workerama

workerama

Feed input data array items one by one to provided function that is automatically spread across Worker Threads.

When it fails then it fails hard, i.e. it terminates entire threads pool. It's up to a consumer's worker function to retry or even swallow errors to keep things going.

:warning: Node.js v10 needs an --experimental-worker flag.

Install

$ yarn add workerama

Usage

type TOptions = {
  items: any[],
  maxThreadCount: number,
  fnFilePath: string,
  fnName: string,
  fnArgs: any[]
}

const workerama: <T>(options: TOptions) => AsyncIterable<T>

import { workerama } from 'workerama'
import { cpus } from 'os'

const resultsIterable = workerama({
  items: Array.from({ length: 1000 }, (_, i) => i),
  maxThreadCount: cpus().length,
  fnFilePath: './test',
  fnName: 'add',
  fnArgs: [1],
})

for await (const result of resultsIterable) {
  process.stdout.write(`${workerId}:${result} `)
}

process.stdout.write('\n')

// test.js

// factory function that receives `fnArgs`
exports.add = (arg1) => {
  // actual function that receives IteratorResult with per-item value
  return (item) => {
    if (!item.done) {
      Promise.resolve({
        value: item.value + arg1,
        transferList: []
      })
    }
  }
}

where:

• factory function – called once per worker, useful to initialize/instantiate something
• actual function – called once per item + once when it's done, must return special object:
  • value – actual result produced by function
  • transferList – optional array of ArrayBuffer or SharedArrayBuffer (not to be confused with Node.js Buffer, but rather Buffer.from([1, 2, 3]).buffer) to be moved from worker to parent to avoid cloning it and consuming double amount of memory