Crate cadence [] [src]

An extensible Statsd client for Rust!

Statsd is a network server that listens for metrics (things like counters and timers) sent over UDP and sends aggregates of these metrics to a backend service of some kind (often Graphite).

Cadence is a client written in Rust for interacting with a Statsd server. You might want to emit metrics (using Cadence, sending them to a Statsd server) in your Rust server application.

For example, if you are running a Rust web service you might want to record:

  • Number of succesful requests
  • Number of error requests
  • Time taken for each request

Cadence is a flexible and easy way to do this!


  • Support for emitting counters, timers, histograms, gauges, and meters to Statsd over UDP.
  • Support for alternate backends via the MetricSink trait.
  • Support for Datadog style metric tags.
  • A simple yet flexible API for sending metrics.


To make use of Cadence in your project, add it as a dependency in your Cargo.toml file.

cadence = "x.y.z"

Then, link to it in your library or application.

// or
extern crate cadence;

// rest of your library or application


Some examples of how to use Cadence are shown below. The examples start simple and work up to how you should be using Cadence in a production application.

Simple Use

Simple usage of Cadence is shown below. In this example, we just import the client, create an instance that will write to some imaginary metrics server, and send a few metrics.

// Import the client.
use cadence::prelude::*;
use cadence::{StatsdClient, UdpMetricSink, DEFAULT_PORT};

// Create client that will write to the given host over UDP.
// Note that you'll probably want to actually handle any errors creating
// the client when you use it for real in your application. We're just
// using .unwrap() here since this is an example!
let host = ("", DEFAULT_PORT);
let client = StatsdClient::from_udp_host("my.metrics", host).unwrap();

// Emit metrics!
client.time("some.methodCall", 42);
client.gauge("some.thing", 7);
client.meter("some.value", 5);

Buffered UDP Sink

While sending a metric over UDP is very fast, the overhead of frequent network calls can start to add up. This is especially true if you are writing a high performance application that emits a lot of metrics.

To make sure that metrics aren't interfering with the performance of your application, you may want to use a MetricSink implementation that buffers multiple metrics before sending them in a single network operation. For this, there's BufferedUdpMetricSink. An example of using this sink is given below.

use std::net::UdpSocket;
use cadence::prelude::*;
use cadence::{StatsdClient, BufferedUdpMetricSink, DEFAULT_PORT};

let socket = UdpSocket::bind("").unwrap();

let host = ("", DEFAULT_PORT);
let sink = BufferedUdpMetricSink::from(host, socket).unwrap();
let client = StatsdClient::from_sink("my.prefix", sink);

client.count("my.counter.thing", 29);
client.time("", 214);

As you can see, using this buffered UDP sink is no more complicated than using the regular, non-buffered, UDP sink.

The only downside to this sink is that metrics aren't written to the Statsd server until the buffer is full. If you have a busy application that is constantly emitting metrics, this shouldn't be a problem. However, if your application only occasionally emits metrics, this sink might result in the metrics being delayed for a little while until the buffer fills.

Queuing Asynchronous Metric Sink

To make sure emitting metrics doesn't interfere with the performance of your application (even though emitting metrics is generally quite fast), it's probably a good idea to make sure metrics are emitted in in a different thread than your application thread.

To allow you do this, there is QueuingMetricSink. This sink allows you to wrap any other metric sink and send metrics to it via a queue, as it emits metrics in another thread, asynchronously from the flow of your application.

The requirements for the wrapped metric sink are that it is thread safe, meaning that it implements the Send and Sync traits. If you're using the QueuingMetricSink with another sink from Cadence, you don't need to worry: they are all thread safe.

An example of using the QueuingMetricSink to wrap a buffered UDP metric sink is given below. This is the preferred way to use Cadence in production.

use std::net::UdpSocket;
use cadence::prelude::*;
use cadence::{StatsdClient, QueuingMetricSink, BufferedUdpMetricSink,

let socket = UdpSocket::bind("").unwrap();

let host = ("", DEFAULT_PORT);
let udp_sink = BufferedUdpMetricSink::from(host, socket).unwrap();
let queuing_sink = QueuingMetricSink::from(udp_sink);
let client = StatsdClient::from_sink("my.prefix", queuing_sink);

client.count("my.counter.thing", 29);
client.time("", 214);

Use With Tags

Adding tags to metrics is accomplished via the use of each of the _with_tags methods that are part of the Cadence StatsdClient struct. An example of using these methods is given below. Note that tags are an extension to the Statsd protocol and so may not be supported by all servers.

See the Datadog docs for more information.

use cadence::prelude::*;
use cadence::{StatsdClient, NopMetricSink};

let client = StatsdClient::from_sink("my.prefix", NopMetricSink);

client.count_with_tags("my.counter.thing", 29)
    .with_tag("host", "")

Counted, Timed, Gauged, Metered, Histogrammed, and MetricClient Traits

Each of the methods that the Cadence StatsdClient struct uses to send metrics are implemented as a trait. There is also a trait that combines all of these other traits. If we want, we can just use one of the trait types to refer to the client instance. This might be useful to you if you'd like to swap out the actual Cadence client with a dummy version when you are unit testing your code or want to abstract away all the implementation details of the client being used behind a trait and pointer.

Each of these traits are exported in the prelude module. They are also available in the main module but aren't typically used like that.

use cadence::prelude::*;
use cadence::{StatsdClient, UdpMetricSink, DEFAULT_PORT};

pub struct User {
    id: u64,
    username: String,
    email: String

// Here's a simple DAO (Data Access Object) that doesn't do anything but
// uses a metric client to keep track of the number of times the
// 'getUserById' method gets called.
pub struct MyUserDao {
    metrics: Box<MetricClient>

impl MyUserDao {
    // Create a new instance that will use the StatsdClient
    pub fn new<T: MetricClient + 'static>(metrics: T) -> MyUserDao {
        MyUserDao { metrics: Box::new(metrics) }

    /// Get a new user by their ID
    pub fn get_user_by_id(&self, id: u64) -> Option<User> {

// Create a new Statsd client that writes to ""
let host = ("", DEFAULT_PORT);
let metrics = StatsdClient::from_udp_host("counter.example", host).unwrap();

// Create a new instance of the DAO that will use the client
let dao = MyUserDao::new(metrics);

// Try to lookup a user by ID!
match dao.get_user_by_id(123) {
    Some(u) => println!("Found a user!"),
    None => println!("No user!")

Custom Metric Sinks

The Cadence StatsdClient uses implementations of the MetricSink trait to send metrics to a metric server. Most users of the Candence library probably want to use the QueuingMetricSink wrapping an instance of the BufferedMetricSink.

However, maybe you want to do something not covered by an existing sink. An example of creating a custom sink is below.

use std::io;
use cadence::prelude::*;
use cadence::{StatsdClient, MetricSink, DEFAULT_PORT};

pub struct MyMetricSink;

impl MetricSink for MyMetricSink {
    fn emit(&self, metric: &str) -> io::Result<usize> {
        // Your custom metric sink implementation goes here!

let sink = MyMetricSink;
let client = StatsdClient::from_sink("my.prefix", sink);

client.count("my.counter.thing", 42);
client.time("my.method.time", 25);

Custom UDP Socket

Most users of the Cadence StatsdClient will be using it to send metrics over a UDP socket. If you need to customize the socket, for example you want to use the socket in blocking mode but set a write timeout, you can do that as demonstrated below.

use std::net::UdpSocket;
use std::time::Duration;
use cadence::prelude::*;
use cadence::{StatsdClient, UdpMetricSink, DEFAULT_PORT};

let socket = UdpSocket::bind("").unwrap();

let host = ("", DEFAULT_PORT);
let sink = UdpMetricSink::from(host, socket).unwrap();
let client = StatsdClient::from_sink("my.prefix", sink);

client.count("my.counter.thing", 29);
client.time("", 214);



Export commonly used parts of Cadence for easy glob imports



Implementation of a MetricSink that buffers metrics before sending them to a UDP socket.


Counters are simple values incremented or decremented by a client.


Gauges are an instantaneous value determined by the client.


Histograms are values whose distribution is calculated by the server.


Meters measure the rate at which events occur as determined by the server.


Builder for adding tags to in-progress metrics.


Error generated by this library potentially wrapping another type of error (exposed via the Error trait).


Implementation of a MetricSink that discards all metrics.


Implementation of a MetricSink that wraps another implementation and uses it to emit metrics asynchronously, in another thread.


Client for Statsd that implements various traits to record metrics.


Timers are a positive number of milliseconds between a start and end point.


Implementation of a MetricSink that emits metrics over UDP.



Potential categories an error from this library falls into.





Trait for incrementing and decrementing counters.


Trait for recording gauge values.


Trait for recording histogram values.


Trait for recording meter values.


Trait for metrics to expose Statsd metric string slice representation.


Trait that encompasses all other traits for sending metrics.


Trait for various backends that send Statsd metrics somewhere.


Trait for recording timings in milliseconds.

Type Definitions