Skip to content

Don't let your Node.JS server fall over when it's too busy.

Notifications You must be signed in to change notification settings

luislhl/node-toobusy

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Build Status npm version

Is Your Node Process Too Busy?

node-toobusy is a fork of lloyd's node-toobusy that removes native dependencies in favor of using the unref introduced in node 0.9.1.

This package is a simpler install without native dependencies, but requires node >= 0.9.1.

Node-Toobusy

What happens when your service is overwhelmed with traffic? Your server can do one of two things:

  • Stop working, or...
  • Keep serving as many requests as possible

This library helps you do the latter.

How it works

toobusy polls the node.js event loop and keeps track of "lag", which is long requests wait in node's event queue to be processed. When lag crosses a threshold, toobusy tells you that you're too busy. At this point you can stop request processing early (before you spend too much time on them and compound the problem), and return a "Server Too Busy" response. This allows your server to stay responsive under extreme load, and continue serving as many requests as possible.

installation

npm install node-toobusy

usage

var toobusy = require('node-toobusy'),
    express = require('express');

var app = express();

// middleware which blocks requests when we're too busy
app.use(function(req, res, next) {
  if (toobusy()) {
    res.send(503, "I'm busy right now, sorry.");
  } else {
    next();
  }
});

app.get('/', function(req, res) {
  // processing the request requires some work!
  var i = 0;
  while (i < 1e5) i++;
  res.send("I counted to " + i);
});

var server = app.listen(3000);

tunable parameters

The library exposes a few knobs:

maxLag - This number represents the maximum amount of time in milliseconds that the event queue is behind, before we consider the process too busy.
interval - The check interval for measuring event loop lag, in ms.
smoothingFactor - When a new lag is measured, we smooth its value using the standard exponential smoothing formula.
There are two factors available, the smoothingFactorOnRise, which is used when the new lag is higher than currentLag, and the smoothingFactorOnFall, which is used when the new lag is lower than currentLag.
It's a good idea to keep the factor on fall higher than on rise, to make the currentLag recover faster after spikes.
lagFunction - This is the function used to calculate currentLag. You can overwrite it if you need a different behavior. The parameters passed to it are: lag, currentLag, smoothingFactorOnRise and smoothingFactorOnFall.

var toobusy = require('node-toobusy');

// Set maximum lag to an aggressive value.
toobusy.maxLag(10);

// Set check interval to a faster value. This will catch more latency spikes
// but may cause the check to be too sensitive.
toobusy.interval(250);

// Set smoothing factor on rise to a lower value. This will make it less sensible
// to spikes. Default is 1/3.
toobusy.smoothingFactorOnRise(1/4);

// Set smoothing factor on fall to a higher value. This will make it recover faster
// after spikes. Default is 2/3.
toobusy.smoothingFactorOnFall(3/4);

// You can overwrite this function to change the way currentLag is calculated.
// This is the default implementation.
toobusy.lagFunction = function(lag, cLag, sFactorRise, sFactorFall) {
  var factor = lag > cLag ? sFactorRise : sFactorFall;
  return factor * lag + (1 - factor) * cLag;
}

// Get current maxLag or interval setting by calling without parameters.
var currentMaxLag = toobusy.maxLag(), interval = toobusy.interval();

toobusy.onLag(function(currentLag) {
  console.log("Event loop lag detected! Latency: " + currentLag + "ms");
});

The default maxLag value is 70ms, and the default check interval is 500ms. This allows an "average" server to run at 90-100% CPU and keeps request latency at around 200ms. For comparison, a maxLag value of 10ms results in 60-70% CPU usage, while latency for "average" requests stays at about 40ms.

These numbers are only examples, and the specifics of your hardware and application can change them drastically, so experiment! The default of 70 should get you started.

Events

As of 0.5.0, node-toobusy exposes an onLag method. Pass it a callback to be notified when a slow event loop tick has been detected.

references

There is nothing new under the sun. (Ecclesiastes 1:9)

Though applying "event loop latency" to node.js was not directly inspired by anyone else's work, this concept is not new. Here are references to others who apply the same technique:

license

WTFPL

About

Don't let your Node.JS server fall over when it's too busy.

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages

  • JavaScript 100.0%