jest-worker

Module for executing heavy tasks under forked processes in parallel, by providing a Promise based interface, minimum overhead, and bound workers.

The module works by providing an absolute path of the module to be loaded in all forked processes. All methods are exposed on the parent process as promises, so they can be await'ed. Child (worker) methods can either be synchronous or asynchronous.

The module also implements support for bound workers. Binding a worker means that, based on certain parameters, the same task will always be executed by the same worker. The way bound workers work is by using the returned string of the computeWorkerKey method. If the string was used before for a task, the call will be queued to the related worker that processed the task earlier; if not, it will be executed by the first available worker, then sticked to the worker that executed it; so the next time it will be processed by the same worker. If you have no preference on the worker executing the task, but you have defined a computeWorkerKey method because you want some of the tasks to be sticked, you can return null from it.

The list of exposed methods can be explicitly provided via the exposedMethods option. If it is not provided, it will be obtained by requiring the child module into the main process, and analyzed via reflection. Check the "minimal example" section for a valid one.

Install

1yarn add jest-worker

Example

This example covers the minimal usage:

File parent.js

1import {Worker as JestWorker} from 'jest-worker';
2
3async function main() {
4  const worker = new JestWorker(require.resolve('./worker'));
5  const result = await worker.hello('Alice'); // "Hello, Alice"
6}
7
8main();

File worker.js

1export function hello(param) {
2  return `Hello, ${param}`;
3}

Experimental worker

Node shipped with worker_threads, a "threading API" that uses SharedArrayBuffers to communicate between the main process and its child threads. This feature can significantly improve the communication time between parent and child processes in jest-worker.

To use worker_threads instead of default child_process you have to pass enableWorkerThreads: true when instantiating the worker.

API

The Worker export is a constructor that is initialized by passing the worker path, plus an options object.

workerPath: string | URL (required)

Node module name or absolute path or file URL of the file to be loaded in the child processes. You can use require.resolve to transform a relative path into an absolute one.

options: Object (optional)

computeWorkerKey: (method: string, ...args: Array<unknown>) => string | null (optional)

Every time a method exposed via the API is called, computeWorkerKey is also called in order to bound the call to a worker. This is useful for workers that are able to cache the result or part of it. You bound calls to a worker by making computeWorkerKey return the same identifier for all different calls. If you do not want to bind the call to any worker, return null.

The callback you provide is called with the method name, plus all the rest of the arguments of the call. Thus, you have full control to decide what to return. Check a practical example on bound workers under the "bound worker usage" section.

By default, no process is bound to any worker.

enableWorkerThreads: boolean (optional)

By default, jest-worker will use child_process threads to spawn new Node.js processes. If you prefer worker_threads instead, pass enableWorkerThreads: true.

exposedMethods: ReadonlyArray<string> (optional)

List of method names that can be called on the child processes from the parent process. You cannot expose any method named like a public Worker method, or starting with _. If you use method auto-discovery, then these methods will not be exposed, even if they exist.

forkOptions: ForkOptions (optional)

Allow customizing all options passed to child_process.fork. By default, some values are set (cwd, env, execArgv and serialization), but you can override them and customize the rest. For a list of valid values, check the Node documentation.

idleMemoryLimit: number (optional)

Specifies the memory limit for workers before they are recycled and is primarily a work-around for this issue;

After the worker has executed a task the memory usage of it is checked. If it exceeds the value specified the worker is killed and restarted. If no limit is set this process does not occur. The limit can be specified in 2 ways:

• <= 1 - The value is assumed to be a percentage of system memory. So 0.5 sets the memory limit of the worker to half of the total system memory
• \> 1 - Assumed to be a fixed byte value. Because of the previous rule if you wanted a value of 1 byte (I don't know why) you could use 1.1.

maxRetries: number (optional)

Maximum amount of times that a dead child can be re-spawned, per call. Defaults to 3, pass Infinity to allow endless retries.

numWorkers: number (optional)

Amount of workers to spawn. Defaults to the number of CPUs minus 1.

resourceLimits: ResourceLimits (optional)

The resourceLimits option which will be passed to worker_threads workers.

silent: Boolean (optional)

Set to false for stdout and stderr to be logged to console.

By default this is true.

setupArgs: Array<unknown> (optional)

The arguments that will be passed to the setup method during initialization.

taskQueue: TaskQueue (optional)

The task queue defines in which order tasks (method calls) are processed by the workers. jest-worker ships with a FifoQueue and PriorityQueue:

• FifoQueue (default): Processes the method calls (tasks) in the call order.
• PriorityQueue: Processes the method calls by a computed priority in natural ordering (lower priorities first). Tasks with the same priority are processed in any order (FIFO not guaranteed). The constructor accepts a single argument, the function that is passed the name of the called function and the arguments and returns a numerical value for the priority: new require('jest-worker').PriorityQueue((method, filename) => filename.length).

WorkerPool: new (workerPath: string, options?: WorkerPoolOptions) => WorkerPoolInterface (optional)

Provide a custom WorkerPool class to be used for spawning child processes.

workerSchedulingPolicy: 'round-robin' | 'in-order' (optional)

Specifies the policy how tasks are assigned to workers if multiple workers are idle:

• round-robin (default): The task will be sequentially distributed onto the workers. The first task is assigned to the worker 1, the second to the worker 2, to ensure that the work is distributed across workers.
• in-order: The task will be assigned to the first free worker starting with worker 1 and only assign the work to worker 2 if the worker 1 is busy.

Tasks are always assigned to the first free worker as soon as tasks start to queue up. The scheduling policy does not define the task scheduling which is always first-in, first-out.

JestWorker

Methods

The returned JestWorker instance has all the exposed methods, plus some additional ones to interact with the workers itself:

getStdout(): Readable

Returns a ReadableStream where the standard output of all workers is piped. Note that the silent option of the child workers must be set to true to make it work. This is the default set by jest-worker, but keep it in mind when overriding options through forkOptions.

getStderr(): Readable

Returns a ReadableStream where the standard error of all workers is piped. Note that the silent option of the child workers must be set to true to make it work. This is the default set by jest-worker, but keep it in mind when overriding options through forkOptions.

start()

Starts up every worker and calls their setup function, if it exists. Returns a Promise which resolves when all workers are running and have completed their setup.

This is useful if you want to start up all your workers eagerly before they are used to call any other functions.

end()

Finishes the workers by killing all workers. No further calls can be done to the Worker instance.

Returns a Promise that resolves with { forceExited: boolean } once all workers are dead. If forceExited is true, at least one of the workers did not exit gracefully, which likely happened because it executed a leaky task that left handles open. This should be avoided, force exiting workers is a last resort to prevent creating lots of orphans.

Note:

awaiting the end() Promise immediately after the workers are no longer needed before proceeding to do other useful things in your program may not be a good idea. If workers have to be force exited, jest-worker may go through multiple stages of force exiting (e.g. SIGTERM, later SIGKILL) and give the worker overall around 1 second time to exit on its own. During this time, your program will wait, even though it may not be necessary that all workers are dead before continuing execution.

Consider deliberately leaving this Promise floating (unhandled resolution). After your program has done the rest of its work and is about to exit, the Node process will wait for the Promise to resolve after all workers are dead as the last event loop task. That way you parallelized computation time of your program and waiting time and you didn't delay the outputs of your program unnecessarily.

Worker IDs

Each worker has a unique id (index that starts with '1'), which is available inside the worker as process.env.JEST_WORKER_ID.

Setting up and tearing down the child process

The child process can define two special methods (both of them can be asynchronous):

• setup(): If defined, it's executed before the first call to any method in the child.
• teardown(): If defined, it's executed when the farm ends.

More examples

Standard usage

This example covers the standard usage:

File parent.js

1import {Worker as JestWorker} from 'jest-worker';
2
3async function main() {
4  const myWorker = new JestWorker(require.resolve('./worker'), {
5    exposedMethods: ['foo', 'bar', 'getWorkerId'],
6    numWorkers: 4,
7  });
8
9  console.log(await myWorker.foo('Alice')); // "Hello from foo: Alice"
10  console.log(await myWorker.bar('Bob')); // "Hello from bar: Bob"
11  console.log(await myWorker.getWorkerId()); // "3" -> this message has sent from the 3rd worker
12
13  const {forceExited} = await myWorker.end();
14  if (forceExited) {
15    console.error('Workers failed to exit gracefully');
16  }
17}
18
19main();

File worker.js

1export function foo(param) {
2  return `Hello from foo: ${param}`;
3}
4
5export function bar(param) {
6  return `Hello from bar: ${param}`;
7}
8
9export function getWorkerId() {
10  return process.env.JEST_WORKER_ID;
11}

Bound worker usage:

This example covers the usage with a computeWorkerKey method:

File parent.js

1import {Worker as JestWorker} from 'jest-worker';
2
3async function main() {
4  const myWorker = new JestWorker(require.resolve('./worker'), {
5    computeWorkerKey: (method, filename) => filename,
6  });
7
8  // Transform the given file, within the first available worker.
9  console.log(await myWorker.transform('/tmp/foo.js'));
10
11  // Wait a bit.
12  await sleep(10000);
13
14  // Transform the same file again. Will immediately return because the
15  // transformed file is cached in the worker, and `computeWorkerKey` ensures
16  // the same worker that processed the file the first time will process it now.
17  console.log(await myWorker.transform('/tmp/foo.js'));
18
19  const {forceExited} = await myWorker.end();
20  if (forceExited) {
21    console.error('Workers failed to exit gracefully');
22  }
23}
24
25main();

File worker.js

1import babel from '@babel/core';
2
3const cache = Object.create(null);
4
5export function transform(filename) {
6  if (cache[filename]) {
7    return cache[filename];
8  }
9
10  // jest-worker can handle both immediate results and thenables. If a
11  // thenable is returned, it will be await'ed until it resolves.
12  return babel.transformFileAsync(filename).then(result => {
13    cache[filename] = result;
14
15    return result;
16  });
17}