Http Cluster
Multi-process clustering for http.listen: a third, optional { workers: N } argument forks N-1 additional processes right after bind+listen succeeds, all of them (the original process included) sharing the one listening socket and served round-robin by the kernel — letting a single http.listen service use every core on the machine instead of just one. os.cpus.length is the natural value to pass, not an arbitrary constant: it closes the loop between this feature and the os module.
cluster.isPrimary/cluster.workerId (0 for the original process, 1..N-1 for each fork) exist purely for cases like the startup banner below — logging it from every one of N processes would be noisy, so real programs typically want exactly one of them to do it.
Like the plain single-process example (http_server.ts), http.listen never returns on its own, so each process (the original and every fork alike — there's no dedicated non-serving primary in this V1) independently schedules its own setTimeout that exits after a short delay, so this example runs to completion under `make examples` without needing a real client to connect. Point curl at it while it's running to see requests actually load-balanced across workers, e.g. run several times in a row: curl "http://localhost:8081/"
import http from 'http'
import os from 'os'
import cluster from 'cluster'
interface Res {
status: number
body: string
}
setTimeout(() => {
process.exit(0)
}, 300)
if (cluster.isPrimary) {
console.log('primary starting ' + os.cpus().length + ' worker(s) on :8081')
}
http.listen(8081, (req: HttpRequest): Res => {
return { status: 200, body: 'served by worker ' + cluster.workerId.toString() + ' (pid ' + process.pid.toString() + ')' }
}, { workers: os.cpus().length })