Awesome Graceful Cluster

• This repo is folked from graceful-cluster to add support of graceful server shotdown (if remaing connection, not shutting down soon).
• This repo is very EXPERIMENTAL and not well documented yet. if you consider to use in production, please be careful.
• I only change 'graceful-cluster.js' file from original source code;

Install:

npm install awesome-graceful-cluster

How to use

1. Enable graceful server shutdown

This patch will prevent active connections reset when server receives SIGKILL or SIGTERM. Idle (keep-alive) inbound connections without active requests will be destroyed.

Example 'server.js':

// Example server with 'express'.
var express = require('express');
var app = express();
var listener = app.listen(8000);

var GracefulServer = require('awesome-graceful-cluster').GracefulServer;
var gracefulServer = new GracefulServer({
    server: listener,
    shutdownTimeout: 10 * 1000,             // 10 sec.
});

GracefulServer options description:

| option | info | ------------------------ | --- |log | function, custom log function, console.log used by default. |server | required, http server instance. |shutdownTimeout | ms, force worker shutdown on SIGTERM timeout. Defaults to 5000ms.

Also you can initiate graceful shutdown when needed:

gracefulServer.shutdown();

2. Use simplified cluster initialization

This cluster wrapper will send SIGTERM signal to workers and wait till they finished all requests.

Also it can gracefully restart all workers one by one with zero cluster downtime on some conditions:

1. Worker memory used.
2. Worker time online.
3. Your custom condition: just call GracefulCluster.gracefullyRestartCurrentWorker() to restart current worker in serverFunction.
4. On SIGUSR2 signal to cluster process.

Example 'cluster.js':

var GracefulCluster = require('awesome-graceful-cluster').GracefulCluster;

process.title = '<your-cluster-title>';     // Note, process title must be near filename (cluster.js) length, longer title truncated.

GracefulCluster.start({
    shutdownTimeout: 10 * 1000,             // 10 sec.
    restartOnTimeout: 5 * 3600 * 1000,      // 5 hours.
    restartOnMemory: 150 * 1024 * 1024,     // 150 MB.
    serverFunction: function() {  // Your 'server.js' code module with server logic. you must return [server, app]
        var express = require('express');
        var app = express();
        var server = app.listen(8000);
        return [server, app];
    }         
});

Example 'app.js': (insert kind of below code so that you can close existing connection)

...
    app.use(function(req, res, next){
        if (app.get('graceful_shutdown') === true) {
            console.log('Connection:close header was sent.')
            res.set('Connection', 'close');
        }
        next();
    });
...

GracefulCluster options description:

| option | info | ------------------------ | --- | disableGraceful | disable graceful shutdown for faster debug. | exitFunction | optional, function that is called when the master needs to exit. The default function exits with exit code 0. | log | function, custom log function, console.log used by default. | restartOnMemory | bytes, optional. restart worker on memory usage. | restartOnTimeout | ms, optional. restart worker by timer. | serverFunction | required, function with worker logic. | shutdownTimeout | ms, optional. force worker shutdown on SIGTERM timeout. Defaults to 5000ms. | workersCount | workers count, if not specified os.cpus().length will be used. | minimumWorkerProcessHealthPercent | represents a lower limit on the number of worker servers that must remain in the RUNNING state during restarting, as a percentage of the desired number of servers . | workerProcessDisconnectDelay | worker process graceful showdown delay time. | workerProcessForceShutdownTimeout | worker process force showdown timeout (after workerProcessDisconnectDelay time).

Gracefully restart cluster

Graceful restart performed by USR2 signal:

pkill -USR2 <your-cluster-title>

or

kill -s SIGUSR2 <cluster-pid>

This method is also good if your app is launched with forever:

forever start cluster.js

Using with PM2

If you prefer PM2 you should use 'server.js' patch only. This will force PM2 to wait until active connections are closed when using:

pm2 reload <id>

With PM2 graceful reload don`t forget to set important process parameters:

• "instances": 0 - use cluster with multiple instances, so one instance will still work when another is reloaded.
• "kill_timeout": 5000 - wait more time to allow active connections finish their responses.