can be customised via the RUST_METRICS_PORT environment variable.
• The code is very readable - under 200 sloc. Take a look!
• Public domain.

Performance

• Updating an unlabelled metric (like my_data_len above) is fast (~500ns uncontended, ~5us contended).
• Updating a labelled metric (like responses above) is slower (~1us uncontended, ~10us contended).
• The HTTP server is extremely dumb and can only handle one client at a time; typically this doesn't matter.

Other stuff

Epimetheus is the brother of Prometheus, hence the crate's name.