@florajs/cluster


Simple cluster manager module with status tracking for HTTP servers.

Features

- Updates in production with zero downtime - complete worker code is replaceable without shutdown
- Self-monitoring process-management - respawn workers on error (as you would expect)
- Rollback generation if reload fails - best chances to have a running version everytime
- Aggregated status with connection-/request-tracking - see in realtime what happens in production
- Status extendible (global counters and per-request infos) - trace your app in realtime - in production
- Node.js cluster based implementation

Examples

$3

The master process will start 3 workers and restarts them on crashes:

``js
const path = require('path');
const { Master } = require('@florajs/cluster');

const master = new Master({
exec: path.join(__dirname, 'worker.js'),
workers: 3 // defaults to os.cpus().length
});

master.run();
`

$3

`js
const { createServer } = require('http');
const { Worker } = require('@florajs/cluster');

const worker = new Worker();

const httpServer = createServer((req, res) => {
res.writeHead(200, { 'Content-Type': 'text/plain' });
res.end('Hello World\n');
});

worker.attach(httpServer);

worker.on('close', () => httpServer.close());

httpServer.listen(3000);

httpServer.on('listening', () => worker.ready());
`

$3

You can retrieve an aggregated status from all workers from any worker:

`js
console.log(await worker.serverStatus());
``

$3

See "example" folder.

Best practices

$3

You can perform extended tests in your worker on startup - just exit if something fails. On a graceful
reload, the master process will cancel the reload and shutdown all workers of the new generation.
Reload is assumed to be successful when all workers called their ready() function. The old generation
of workers is shutdown not before all new workers are up and running.

$3

Aggregated status over all workers can be requested from inside a worker and from the master process.
So it is possible to integrate the status into your application. Another possibility is to start a
HTTP server inside the master process on a management port for those requests.

License

MIT