@gamedevsam/throng

Throng

Dead-simple one-liner for clustered Node.js apps.

Forks N workers and creates new ones if they go down. Correctly handles signals from the OS.

// CJS
const throng = require('throng');

throng((id) => console.log(`Started worker ${id}`));

// ESM
import { throng } from 'throng';

throng((id) => console.log(`Started worker ${id}`));

$ node examples/basic
Started worker 1
Started worker 2
Started worker 3
Started worker 4

Installation

$ npm install --save throng

Use

Fork 1 worker per CPU:

throng(workerStartFunction);

Specify the number of workers:

throng({ worker: workerStartFunction, count: 3 });

More options:

throng({
  primary: primaryStartFunction,
  worker: workerStartFunction,
  count: 16,
  grace: 1000
});

Handling signals:

(for cleaning up before disconnecting a worker on a SIGTERM, for instance)

throng({ worker })

function worker(id, disconnect) {
  console.log(`Started worker ${id}`)

  process.on('SIGTERM', () => {
    console.log(`Worker ${id} exiting (cleanup here)`)
    disconnect()
  })
})

All options (with defaults)

throng({
  primary: () => {},              // Fn to call in master process (can be async)
  worker: yourWorkerFunc,         // Fn to call in cluster workers (can be async)
  workerCreated: undefined        // Fn to call when a worker is created
  count: os.cpus().length,        // Number of workers
  lifetime: Infinity,             // Min time to keep cluster alive (ms)
  grace: 5000,                    // Grace period between signal and hard shutdown (ms)
  signals: ['SIGTERM', 'SIGINT']  // Signals that trigger a shutdown (proxied to workers)
})

A complex example

const throng = require('throng');

throng({ primary, worker, count: 4 });

// This will only be called once
function primary() {
  console.log('Started primary');

  process.on('beforeExit', () => {
    console.log('Primary cleanup.');
  });
}

// This will be called four times
function worker(id, disconnect) {
  let exited = false;

  console.log(`Started worker ${id}`);
  process.on('SIGTERM', shutdown);
  process.on('SIGINT', shutdown);

  async function shutdown() {
    if (exited) return;
    exited = true;

    await new Promise((r) => setTimeout(r, 300)); // simulate async cleanup work
    console.log(`Worker ${id} cleanup done.`);
    disconnect();
  }
}

$ node examples/complex
Started primary
Started worker 1
Started worker 3
Started worker 2
Started worker 4
^C
Worker 1 cleanup done.
Worker 3 cleanup done.
Worker 2 cleanup done.
Worker 4 cleanup done.
Primary cleanup.

Staying alive

Throng forks replacements for workers that crash so your cluster can continue working through failures.

$ node examples/crashy
-1--2--3--4--2--1--3--4--crash!--1--3--4--crash!--5--3--4--6--5--3--4--crash!--6--crash!--crash!--7--6--8--9--7--6--8--9--crash!--7--6--9--10--7--6--9--10--crash!--7--9--10--11--7--crash!--9--crash!--7--12--9--13--crash!--12--9--crash!--crash!--crash!--14--crash!--12--15--crash!--14--18--15--19--14--18--15--crash!--19--14--crash!--15--20--14--21--15--20--14--21--15--20--14--21--15--20--14--21--15-

Test

$ docker-compose run --rm dev

node@docker:/home/app$ npm test