Gracefully restart node.js http cluster with zero downtime. Shutdown server without active inbound connections reset.
npm install graceful-clusterInstall:
npm install graceful-cluster
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':
``js
// Example server with 'express'.
var express = require('express');
var app = express();
var listener = app.listen(8000);
var GracefulServer = require('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:
`js`
gracefulServer.shutdown();
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.SIGUSR2
4. On signal to cluster process.
Example 'cluster.js':
`js
var GracefulCluster = require('graceful-cluster').GracefulCluster;
process.title = '
GracefulCluster.start({
shutdownTimeout: 10 * 1000, // 10 sec.
restartOnTimeout: 5 3600 1000, // 5 hours.
restartOnMemory: 150 1024 1024, // 150 MB.
serverFunction: function() {
require('./server'); // Your 'server.js' code module with server logic.
}
});
`
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.
Graceful restart performed by USR2 signal:
`sh`
pkill -USR2
or
`sh`
kill -s SIGUSR2
This method is also good if your app is launched with forever:
`sh`
forever start cluster.js
If you prefer PM2 you should use 'server.js' patch only. This will force PM2 to wait until active connections are closed when using:
`sh`
pm2 reload
With PM2 graceful reload dont 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.