Re

Do it again, if it doesn't work the first time. Supports various configurable retry strategies, including: constant, exponential backoff and linear backoff.

Functions are styled after the async library.

Install

  npm install re

Usage

If you like the defaults, call it like this:

var retry = require('re'),
    re = new retry.Re();

re.try(function(retryCount, fail, callback){
    if(retryCount < 2) fail(new Error("Not there yet!"));
    else callback(null, retryCount);
  },
  function(err, retryCount){
    console.log("It took this many tries: " + retryCount);
});

The re.try function takes two arguments, a function to call until it works (or we run out of retries) and a function to call when it finally succeeds (or we fail too many times). As the name suggests we automatically wrap your function in a standard try block and, if an exception occurs, call it again according to the retry schedule.

This first function passed to re.try should take 3 arguments like this:

function operation(retryCount, fail, callback)

The retryCount argument is the number of the current retry. It'll be zero the first time and get bigger every time.

The fail argument is a function to call if you encounter a fail condition in your operation. This let's us know we need to try again. You can pass an err argument to the fail function.

The callback argument is the callback function you passed into re.try. It should be a function that takes an err parameter as it's first argument. The rest of the arguments are up to you. Call this when you succeed. We'll call it with the last exception or whatever you passed to the last fail, when too many failures happen.

The re.do function is like re.try expect it doesn't wrap your operation in a try...catch.

Options

The default options look like this:

var options = {
    retries : 10,
    retryStrategy : {
      "type": retry.RETRY_STRATEGY.EXPONENTIAL,
      "initial":100,
      "base":2
    }
}

You pass this options object into the Re constructor.

var retry = require('re'),
    re = new retry.Re(options);

This gives you 10 retries and an exponential backoff strategy with the following progression (in milliseconds): 100, 200, 400, 800, 1600, 3200, 6400, 12800, 25600, 51200

Retry Strategy Examples

The following will retry every 400 milliseconds:

{"type": retry.RETRY_STRATEGY.CONSTANT, "initial": 400}

The following will give a linear backoff strategy that has the following progression (when paired with retries: 10) : 200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 1800

{"type": retry.RETRY_STRATEGY.LINEAR, "initial": 200, "max":1800}

Both progressive strategies take the max option. All strategies also take a rand option. This is a Boolean that adds a random multiplier between 1 and 2. This makes them act like the tradition backoff function. This option is set to false by default.

Technical Details

The traditional exponential backoff function is described here: Exponential Backoff in Distributed Systems. This is equivalent to our exponential backoff function with the rand option set to true.

Our formula for exponential backoff looks something like this, when using all the options:

return Math.min(random * initial * Math.pow(base, retry), max);

Where random is a random number in the half-open interval [1, 2). When randomness is turned off, it's always 1.

If you don't specify a max it looks like this:

return random * initial * Math.pow(base, retry);

I'm shamelessly stealing the following link from node-retry just because it's fun for nerdy math people to play with. You can use it to calculate the exact value for base you need so that all retry intervals sum to a desired amount. Wolfram Alpha.