Node JS class for publishing metrics as time series data (TSD).

MetricsNodeJs Client

Client implementation for publishing metrics from NodeJs applications.

Determine the latest version of the NodeJs client:

metrics-client-nodejs> npm view com.arpnetworking.metrics.metrics-client dist-tags.latest

Add a dependency on the metrics client in your package.json file:

"dependencies" : {
    "com.arpnetworking.metrics.metrics-client" : "VERSION",

If you are using shrinkwrap you need to re-shrink after adding the dependency.

The metrics library must be included before being used.

mc = require("com.arpnetworking.metrics.metrics-client")

This will write the metrics to file named tsd-query.log with rotation size of 32MB and retention of the last 10 logs

Optionally, the client library may be configured by creating custom query log sink.

mc = require("com.arpnetworking.metrics.metrics-client")
var logFilename="filename.log";
var maxlogSize=10*1024*1024;
var backups=10;
mc.init([mc.Sinks.createQueryLogSink(logFilename, maxLogSize, backups)]);

Custom sinks can also be added by extending the Sink class.

var mc = require("com.arpnetworking.metrics.metrics-client");
var util = require("util");
function MySink() {
util.inherits(MySink, mc.Sink);
MySink.prototype.record = function (metricsEvent) {
var mySink =  new MySink();
mc.init([mySink, mc.Sinks.createQueryLogSink()])

A new Metrics instance should be created for each unit of work. For example:

var metrics = new mc.TsdMetrics();

Counters, timers and gauges are recorded against a metrics instance which must be closed at the end of a unit of work. After the Metrics instance is closed no further measurements may be recorded against that instance.

// Do something that is being timed 
metrics.setGauge("temperature", 21.7);

Surprisingly, counters are probably more difficult to use and interpret than timers and gauges. In the simplest case you can just starting counting, for example, iterations of a loop:

arrayOfStrings.forEach(function(itemindex) {

However, what happens if listOfString is empty? Then no sample is recorded. To always record a sample the counter should be reset before the loop is executed:

arrayOfStrings.forEach(function(itemindex) {

Next, if the loop is executed multiple times:

arrayOfArrayOfStrings.forEach(function(arrayOfStringsarrayOfStringsIndex) {
    arrayOfStrings.forEach(function(itemindex) {

The above code will produce a number of samples equal to the size of listOfListOfStrings (including no samples if listOfListOfStrings is empty). If you move the resetCounter call outside the outer loop the code always produces a single sample (even if listOfListOfStrings is empty). There is a significant difference between counting the total number of strings and the number of strings per list; especially, when computing and analyzing statistics such as percentiles.

Finally, if the loop is being executed concurrently for the same unit of work, that is for the same Metrics instance, then you could use a Counter object:

final Counter counter = metrics.createCounter("strings");
arrayOfStrings.forEach(function(itemindex) {

The Counter object example extends in a similar way for nested loops.

Timers are very easy to use. The only note worth making is that when using timers - either procedural or via objects - do not forget to stop/close the timer! If you fail to do this the client will log a warning and suppress any unstopped/unclosed samples.

The timer object allows a timer sample to be detached from the Metrics instance. For example:

public void run() {
    final Timer t = metrics.createTimer("operation");
    // Perform your operation 

The one caveat is to ensure the timer objects are stopped/closed before the Metrics instance is closed. Failing to do so will log a warning and suppress any samples stopped/closed after the Metrics instance is closed.

Gauges are the simplest metric to record. Samples for a gauge represent spot measurements. For example, the length of a queue or the number of active threads in a thread pool. Gauges are often used in separate units of work to measure the state of system resources, for example the row count in a database table. However, gauges are also useful in existing units of work, for example recording the memory in use at the beginning and end of each service request.

In lieu of logging errors the TSD library exposes a function tsd.onError(errorCallback) to register a callback that is called whenever an error occurs withing the library. The errorCallback is a function that takes a single parameter of type Error

For example, to log all errors you could do the following:

tsd.onError(function (err) {
    console.log("Error: " + err);


Install the package dependencies:

metrics-client-nodejs> npm install

Install Grunt task runner:

metrics-client-nodejs> npm install -g grunt-cli

Build the project:

metrics-client-nodejs> grunt

Execute the tests:

metrics-client-nodejs> grunt test

Generate documentation:

metrics-client-nodejs> grunt build jsdoc

Install the current version locally:

npm link /path_to_package

Using the local version is intended only for testing or development.

As a collaborator, to release a new version, perform the following:

> npm version patch|minor|major

Travis will build and release the new version to npmjs.org .

Published under Apache Software License 2.0, see LICENSE

© Groupon Inc., 2014