/async-mutex

A mutex for synchronizing async workflows in Javascript

Primary LanguageTypeScriptMIT LicenseMIT

Build status NPM version Coverage Status

What is it?

This package implements primitives for synchronizing asynchronous operations in Javascript.

Mutex

The term "mutex" usually refers to a data structure used to synchronize concurrent processes running on different threads. For example, before accessing a non-threadsafe resource, a thread will lock the mutex. This is guaranteed to block the thread until no other thread holds a lock on the mutex and thus enforces exclusive access to the resource. Once the operation is complete, the thread releases the lock, allowing other threads to acquire a lock and access the resource.

While Javascript is strictly single-threaded, the asynchronous nature of its execution model allows for race conditions that require similar synchronization primitives. Consider for example a library communicating with a web worker that needs to exchange several subsequent messages with the worker in order to achieve a task. As these messages are exchanged in an asynchronous manner, it is perfectly possible that the library is called again during this process. Depending on the way state is handled during the async process, this will lead to race conditions that are hard to fix and even harder to track down.

This library solves the problem by applying the concept of mutexes to Javascript. Locking the mutex will return a promise that resolves once the mutex becomes available. Once the async process is complete (usually taking multiple spins of the event loop), a callback supplied to the caller is called in order to release the mutex, allowing the next scheduled worker to execute.

Semaphore

Imagine a situation where you need to control access to several instances of a shared resource. For example, you might want to distribute images between several worker processes that perform transformations, or you might want to create a web crawler that performs a defined number of requests in parallel.

A semaphore is a data structure that is initialized to an arbitrary integer value and that can be locked multiple times. As long as the semaphore value is positive, locking it will return the current value and the locking process will continue execution immediately; the semaphore will be decremented upon locking. Releasing the lock will increment the semaphore again.

Once the semaphore has reached zero, the next process that attempts to acquire a lock will be suspended until another process releases its lock and this increments the semaphore again.

This library provides a semaphore implementation for Javascript that is similar to the mutex implementation described above.

How to use it?

Installation

You can install the library into your project via npm

npm install async-mutex

The library is written in TypeScript and will work in any environment that supports ES5, ES6 promises and Array.isArray. On ancient browsers, a shim can be used (e.g. core-js). No external typings are required for using this library with TypeScript (version >= 2).

Starting with Node 12.16 and 13.7, native ES6 style imports are supported.

WARNING: Node 13 versions < 13.2.0 fail to import this package correctly. Node 12 and earlier are fine, as are newer versions of Node 13.

Importing

CommonJS:

var Mutex = require('async-mutex').Mutex;
var Semaphore = require('async-mutex').Semaphore;
var withTimeout = require('async-mutex').withTimeout;

ES6:

import {Mutex, Semaphore, withTimeout} from 'async-mutex';

TypeScript:

import {Mutex, MutexInterface, Semaphore, SemaphoreInterface, withTimeout} from 'async-mutex';

With the latest version of Node, native ES6 style imports are supported.

Mutex API

Creating

const mutex = new Mutex();

Create a new mutex.

Synchronized code execution

Promise style:

mutex
    .runExclusive(function() {
        // ...
    })
    .then(function(result) {
        // ...
    });

async/await:

await mutex.runExclusive(async () => {
    // ...
});

runExclusive schedules the supplied callback to be run once the mutex is unlocked. The function may return a promise. Once the promise is resolved or rejected (or immediately after execution if an immediate value was returned), the mutex is released. runExclusive returns a promise that adopts the state of the function result.

The mutex is released and the result rejected if an exception occurs during execution if the callback.

Manual locking / releasing

Promise style:

mutex
    .acquire()
    .then(function(release) {
        // ...

        release();
    });

async/await:

const release = await mutex.acquire();
try {
    // ...
} finally {
    release();
}

acquire returns an (ES6) promise that will resolve as soon as the mutex is available. The promise resolves with a function release that must be called once the mutex should be released again. The release callback is idempotent.

IMPORTANT: Failure to call release will hold the mutex locked and will likely deadlock the application. Make sure to call release under all circumstances and handle exceptions accordingly.

Unscoped release

As an alternative to calling the release callback returned by acquire, the mutex can be released by calling release directly on it:

mutex.release();

Checking whether the mutex is locked

mutex.isLocked();

Cancelling pending locks.

Pending locks can be cancelled by calling cancel() on the mutex. This will reject all pending locks with E_CANCELED:

Promise style:

import {E_CANCELED} from 'async-mutex';

mutex
    .runExclusive(() => {
        // ...
    })
    .then(() => {
        // ...
    })
    .catch(e => {
        if (e === E_CANCELED) {
            // ...
        }
    });

async/await:

import {E_CANCELED} from 'async-mutex';

try {
    await mutex.runExclusive(() => {
        // ...
    });
} catch (e) {
    if (e === E_CANCELED) {
        // ...
    }
}

This works with acquire, too: if acquire is used for locking, the resulting promise will reject with E_CANCELED.

The error that is thrown can be customized by passing a different error to the Mutex constructor:

const mutex = new Mutex(new Error('fancy custom error'));

Note that while all pending locks are cancelled, a currently held lock will not be revoked. In consequence, the mutex may not be available even after cancel() has been called.

Waiting until the mutex is available

You can wait until the mutex is available without locking it by calling waitForUnlock(). This will return a promise that resolve once the mutex can be acquired again. This operation will not lock the mutex, and there is no guarantee that the mutex will still be available once a async barrier has been encountered.

Promise style:

mutex
    .waitForUnlock()
    .then(() => {
        // ...
    });

Async/await:

await mutex.waitForUnlock();
// ...

Semaphore API

Creating

const semaphore = new Semaphore(initialValue);

Creates a new semaphore. initialValue is an arbitrary integer that defines the initial value of the semaphore.

Synchronized code execution

Promise style:

semaphore
    .runExclusive(function(value) {
        // ...
    })
    .then(function(result) {
        // ...
    });

async/await:

await semaphore.runExclusive(async (value) => {
    // ...
});

runExclusive schedules the supplied callback to be run once the semaphore is available. The callback will receive the current value of the semaphore as its argument. The function may return a promise. Once the promise is resolved or rejected (or immediately after execution if an immediate value was returned), the semaphore is released. runExclusive returns a promise that adopts the state of the function result.

The semaphore is released and the result rejected if an exception occurs during execution of the callback.

runExclusive accepts an optional argument weight. Specifying a weight will decrement the semaphore by the specified value, and the callback will only be invoked once the semaphore's value greater or equal to weight.

Manual locking / releasing

Promise style:

semaphore
    .acquire()
    .then(function([value, release]) {
        // ...

        release();
    });

async/await:

const [value, release] = await semaphore.acquire();
try {
    // ...
} finally {
    release();
}

acquire returns an (ES6) promise that will resolve as soon as the semaphore is available. The promise resolves to an array with the first entry being the current value of the semaphore, and the second value a function that must be called to release the semaphore once the critical operation has completed. The release callback is idempotent.

IMPORTANT: Failure to call release will hold the semaphore locked and will likely deadlock the application. Make sure to call release under all circumstances and handle exceptions accordingly.

runExclusive accepts an optional argument weight. Specifying a weight will decrement the semaphore by the specified value, and the semaphore will only be acquired once the its value is greater or equal to.

Unscoped release

As an alternative to calling the release callback return by acquire, the mutex can be released by calling release directly on it:

mutex.release();

release accepts an optional argument weight and increments the semaphore accordingly.

IMPORTANT: Releasing a previously acquired semaphore with the releaser that was returned by acquire will automatically increment the semaphore by the correct weight. If you release by calling the unscoped release you have to supply the correct weight yourself!

Getting the semaphore value

semaphore.getValue()

Checking whether the semaphore is locked

semaphore.isLocked();

The semaphore is considered to be locked if its value is either zero or negative;

Setting the semaphore value

The value of a semaphore can be set directly to a desired value. A positive value will cause the semaphore to schedule any pending waiters accordingly.

semaphore.setValue();

Cancelling pending locks.

Pending locks can be cancelled by calling cancel() on the semaphore. This will reject all pending locks with E_CANCELED:

Promise style:

import {E_CANCELED} from 'async-mutex';

semaphore
    .runExclusive(() => {
        // ...
    })
    .then(() => {
        // ...
    })
    .catch(e => {
        if (e === E_CANCELED) {
            // ...
        }
    });

async/await:

import {E_CANCELED} from 'async-mutex';

try {
    await semaphore.runExclusive(() => {
        // ...
    });
} catch (e) {
    if (e === E_CANCELED) {
        // ...
    }
}

This works with acquire, too: if acquire is used for locking, the resulting promise will reject with E_CANCELED.

The error that is thrown can be customized by passing a different error to the Semaphore constructor:

const semaphore = new Semaphore(2, new Error('fancy custom error'));

Note that while all pending locks are cancelled, any currently held locks will not be revoked. In consequence, the semaphore may not be available even after cancel() has been called.

Waiting until the semaphore is available

You can wait until the semaphore is available without locking it by calling waitForUnlock(). This will return a promise that resolve once the semaphore can be acquired again. This operation will not lock the semaphore, and there is no guarantee that the semaphore will still be available once a async barrier has been encountered.

Promise style:

semaphore
    .waitForUnlock()
    .then(() => {
        // ...
    });

Async/await:

await semaphore.waitForUnlock();
// ...

waitForUnlock accepts an optional argument weight. If weight is specified the promise will only resolve once the semaphore's value is greater or equal to weight;

Limiting the time waiting for a mutex or semaphore to become available

Sometimes it is desirable to limit the time a program waits for a mutex or semaphore to become available. The withTimeout decorator can be applied to both semaphores and mutexes and changes the behavior of acquire and runExclusive accordingly.

import {withTimeout, E_TIMEOUT} from 'async-mutex';

const mutexWithTimeout = withTimeout(new Mutex(), 100);
const semaphoreWithTimeout = withTimeout(new Semaphore(5), 100);

The API of the decorated mutex or semaphore is unchanged.

The second argument of withTimeout is the timeout in milliseconds. After the timeout is exceeded, the promise returned by acquire and runExclusive will reject with E_TIMEOUT. The latter will not run the provided callback in case of an timeout.

The third argument of withTimeout is optional and can be used to customize the error with which the promise is rejected.

const mutexWithTimeout = withTimeout(new Mutex(), 100, new Error('new fancy error'));
const semaphoreWithTimeout = withTimeout(new Semaphore(5), 100, new Error('new fancy error'));

Failing early if the mutex or semaphore is not available

A shortcut exists for the case where you do not want to wait for a lock to be available at all. The tryAcquire decorator can be applied to both mutexes and semaphores and changes the behavior of acquire and runExclusive to immediately throw E_ALREADY_LOCKED if the mutex is not available.

Promise style:

import {tryAcquire, E_ALREADY_LOCKED} from 'async-mutex';

tryAcquire(semaphoreOrMutex)
    .runExclusive(() => {
        // ...
    })
    .then(() => {
        // ...
    })
    .catch(e => {
        if (e === E_ALREADY_LOCKED) {
            // ...
        }
    });

async/await:

import {tryAcquire, E_ALREADY_LOCKED} from 'async-mutex';

try {
    await tryAcquire(semaphoreOrMutex).runExclusive(() => {
        // ...
    });
} catch (e) {
    if (e === E_ALREADY_LOCKED) {
        // ...
    }
}

Again, the error can be customized by providing a custom error as second argument to tryAcquire.

License

Feel free to use this library under the conditions of the MIT license.