naught

zero downtime deployment for your node.js server

npm install naught
44 downloads in the last day
271 downloads in the last week
926 downloads in the last month

Build Status

Features:

  • Zero downtime code deployment
  • Ability to gracefully handle uncaught exceptions
  • Resuscitation - when a worker dies it is restarted
  • Redirect worker stdout and stderr to rotating gzipped log files
  • Runs as daemon, providing ability to start and stop
  • Clustering - take advantage of multiple CPU cores
  • Properly handles SIGTERM and SIGHUP for integration with service wrappers
  • Supports POSIX operating systems (does not support Windows)

Usage:

To use naught, your node.js server has 2 requirements.

  1. Once the server is fully booted and is readily accepting connections,

    process.send('online');
    

    Usually this is done in the listening event for a node server, for example:

    server = http.createServer(...);
    server.listen(80, function () {
      if (process.send) process.send('online');
    });
    
  2. Listen to the shutdown message and shutdown gracefully. This message is emitted after there is already a newer instance of your server online and taking care of business:

    process.on('message', function(message) {
      if (message === 'shutdown') {
        performCleanup();
        process.exit(0);
      }
    });
    

    If your server has no long-lived connections, you may skip this step. However, note that most node.js apps do have long lived connections. In fact, by default, the connection: keep-alive header is sent with every request.

    When you receive the shutdown message, either close all open connections or call process.exit().

Gracefully Handling Exceptions

Another way you can use naught is to gracefully handle exceptions that would normally cause errors for users other than the one that triggered the exception.

It is common practice to allow an uncaught exception to crash the Node.js process. In the case of a web server, that forcefully ends the execution of all other connections, resulting in more than a single user getting an error.

Using naught a worker can use the 'offline' message to announce that it is dying. At this point, naught prevents it from accepting new connections and spawns a replacement worker, allowing the dying worker to finish up with its current connections and do any cleanup necessary before finally perishing.

To take advantage of this, you need a way of catching the uncaught exceptions that cause crashes. There are two ways:

The documentation says to use Domains, so use that unless you have a better reason.

The below example assumes this is for an express app. If it is not, suit it to your needs.

  1. Setup domain.

    An easy way is to use express-domain-errors

  2. Send offline message to naught from the domain error handler

    var domainError = require('express-domain-errors');
    var domain = require('domain');
    var serverDomain = domain.create();
    var gracefulExit = require('express-graceful-exit');
    var express = require('express');
    var app;
    
    function sendOfflineMsg() {
      if (process.send) process.send('offline')
    }
    
    function doGracefulExit(err) {
      gracefulExit.gracefulExitHandler(app, server)
    }
    
    serverDomain.run(function() {
      app = express()
    
      app.use(domainError(sendOfflineMsg, doGracefulExit))
    
      // Setup app as normal
      // ...
    
      server = app.listen(process.env.PORT || 8000)
      server.on('listening', function() {
         if (process.send) process.send('online')
      })
    })
    

Tip:

If you want to deploy on a restricted port such as 80 or 443 without sudo, try authbind.

Note that there are 3 layers of process spawning between the naught CLI and your server. So you'll want to use the --deep option with authbind.

Using a service wrapper:

It may make sense to use naught with other process monitoring software. For this reason, naught supports listening to SIGTERM to do a stop operation, and SIGHUP to do a deploy operation. You may also run in the foregroun with --daemon-mode false.

When you run with --daemon-mode true (the default), the process tree looks like this:

  • CLI process, spawns the following (detached) and then exits:
    • daemon process, listens for SIGTERM/SIGHUP, spawns the following and stays running:
      • worker 1
      • worker 2
      • etc

When you run with --daemon-mode false, the process tree looks like this:

  • CLI process, listens for SIGTERM/SIGHUP, spawns the following and stays running:
    • cluster master process, spawns the following and stays running:
      • worker 1
      • worker 2
      • etc

CLI:

naught start [options] server.js [script-options]

    Starts server.js as a daemon passing script-options as command
    line arguments.

    Each worker's stdout and stderr are redirected to a log files
    specified by the `stdout` and `stderr` parameters. When a log file
    becomes larger than `max-log-size`, the log file is renamed using the
    current date and time, and a new log file is opened.

    With naught, you can use `console.log` and friends. Because naught
    pipes the output into a log file, node.js treats stdout and stderr
    as asynchronous streams.

    If you don't want a particular log, use `/dev/null` for the path. Naught
    special cases this filename and disables that log altogether.

    When running in `daemon-mode` `false`, naught will start the master
    process and then block. It listens to SIGHUP for restarting and SIGTERM
    for stopping. In this situation you may use `-` for `stderr` and/or
    `stdout` which will redirect the respective streams to naught's output
    streams instead of a log file.

    Creates an `ipc-file` which naught uses to communicate with your
    server once it has started.

    Available options and their defaults:

    --worker-count 1
    --ipc-file naught.ipc
    --log naught.log
    --stdout stdout.log
    --stderr stderr.log
    --max-log-size 10485760
    --cwd .
    --daemon-mode true
    --node-args ''


naught stop [options] [ipc-file]

    Stops the running server which created `ipc-file`.
    Uses `naught.ipc` by default.

    This sends the 'shutdown' message to all the workers and waits for
    them to exit gracefully.

    If you specify a timeout, naught will forcefully kill your workers
    if they do not shut down gracefully within the timeout.

    Available options and their defaults:

        --timeout none


naught status [ipc-file]

    Displays whether a server is running or not.
    Uses `naught.ipc` by default.


naught deploy [options] [ipc-file]

    Replaces workers with new workers using new code and optionally
    the environment variables from this command.

    Naught spawns all the new workers and waits for them to all become
    online before killing a single old worker. This guarantees zero
    downtime if any of the new workers fail and provides the ability to
    cleanly abort the deployment if it hangs.

    A hanging deploy happens when a new worker fails to emit the 'online'
    message, or when an old worker fails to shutdown upon receiving the
    'shutdown' message. A keyboard interrupt will cause a deploy-abort,
    cleanly and with zero downtime.

    If `timeout` is specified, naught will automatically abort the deploy
    if it does not finish within those seconds.

    If `override-env` is true, the environment varibables that are set with
    this command are used to override the original environment variables
    used with the `start` command. If any variables are missing, the
    original values are left intact.

    `worker-count` can be used to change the number of workers running. A
    value of `0` means to keep the same number of workers.

    Uses `naught.ipc` by default.

    Available options and their defaults:

        --worker-count 0
        --override-env true
        --timeout none
        --cwd .


naught deploy-abort [ipc-file]

    Aborts a hanging deploy. A hanging deploy happens when a new worker
    fails to emit the 'online' message, or when an old worker fails
    to shutdown upon receiving the 'shutdown' message.

    When deploying, a keyboard interrupt will cause a deploy-abort,
    so the times you actually have to run this command will be few and
    far between.

    Uses `naught.ipc` by default.

naught version

    Prints the version of naught and exits.

naught help [cmd]

    Displays help for cmd.
npm loves you