Yet another client for Etsy's statsd

Downloads in past


17400.4.72 years ago11 years agoMinified + gzip package size for statsd-client in KB


Node.js client for statsd.
Build Status

Quick tour

var SDC = require('statsd-client'),
	sdc = new SDC({host: ''});

var timer = new Date();
sdc.increment('some.counter'); // Increment by one.
sdc.gauge('some.gauge', 10); // Set gauge to 10
sdc.timing('some.timer', timer); // Calculates time diff
sdc.histogram('some.histogram', 10, {foo: 'bar'}) // Histogram with tags
sdc.distribution('some.distribution', 10, {foo: 'bar'}) // Distribution with tags

sdc.close(); // Optional - stop NOW



var SDC = require('statsd-client'),
	sdc = new SDC({host: '', port: 8124});

Global options: prefix: Prefix all stats with this value (default ""). tcp: User specifically wants to use tcp (default false). socketTimeout: Dual-use timer. Will flush metrics every interval. For UDP,
it auto-closes the socket after this long without activity (default 1000 ms;
0 disables this). For TCP, it auto-closes the socket after `socketTimeoutsToClose` number of timeouts have elapsed without activity.
tags: Object of string key/value pairs which will be appended on to all StatsD payloads (excluding raw payloads) (default {})
UDP options: host: Where to send the stats (default localhost). port: Port to contact the statsd-daemon on (default 8125). ipv6: Use IPv6 instead of IPv4 (default false).
TCP options:
host: Where to send the stats (default localhost). port: Port to contact the statsd-daemon on (default 8125). socketTimeoutsToClose: Number of timeouts in which the socket auto-closes if it has been inactive. (default 10; 1 to auto-close after a single timeout).
HTTP options: host: The URL to send metrics to (default: http://localhost). headers: Additional headers to send (default {}) method: What HTTP method to use (default PUT)
To debug, set the environment variable NODE_DEBUG=statsd-client when running your program.

Counting stuff

Counters are supported, both as raw .counter(metric, delta) and with the shortcuts .increment(metric, [delta=1]) and .decrement(metric, [delta=-1]):
sdc.increment('systemname.subsystem.value'); // Increment by one
sdc.decrement('systemname.subsystem.value', -10); // Decrement by 10
sdc.counter('systemname.subsystem.value', 100); // Increment by 100


Sends an arbitrary number to the back-end:
sdc.gauge('', 100);
sdc.gaugeDelta('', 20);  // Will now count 120
sdc.gaugeDelta('', -70); // Will now count 50
sdc.gauge('', 10);       // Will now count 10


Send unique occurences of events between flushes to the back-end:
sdc.set('your.set', 200);


Keep track of how fast (or slow) your stuff is:
var start = new Date();
setTimeout(function () {
	sdc.timing('random.timeout', start);
}, 100 * Math.random());

If it is given a Date, it will calculate the difference, and anything else will be passed straight through.
And don't let the name (or nifty interface) fool you - it can measure any kind of number, where you want to see the distribution (content lengths, list items, query sizes, ...)


Many implementations (though not the official one from Etsy) support histograms as an alias/alternative for timers. So aside from the fancy bits with handling dates, this is much the same as .timing().


Datadog's specific implementation supports another alternative to timers/histograms, called the distribution metric type
. From the client's perspective, this is pretty much an alias to histograms and can be used via .distribution().


Passes a raw string to the underlying socket. Useful for dealing with custom statsd-extensions in a pinch.


All the methods above support metric level tags as their last argument. Just like global tags, the format for metric tags is an object of string key/value pairs. Tags at the metric level overwrite global tags with the same key.
sdc.gauge('gauge.with.tags', 100, {foo: 'bar'});

Express helper

There's also a helper for measuring stuff in Express.js via middleware:
var SDC = require('statsd-client');
var app = express();
	sdc = new SDC({...});

// or
	function (req, res, next) { req.pipe(res); });


This will count responses by status-code (prefix.<statuscode>) and the overall response-times.
It can also measure per-URL (e.g. PUT to /:user/:thing will become PUT_user_thing by setting the timeByUrl: true in the options-object:
app.use(sdc.helpers.getExpressMiddleware('prefix', { timeByUrl: true }));

As the names can become rather odd in corner-cases (esp. regexes and non-REST interfaces), you can specify another value by setting res.locals.statsdUrlKey at a later point.
The / page will appear as root (e.g. GET_root) in metrics while any not found route will appear as {METHOD}_unknown_express_route. You can change that name by setting the notFoundRouteName in the middleware options.

Callback helper

There's also a helper for measuring stuff with regards to a callback:
var SDC = requrire('statsd-client');
	sdc = new SDC({...});

function doSomethingAsync(arg, callback) {
	callback = sdc.helpers.wrapCallback('somePrefix', callback);
	// ... do something ...
	return callback(null);

The callback is overwritten with a shimmed version that counts the number of errors (prefix.err) and successes (prefix.ok) and the time of execution of the function (prefix.time). You invoked the shimmed callback exactly the same way as though there was no shim at all. Yes, you get metrics for your function in a single line of code.
Note that the start of execution time is marked as soon as you invoke sdc.helpers.wrapCallback().
You can also provide more options:
sdc.helpers.wrapCallback('somePrefix', callback, {
	tags: { foo: 'bar' }

Stopping gracefully

By default, the socket is closed if it hasn't been used for a second (see socketTimeout in the init-options), but it can also be force-closed with .close():
var start = new Date();
setTimeout(function () {
	sdc.timing('random.timeout', start); // 2 - implicitly re-creates socket.
	sdc.close(); // 3 - Closes socket after last use.
}, 100 * Math.random());
sdc.close(); // 1 - Closes socket early.

The call is idempotent, so you can call it "just to be sure". And if you submit new metrics later, the socket will automatically be re-created, and a new timeout-timer started.


The library supports getting "child" clients with extra prefixes, to help with making sane name-spacing in apps:
// Create generic client
var sdc = new StatsDClient({host: '', prefix: 'systemname'});
sdc.increment('foo'); // Increments ''
... do great stuff ...

// Subsystem A
var sdcA = sdc.getChildClient('a');
sdcA.increment('foo'); // Increments ''

// Subsystem B
var sdcB = sdc.getChildClient('b');
sdcB.increment('foo'); // Increments ''

Internally, they all use the same socket, so calling .close() on any of them will allow the entire program to stop gracefully.

What's broken

Check the GitHub issues.

Other resources

statsd-tail - A simple program to grab statsd-data on localhost hot-shots - Another popular statsd client for Node.js statsd - The canonical server


See the changelog.