@captemulation/throng
v6.0.2
Published
A simple worker-manager for clustered apps
Downloads
8
Readme
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.
const throng = require('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({
master: masterStartFunction,
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({
master: () => {}, // Fn to call in master process (can be async)
worker: yourWorkerFunc, // Fn to call in cluster workers (can be async)
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({ master, worker, count: 4 })
// This will only be called once
function master() {
console.log('Started master')
process.on('beforeExit', () => {
console.log('Master 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 master
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.
Master 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