UNB/ CS/ David Bremner/ teaching/ cs2613/ books/ nodejs-api/ worker threads

Worker threads

Stability: 2 - Stable

The worker_threads module enables the use of threads that execute JavaScript in parallel. To access it:

const worker = require('worker_threads');

Workers (threads) are useful for performing CPU-intensive JavaScript operations. They will not help much with I/O-intensive work. Node.js’s built-in asynchronous I/O operations are more efficient than Workers can be.

Unlike child_process or cluster, worker_threads can share memory. They do so by transferring ArrayBuffer instances or sharing SharedArrayBuffer instances.

const {
  Worker, isMainThread, parentPort, workerData
} = require('worker_threads');

if (isMainThread) {
  module.exports = function parseJSAsync(script) {
    return new Promise((resolve, reject) => {
      const worker = new Worker(__filename, {
        workerData: script
      });
      worker.on('message', resolve);
      worker.on('error', reject);
      worker.on('exit', (code) => {
        if (code !== 0)
          reject(new Error(`Worker stopped with exit code ${code}`));
      });
    });
  };
} else {
  const { parse } = require('some-js-parsing-library');
  const script = workerData;
  parentPort.postMessage(parse(script));
}

The above example spawns a Worker thread for each parse() call. In actual practice, use a pool of Workers instead for these kinds of tasks. Otherwise, the overhead of creating Workers would likely exceed their benefit.

When implementing a worker pool, use the AsyncResource API to inform diagnostic tools (e.g. in order to provide asynchronous stack traces) about the correlation between tasks and their outcomes. See "Using AsyncResource for a Worker thread pool" in the async_hooks documentation for an example implementation.

Worker threads inherit non-process-specific options by default. Refer to Worker constructor options to know how to customize worker thread options, specifically argv and execArgv options.

worker.isMainThread

Is true if this code is not running inside of a Worker thread.

const { Worker, isMainThread } = require('worker_threads');

if (isMainThread) {
  // This re-loads the current file inside a Worker instance.
  new Worker(__filename);
} else {
  console.log('Inside Worker!');
  console.log(isMainThread);  // Prints 'false'.
}

worker.markAsUntransferable(object)

Mark an object as not transferable. If object occurs in the transfer list of a port.postMessage() call, it will be ignored.

In particular, this makes sense for objects that can be cloned, rather than transferred, and which are used by other objects on the sending side. For example, Node.js marks the ArrayBuffers it uses for its Buffer pool with this.

This operation cannot be undone.

const { MessageChannel, markAsUntransferable } = require('worker_threads');

const pooledBuffer = new ArrayBuffer(8);
const typedArray1 = new Uint8Array(pooledBuffer);
const typedArray2 = new Float64Array(pooledBuffer);

markAsUntransferable(pooledBuffer);

const { port1 } = new MessageChannel();
port1.postMessage(typedArray1, [ typedArray1.buffer ]);

// The following line prints the contents of typedArray1 -- it still owns
// its memory and has been cloned, not transferred. Without
// `markAsUntransferable()`, this would print an empty Uint8Array.
// typedArray2 is intact as well.
console.log(typedArray1);
console.log(typedArray2);

There is no equivalent to this API in browsers.

worker.moveMessagePortToContext(port, contextifiedSandbox)

Transfer a MessagePort to a different vm Context. The original port object will be rendered unusable, and the returned MessagePort instance will take its place.

The returned MessagePort will be an object in the target context, and will inherit from its global Object class. Objects passed to the port.onmessage() listener will also be created in the target context and inherit from its global Object class.

However, the created MessagePort will no longer inherit from EventEmitter, and only port.onmessage() can be used to receive events using it.

worker.parentPort

If this thread was spawned as a Worker, this will be a MessagePort allowing communication with the parent thread. Messages sent using parentPort.postMessage() will be available in the parent thread using worker.on('message'), and messages sent from the parent thread using worker.postMessage() will be available in this thread using parentPort.on('message').

const { Worker, isMainThread, parentPort } = require('worker_threads');

if (isMainThread) {
  const worker = new Worker(__filename);
  worker.once('message', (message) => {
    console.log(message);  // Prints 'Hello, world!'.
  });
  worker.postMessage('Hello, world!');
} else {
  // When a message from the parent thread is received, send it back:
  parentPort.once('message', (message) => {
    parentPort.postMessage(message);
  });
}

worker.receiveMessageOnPort(port)

Receive a single message from a given MessagePort. If no message is available, undefined is returned, otherwise an object with a single message property that contains the message payload, corresponding to the oldest message in the MessagePort’s queue.

const { MessageChannel, receiveMessageOnPort } = require('worker_threads');
const { port1, port2 } = new MessageChannel();
port1.postMessage({ hello: 'world' });

console.log(receiveMessageOnPort(port2));
// Prints: { message: { hello: 'world' } }
console.log(receiveMessageOnPort(port2));
// Prints: undefined

When this function is used, no 'message' event will be emitted and the onmessage listener will not be invoked.

worker.resourceLimits

Provides the set of JS engine resource constraints inside this Worker thread. If the resourceLimits option was passed to the Worker constructor, this matches its values.

If this is used in the main thread, its value is an empty object.

worker.SHARE_ENV

A special value that can be passed as the env option of the Worker constructor, to indicate that the current thread and the Worker thread should share read and write access to the same set of environment variables.

const { Worker, SHARE_ENV } = require('worker_threads');
new Worker('process.env.SET_IN_WORKER = "foo"', { eval: true, env: SHARE_ENV })
  .on('exit', () => {
    console.log(process.env.SET_IN_WORKER);  // Prints 'foo'.
  });

worker.threadId

An integer identifier for the current thread. On the corresponding worker object (if there is any), it is available as worker.threadId. This value is unique for each Worker instance inside a single process.

worker.workerData

An arbitrary JavaScript value that contains a clone of the data passed to this thread’s Worker constructor.

The data is cloned as if using postMessage(), according to the HTML structured clone algorithm.

const { Worker, isMainThread, workerData } = require('worker_threads');

if (isMainThread) {
  const worker = new Worker(__filename, { workerData: 'Hello, world!' });
} else {
  console.log(workerData);  // Prints 'Hello, world!'.
}

Class: MessageChannel

Instances of the worker.MessageChannel class represent an asynchronous, two-way communications channel. The MessageChannel has no methods of its own. new MessageChannel() yields an object with port1 and port2 properties, which refer to linked MessagePort instances.

const { MessageChannel } = require('worker_threads');

const { port1, port2 } = new MessageChannel();
port1.on('message', (message) => console.log('received', message));
port2.postMessage({ foo: 'bar' });
// Prints: received { foo: 'bar' } from the `port1.on('message')` listener

Class: MessagePort

Instances of the worker.MessagePort class represent one end of an asynchronous, two-way communications channel. It can be used to transfer structured data, memory regions and other MessagePorts between different Workers.

With the exception of MessagePorts being EventEmitters rather than EventTargets, this implementation matches browser MessagePorts.

Event: 'close'

The 'close' event is emitted once either side of the channel has been disconnected.

const { MessageChannel } = require('worker_threads');
const { port1, port2 } = new MessageChannel();

// Prints:
//   foobar
//   closed!
port2.on('message', (message) => console.log(message));
port2.on('close', () => console.log('closed!'));

port1.postMessage('foobar');
port1.close();

Event: 'message'

The 'message' event is emitted for any incoming message, containing the cloned input of port.postMessage().

Listeners on this event will receive a clone of the value parameter as passed to postMessage() and no further arguments.

Event: 'messageerror'

The 'messageerror' event is emitted when deserializing a message failed.

port.close()

Disables further sending of messages on either side of the connection. This method can be called when no further communication will happen over this MessagePort.

The 'close' event will be emitted on both MessagePort instances that are part of the channel.

port.postMessage(value[, transferList])

Sends a JavaScript value to the receiving side of this channel. value will be transferred in a way which is compatible with the HTML structured clone algorithm.

In particular, the significant differences to JSON are:

const { MessageChannel } = require('worker_threads');
const { port1, port2 } = new MessageChannel();

port1.on('message', (message) => console.log(message));

const circularData = {};
circularData.foo = circularData;
// Prints: { foo: [Circular] }
port2.postMessage(circularData);

transferList may be a list of ArrayBuffer, MessagePort and FileHandle objects. After transferring, they will not be usable on the sending side of the channel anymore (even if they are not contained in value). Unlike with child processes, transferring handles such as network sockets is currently not supported.

If value contains SharedArrayBuffer instances, those will be accessible from either thread. They cannot be listed in transferList.

value may still contain ArrayBuffer instances that are not in transferList; in that case, the underlying memory is copied rather than moved.

const { MessageChannel } = require('worker_threads');
const { port1, port2 } = new MessageChannel();

port1.on('message', (message) => console.log(message));

const uint8Array = new Uint8Array([ 1, 2, 3, 4 ]);
// This posts a copy of `uint8Array`:
port2.postMessage(uint8Array);
// This does not copy data, but renders `uint8Array` unusable:
port2.postMessage(uint8Array, [ uint8Array.buffer ]);

// The memory for the `sharedUint8Array` will be accessible from both the
// original and the copy received by `.on('message')`:
const sharedUint8Array = new Uint8Array(new SharedArrayBuffer(4));
port2.postMessage(sharedUint8Array);

// This transfers a freshly created message port to the receiver.
// This can be used, for example, to create communication channels between
// multiple `Worker` threads that are children of the same parent thread.
const otherChannel = new MessageChannel();
port2.postMessage({ port: otherChannel.port1 }, [ otherChannel.port1 ]);

Because the object cloning uses the structured clone algorithm, non-enumerable properties, property accessors, and object prototypes are not preserved. In particular, Buffer objects will be read as plain Uint8Arrays on the receiving side.

The message object will be cloned immediately, and can be modified after posting without having side effects.

For more information on the serialization and deserialization mechanisms behind this API, see the serialization API of the v8 module.

Considerations when transferring TypedArrays and Buffers

All TypedArray and Buffer instances are views over an underlying ArrayBuffer. That is, it is the ArrayBuffer that actually stores the raw data while the TypedArray and Buffer objects provide a way of viewing and manipulating the data. It is possible and common for multiple views to be created over the same ArrayBuffer instance. Great care must be taken when using a transfer list to transfer an ArrayBuffer as doing so will cause all TypedArray and Buffer instances that share that same ArrayBuffer to become unusable.

const ab = new ArrayBuffer(10);

const u1 = new Uint8Array(ab);
const u2 = new Uint16Array(ab);

console.log(u2.length);  // prints 5

port.postMessage(u1, [u1.buffer]);

console.log(u2.length);  // prints 0

For Buffer instances, specifically, whether the underlying ArrayBuffer can be transferred or cloned depends entirely on how instances were created, which often cannot be reliably determined.

An ArrayBuffer can be marked with markAsUntransferable() to indicate that it should always be cloned and never transferred.

Depending on how a Buffer instance was created, it may or may not own its underlying ArrayBuffer. An ArrayBuffer must not be transferred unless it is known that the Buffer instance owns it. In particular, for Buffers created from the internal Buffer pool (using, for instance Buffer.from() or Buffer.alloc()), transferring them is not possible and they will always be cloned, which sends a copy of the entire Buffer pool. This behavior may come with unintended higher memory usage and possible security concerns.

See Buffer.allocUnsafe() for more details on Buffer pooling.

The ArrayBuffers for Buffer instances created using Buffer.alloc() or Buffer.allocUnsafeSlow() can always be transferred but doing so will render all other existing views of those ArrayBuffers unusable.

port.ref()

Opposite of unref(). Calling ref() on a previously unref()ed port will not let the program exit if it's the only active handle left (the default behavior). If the port is ref()ed, calling ref() again will have no effect.

If listeners are attached or removed using .on('message'), the port will be ref()ed and unref()ed automatically depending on whether listeners for the event exist.

port.start()

Starts receiving messages on this MessagePort. When using this port as an event emitter, this will be called automatically once 'message' listeners are attached.

This method exists for parity with the Web MessagePort API. In Node.js, it is only useful for ignoring messages when no event listener is present. Node.js also diverges in its handling of .onmessage. Setting it will automatically call .start(), but unsetting it will let messages queue up until a new handler is set or the port is discarded.

port.unref()

Calling unref() on a port will allow the thread to exit if this is the only active handle in the event system. If the port is already unref()ed calling unref() again will have no effect.

If listeners are attached or removed using .on('message'), the port will be ref()ed and unref()ed automatically depending on whether listeners for the event exist.

Class: Worker

The Worker class represents an independent JavaScript execution thread. Most Node.js APIs are available inside of it.

Notable differences inside a Worker environment are:

Creating Worker instances inside of other Workers is possible.

Like Web Workers and the cluster module, two-way communication can be achieved through inter-thread message passing. Internally, a Worker has a built-in pair of MessagePorts that are already associated with each other when the Worker is created. While the MessagePort object on the parent side is not directly exposed, its functionalities are exposed through worker.postMessage() and the worker.on('message') event on the Worker object for the parent thread.

To create custom messaging channels (which is encouraged over using the default global channel because it facilitates separation of concerns), users can create a MessageChannel object on either thread and pass one of the MessagePorts on that MessageChannel to the other thread through a pre-existing channel, such as the global one.

See port.postMessage() for more information on how messages are passed, and what kind of JavaScript values can be successfully transported through the thread barrier.

const assert = require('assert');
const {
  Worker, MessageChannel, MessagePort, isMainThread, parentPort
} = require('worker_threads');
if (isMainThread) {
  const worker = new Worker(__filename);
  const subChannel = new MessageChannel();
  worker.postMessage({ hereIsYourPort: subChannel.port1 }, [subChannel.port1]);
  subChannel.port2.on('message', (value) => {
    console.log('received:', value);
  });
} else {
  parentPort.once('message', (value) => {
    assert(value.hereIsYourPort instanceof MessagePort);
    value.hereIsYourPort.postMessage('the worker is sending this');
    value.hereIsYourPort.close();
  });
}

new Worker(filename[, options])

Event: 'error'

The 'error' event is emitted if the worker thread throws an uncaught exception. In that case, the worker will be terminated.

Event: 'exit'

The 'exit' event is emitted once the worker has stopped. If the worker exited by calling process.exit(), the exitCode parameter will be the passed exit code. If the worker was terminated, the exitCode parameter will be 1.

This is the final event emitted by any Worker instance.

Event: 'message'

The 'message' event is emitted when the worker thread has invoked require('worker_threads').parentPort.postMessage(). See the port.on('message') event for more details.

All messages sent from the worker thread will be emitted before the 'exit' event is emitted on the Worker object.

Event: 'messageerror'

The 'messageerror' event is emitted when deserializing a message failed.

Event: 'online'

The 'online' event is emitted when the worker thread has started executing JavaScript code.

worker.getHeapSnapshot()

Returns a readable stream for a V8 snapshot of the current state of the Worker. See v8.getHeapSnapshot() for more details.

If the Worker thread is no longer running, which may occur before the 'exit' event is emitted, the returned Promise will be rejected immediately with an ERR_WORKER_NOT_RUNNING error.

worker.performance

An object that can be used to query performance information from a worker instance. Similar to perf_hooks.performance.

performance.eventLoopUtilization([utilization1[, utilization2]])

The same call as perf_hooks eventLoopUtilization(), except the values of the worker instance are returned.

One difference is that, unlike the main thread, bootstrapping within a worker is done within the event loop. So the event loop utilization will be immediately available once the worker's script begins execution.

An idle time that does not increase does not indicate that the worker is stuck in bootstrap. The following examples shows how the worker's entire lifetime will never accumulate any idle time, but is still be able to process messages.

const { Worker, isMainThread, parentPort } = require('worker_threads');

if (isMainThread) {
  const worker = new Worker(__filename);
  setInterval(() => {
    worker.postMessage('hi');
    console.log(worker.performance.eventLoopUtilization());
  }, 100).unref();
  return;
}

parentPort.on('message', () => console.log('msg')).unref();
(function r(n) {
  if (--n < 0) return;
  const t = Date.now();
  while (Date.now() - t < 300);
  setImmediate(r, n);
})(10);

The event loop utilization of a worker is available only after the 'online' event emitted, and if called before this, or after the 'exit' event, then all properties have the value of 0.

worker.postMessage(value[, transferList])

Send a message to the worker that will be received via require('worker_threads').parentPort.on('message'). See port.postMessage() for more details.

worker.ref()

Opposite of unref(), calling ref() on a previously unref()ed worker will not let the program exit if it's the only active handle left (the default behavior). If the worker is ref()ed, calling ref() again will have no effect.

worker.resourceLimits

Provides the set of JS engine resource constraints for this Worker thread. If the resourceLimits option was passed to the Worker constructor, this matches its values.

If the worker has stopped, the return value is an empty object.

worker.stderr

This is a readable stream which contains data written to process.stderr inside the worker thread. If stderr: true was not passed to the Worker constructor, then data will be piped to the parent thread's process.stderr stream.

worker.stdin

If stdin: true was passed to the Worker constructor, this is a writable stream. The data written to this stream will be made available in the worker thread as process.stdin.

worker.stdout

This is a readable stream which contains data written to process.stdout inside the worker thread. If stdout: true was not passed to the Worker constructor, then data will be piped to the parent thread's process.stdout stream.

worker.terminate()

Stop all JavaScript execution in the worker thread as soon as possible. Returns a Promise for the exit code that is fulfilled when the 'exit' event is emitted.

worker.threadId

An integer identifier for the referenced thread. Inside the worker thread, it is available as require('worker_threads').threadId. This value is unique for each Worker instance inside a single process.

worker.unref()

Calling unref() on a worker will allow the thread to exit if this is the only active handle in the event system. If the worker is already unref()ed calling unref() again will have no effect.