poolifier

Node Thread Pool and Cluster Pool :arrow_double_up: :on:

Why Poolifier?

Poolifier is used to perform CPU intensive and I/O intensive tasks on nodejs servers, it implements worker pools using worker-threads and cluster pools using Node.js cluster modules.
With poolifier you can improve your performance and resolve problems related to the event loop.
Moreover you can execute your tasks using an API designed to improve the developer experience.
Please consult our general guidelines.

• Performance :racehorse: benchmarks
• Security :bank: :cop:
• Easy to use :couple:
• Dynamic pool size :white_check_mark:
• Easy switch from a pool to another :white_check_mark:
• No runtime dependencies :white_check_mark:
• Proper async integration with node async hooks :white_check_mark:
• Support for worker threads and cluster node modules :white_check_mark:
• Support sync and async tasks :white_check_mark:
• Tasks distribution strategies :white_check_mark:
• General guidance on pools to use :white_check_mark:
• Widely tested :white_check_mark:
• Error handling out of the box :white_check_mark:
• Active community :white_check_mark:
• Code quality :octocat:

Contents

Overview · Installation · Usage · Node versions · API · General guidance · Contribute · Team · License

Overview

Node pool contains two worker-threads/cluster worker pool implementations, you don't have to deal with worker-threads/cluster worker complexity.
The first implementation is a static worker pool, with a defined number of workers that are started at creation time and will be reused.
The second implementation is a dynamic worker pool with a number of worker started at creation time (these workers will be always active and reused) and other workers created when the load will increase (with an upper limit, these workers will be reused when active), the new created workers will be stopped after a configurable period of inactivity.
You have to implement your worker extending the ThreadWorker or ClusterWorker class.

Installation

npm install poolifier --save

Usage

You can implement a worker-threads worker in a simple way by extending the class ThreadWorker:

'use strict'
const { ThreadWorker } = require('poolifier')

function yourFunction(data) {
  // this will be executed in the worker thread,
  // the data will be received by using the execute method
  return { ok: 1 }
}

module.exports = new ThreadWorker(yourFunction, {
  maxInactiveTime: 60000
})

Instantiate your pool based on your needs :

'use strict'
const { DynamicThreadPool, FixedThreadPool, PoolEvents } = require('poolifier')

// a fixed worker-threads pool
const pool = new FixedThreadPool(15,
  './yourWorker.js',
  { errorHandler: (e) => console.error(e), onlineHandler: () => console.log('worker is online') })

pool.emitter.on(PoolEvents.busy, () => console.log('Pool is busy'))

// or a dynamic worker-threads pool
const pool = new DynamicThreadPool(10, 100,
  './yourWorker.js',
  { errorHandler: (e) => console.error(e), onlineHandler: () => console.log('worker is online') })

pool.emitter.on(PoolEvents.full, () => console.log('Pool is full'))
pool.emitter.on(PoolEvents.busy, () => console.log('Pool is busy'))

// the execute method signature is the same for both implementations,
// so you can easy switch from one to another
pool.execute({}).then(res => {
  console.log(res)
}).catch ....

You can do the same with the classes ClusterWorker, FixedClusterPool and DynamicClusterPool.

See examples folder for more details (in particular if you want to use a pool with multiple worker functions).

Remember that workers can only send and receive serializable data.

Node versions

Node versions >= 16.14.x are supported.

API

Pool options

An object with these properties:

• messageHandler (optional) - A function that will listen for message event on each worker

• errorHandler (optional) - A function that will listen for error event on each worker

• onlineHandler (optional) - A function that will listen for online event on each worker

• exitHandler (optional) - A function that will listen for exit event on each worker

• workerChoiceStrategy (optional) - The worker choice strategy to use in this pool:

  • WorkerChoiceStrategies.ROUND_ROBIN: Submit tasks to worker in a round robin fashion
  • WorkerChoiceStrategies.LEAST_USED: Submit tasks to the worker with the minimum number of executed, executing and queued tasks
  • WorkerChoiceStrategies.LEAST_BUSY: Submit tasks to the worker with the minimum tasks total execution and wait time
  • WorkerChoiceStrategies.LEAST_ELU: Submit tasks to the worker with the minimum event loop utilization (ELU) (experimental)
  • WorkerChoiceStrategies.WEIGHTED_ROUND_ROBIN: Submit tasks to worker by using a weighted round robin scheduling algorithm based on tasks execution time
  • WorkerChoiceStrategies.INTERLEAVED_WEIGHTED_ROUND_ROBIN: Submit tasks to worker by using an interleaved weighted round robin scheduling algorithm based on tasks execution time (experimental)
  • WorkerChoiceStrategies.FAIR_SHARE: Submit tasks to worker by using a fair share tasks scheduling algorithm based on tasks execution time or ELU

  WorkerChoiceStrategies.WEIGHTED_ROUND_ROBIN, WorkerChoiceStrategies.INTERLEAVED_WEIGHTED_ROUND_ROBIN and WorkerChoiceStrategies.FAIR_SHARE strategies are targeted to heavy and long tasks.
  Default: WorkerChoiceStrategies.ROUND_ROBIN

• workerChoiceStrategyOptions (optional) - The worker choice strategy options object to use in this pool.
  Properties:

  • measurement (optional) - The measurement to use in worker choice strategies: runTime, waitTime or elu.
  • runTime (optional) - Use the tasks median runtime instead of the tasks average runtime in worker choice strategies.
  • waitTime (optional) - Use the tasks median wait time instead of the tasks average wait time in worker choice strategies.
  • elu (optional) - Use the tasks median ELU instead of the tasks average ELU in worker choice strategies.
  • weights (optional) - The worker weights to use in the weighted round robin worker choice strategy: { 0: 200, 1: 300, ..., n: 100 }.

  Default: { runTime: { median: false }, waitTime: { median: false }, elu: { median: false } }

• restartWorkerOnError (optional) - Restart worker on uncaught error in this pool.
  Default: true

• enableEvents (optional) - Events emission enablement in this pool.
  Default: true

• enableTasksQueue (optional) - Tasks queue per worker enablement in this pool.
  Default: false

• tasksQueueOptions (optional) - The worker tasks queue options object to use in this pool.
  Properties:

  • concurrency (optional) - The maximum number of tasks that can be executed concurrently on a worker.

  Default: { concurrency: 1 }

Thread pool specific options

• workerOptions (optional) - An object with the worker options. See worker_threads for more details.

Cluster pool specific options

• env (optional) - An object with the environment variables to pass to the worker. See cluster for more details.

• settings (optional) - An object with the cluster settings. See cluster for more details.

pool = new FixedThreadPool/FixedClusterPool(numberOfThreads/numberOfWorkers, filePath, opts)

numberOfThreads/numberOfWorkers (mandatory) Number of workers for this pool
filePath (mandatory) Path to a file with a worker implementation
opts (optional) An object with the pool options properties described above

pool = new DynamicThreadPool/DynamicClusterPool(min, max, filePath, opts)

min (mandatory) Same as FixedThreadPool/FixedClusterPool numberOfThreads/numberOfWorkers, this number of workers will be always active
max (mandatory) Max number of workers that this pool can contain, the new created workers will die after a threshold (default is 1 minute, you can override it in your worker implementation).
filePath (mandatory) Path to a file with a worker implementation
opts (optional) An object with the pool options properties described above

pool.execute(data, name)

data (optional) An object that you want to pass to your worker implementation
name (optional) A string with the task function name that you want to execute on the worker. Default: 'default'
This method is available on both pool implementations and returns a promise.

pool.destroy()

Destroy method is available on both pool implementations.
This method will call the terminate method on each worker.

class YourWorker extends ThreadWorker/ClusterWorker

taskFunctions (mandatory) The task function or task functions object that you want to execute on the worker
opts (optional) An object with these properties:

• maxInactiveTime (optional) - Max time to wait tasks to work on in milliseconds, after this period the new worker will die.
  The last active time of your worker unit will be updated when a task is submitted to a worker or when a worker terminate a task.
  If killBehavior is set to KillBehaviors.HARD this value represents also the timeout for the tasks that you submit to the pool, when this timeout expires your tasks is interrupted and the worker is killed if is not part of the minimum size of the pool.
  If killBehavior is set to KillBehaviors.SOFT your tasks have no timeout and your workers will not be terminated until your task is completed.
  Default: 60000

• killBehavior (optional) - Dictates if your async unit (worker/process) will be deleted in case that a task is active on it.
  KillBehaviors.SOFT: If currentTime - lastActiveTime is greater than maxInactiveTime but a task is still executing, then the worker won't be deleted.
  KillBehaviors.HARD: If currentTime - lastActiveTime is greater than maxInactiveTime but a task is still executing, then the worker will be deleted.
  This option only apply to the newly created workers.
  Default: KillBehaviors.SOFT

General guidance

Performance is one of the main target of these worker pool implementations, we want to have a strong focus on this.
We already have a bench folder where you can find some comparisons.

Internal Node.js thread pool

Before to jump into each poolifier pool type, let highlight that Node.js comes with a thread pool already, the libuv thread pool where some particular tasks already run by default.
Please take a look at which tasks run on the libuv thread pool.

If your task runs on libuv thread pool, you can try to:

• Tune the libuv thread pool size setting the UV_THREADPOOL_SIZE.

and/or

• Use poolifier cluster pool that spawning child processes will also increase the number of libuv threads since that any new child process comes with a separated libuv thread pool. More threads does not mean more fast, so please tune your application.

Cluster vs Threads worker pools

If your task does not run into libuv thread pool and is CPU intensive then poolifier thread pools (FixedThreadPool and DynamicThreadPool) are suggested to run CPU intensive tasks, you can still run I/O intensive tasks into thread pools, but performance enhancement is expected to be minimal.
Thread pools are built on top of Node.js worker-threads module.

If your task does not run into libuv thread pool and is I/O intensive then poolifier cluster pools (FixedClusterPool and DynamicClusterPool) are suggested to run I/O intensive tasks, again you can still run CPU intensive tasks into cluster pools, but performance enhancement is expected to be minimal.
Consider that by default Node.js already has great performance for I/O tasks (asynchronous I/O).
Cluster pools are built on top of Node.js cluster module.

If your task contains code that runs on libuv plus code that is CPU intensive or I/O intensive you either split it either combine more strategies (i.e. tune the number of libuv threads and use cluster/thread pools).
But in general, always profile your application.

Fixed vs Dynamic pools

To choose your pool consider that with a FixedThreadPool/FixedClusterPool or a DynamicThreadPool/DynamicClusterPool (in this case is important the min parameter passed to the constructor) your application memory footprint will increase.
Increasing the memory footprint, your application will be ready to accept more tasks, but during idle time your application will consume more memory.
One good choose from my point of view is to profile your application using Fixed/Dynamic worker pool, and to see your application metrics when you increase/decrease the num of workers.
For example you could keep the memory footprint low choosing a DynamicThreadPool/DynamicClusterPool with 5 workers, and allow to create new workers until 50/100 when needed, this is the advantage to use the DynamicThreadPool/DynamicClusterPool.
But in general, always profile your application.

Contribute

Choose your task here 2.5.x, propose an idea, a fix, an improvement.

See CONTRIBUTING guidelines.

Team

Creator/Owner:

• Alessandro Pio Ardizio

Contributors

• Shinigami92
• Jérôme Benoit

License

MIT

Changelog

[2.6.0] - 2023-06-09

Added

• Add LEAST_ELU worker choice strategy (experimental).
• Add tasks ELU instead of runtime support to FAIR_SHARE worker choice strategy.

Changed

• Refactor pool worker node usage internals.
• Breaking change: refactor worker choice strategy statistics requirements: the syntax of the worker choice strategy options has changed.
• Breaking change: pool information info property object fields have been renamed.

Fixed

• Fix wait time accounting.
• Ensure worker choice strategy LEAST_BUSY accounts also tasks wait time.
• Ensure worker choice strategy LEAST_USED accounts also queued tasks.