process-pool

process-pool allows you to maintain a set of sub-processes with a cached state, creating a process pool that can be used to efficiently delegate work over multiple CPU cores.

Using process pool

var moment = require('moment')
var ProcessPool = require('process-pool')

// Limit number of running processes to two.
var pool = new ProcessPool({ processLimit: 2 })

function time() { return moment().diff(time.start, 'seconds') }
time.start = moment()

var func = pool.prepare(function() {
  // code here is run in the subprocess before it is first called, this allows you
  // to cache state in the subprocess so that it is immediately available.

  // this is the function run in the sub-process whenever the wrapping function
  // is called from a sub-process.
  return function(value) {
    // the promise is used to keep the process active for a second, usually
    // promises would not be used for this purpose in a process pool.
    return new Promise(function(resolve) {
      console.log('begin %s: %s', time(), returnValue)
      setTimeout(function() { resolve(p * 10) }, 1000)
    })
  }
})

for (var i = 1; i < 4; ++i) {
  func(i).then(function(returnValue) {
    console.log('end %s: %s', time(), returnValue)
  })
}

This would print:

begin 0: 1
begin 0: 2
end 1: 10
end 1: 20
begin 1: 3
end 2: 30

The process pool is set to run two processes concurrently, this delays the execution of the third call by a second.

Functions past to prepare are not closures and do not have access to surrounding scope. The following would fail:

var ProcessPool = require('process-pool')
var global = 5

var pool = new ProcessPool
var pooled = pool.prepare(function() {
  return function(argument) {
    return argument + global
  }
})

global is not available within the call to prepare. To pass context to prepare the two argument version of prepare can be used:

var ProcessPool = require('process-pool')
var global1 = 2, global2 = 10

var pool = new ProcessPool
var pooled = pool.prepare(function(context) {
  // global module requires are not available and must be required.
  var _ = require('lodash')

  return function(args) {
    return context.multiply * _.max(args) + context.add
  }
}, { multiply: global1, add: global2 })

pooled([1, 3]).then(function(value) {
  console.log("The value 16": value)
})

Requiring modules in a sub-process

By the the module path data is inherited from module.parent which is the module that included process-pool, in many cases this may not be the environment the sup-process should use. In order to use the current module path data the module option can be used. In most cases the module global variable provided by node should be passed which will case require to resolve modules according to module of the current source file.

// In this case the 'pooler' module includes 'process-pool', without using
// the `module` argument then require would resolve paths according to the
// 'pooler' module rather than this one.
var pooler = require('pooler')

var pooled = pooler.procPool.prepare(function() {
  var compiler = require('compiler')
  return function(data) {
    return compiler.compile(data)
  }
}, null, { module: module })

Running multiple functions with a single pool

Many functions can be wrapped to run in a subprocess by a single pool via calls to prepare using the processLimit option as shown in the previous example. By default processLimit copies of each prepared function are created. Up to processLimit * number of calls to prepare can be created but only processLimit subprocesses will be running code at any given time, the rest will be sleeping. This can be restricted on a per function basis:

var Promise = require('bluebird')
var ProcessPool = require('process-pool')
var pool = new ProcessPool({ processLimit: 3 })

var twoFunc = pool.prepare(function() {
  var nCalls = 0
  return function() {
    console.log("twoFunc", ++nCalls)
    return Promise.delay(1000)
  }
}, { processLimit: 2 })

var oneFunc = pool.prepare(function() {
  var nCalls = 0
  return function() {
    console.log("oneFunc", ++nCalls)
    return Promise.delay(1000)
  }
}, { processLimit: 1 })

twoFunc()
twoFunc()
twoFunc()
oneFunc()
oneFunc()

This would print:

twoFunc 1
twoFunc 2
oneFunc 1

followed by

twoFunc 3
oneFunc 2

a second later.

Future work

• Killing a pooled function should drain the wait queue.