async_task 1.1.0

Asynchronous tasks and parallel executors (similar to thread pools) for all Dart platforms (transparently), without having to deal with `Isolate` complexity.

async_task

This package brings asynchronous tasks and parallel executors (similar to classic thread pools) for all Dart platforms (JS/Web, Flutter, VM/Native) through transparent internal implementations, based on dart:isolate or only dart:async, without having to deal with the Isolate complexity.

Motivation

Dart parallelism is based on asynchronous, non-blocking and thread-safe code. This creates a language that facilitates the creation of safe concurrency code, but all running in the same thread, using only 1 thread/core of a device.

If you want to use more than 1 thread/core of a device, Dart VM/Native has Isolate, but the paradigm is not easy to use like classical Thread Pools, or like environments that shares all the memory/objects between threads.

This package was created to facilitate the creation and execution of multiple tasks asynchronously in all Dart Platforms, avoiding platform specific codes by the developer.

Usage

import 'dart:async';

import 'package:async_task/async_task.dart';

void main() async {
  // The tasks to execute:
  var tasks = [
    PrimeChecker(8779),
    PrimeChecker(1046527),
    PrimeChecker(3139581), // Not Prime
    PrimeChecker(16769023),
  ];

  // Instantiate the task executor:
  var asyncExecutor = AsyncExecutor(
    sequential: false, // Non-sequential tasks.
    parallelism: 2, // Concurrency with 2 threads.
    taskTypeRegister: _taskTypeRegister, // The top-level function to register the tasks types.
  );

  // Enable logging output:
  asyncExecutor.logger.enabled = true ;

  // Execute all tasks:
  var executions = asyncExecutor.executeAll(tasks);

  // Wait tasks executions:
  await Future.wait(executions);

  for (var task in tasks) {
    var n = task.n; // Number to check for prime.
    var prime = task.result; // Task result: true if is prime.
    print('$n\t-> $prime \t $task');
  }
}

// This top-level function returns the tasks types that will be registered
// for execution. Task instances are returned, but won't be executed and
// will be used only to identify the task type:
List<AsyncTask> _taskTypeRegister() => [PrimeChecker(0)];

// A task that checks if a number is prime:
class PrimeChecker extends AsyncTask<int, bool> {
  // The number to check if is prime.
  final int n;

  PrimeChecker(this.n);

  // Instantiates a `PrimeChecker` task with `parameters`.
  @override
  AsyncTask<int, bool> instantiate(int parameters) {
    return PrimeChecker(parameters);
  }

  // The parameters of this task:
  @override
  int parameters() {
    return n;
  }

  // Runs the task code:
  @override
  FutureOr<bool> run() {
    return isPrime(n);
  }

  // A simple prime check function:
  bool isPrime(int n) {
    if (n < 2) return false;

    var limit = n ~/ 2;
    for (var p = 2; p <= limit; ++p) {
      if (n % p == 0) return false;
    }

    return true;
  }
}

Output:

[INFO] Starting AsyncExecutor{ sequential: false, parallelism: 2, executorThread: _AsyncExecutorMultiThread{ totalThreads: 2, queue: 0 } }
[INFO] Starting _AsyncExecutorMultiThread{ totalThreads: 2, queue: 0 }
[INFO] Created _IsolateThread{ id: 2 ; registeredTasksTypes: [PrimeChecker] }
[INFO] Created _IsolateThread{ id: 1 ; registeredTasksTypes: [PrimeChecker] }
8779     -> true 	 PrimeChecker(8779)[finished]{ result: true ; executionTime: 2 ms }
1046527  -> true 	 PrimeChecker(1046527)[finished]{ result: true ; executionTime: 2 ms }
3139581  -> false 	 PrimeChecker(3139581)[finished]{ result: false ; executionTime: 0 ms }
16769023 -> true 	 PrimeChecker(16769023)[finished]{ result: true ; executionTime: 35 ms }

parallelism

Note that the parameter parallelism at AsyncExecutor shouldn't be greater than the CPU cores (unless you really have a case that justifies a higher number).

The most common error is to put a high number for the parameter parallelism thinking that this will increase tasks execution, but this actually can reduce your application performance. Note that each allocated Isolateneeds a CPU core to execute, and a high number of Isolate instances running tasks will be a bottleneck in your system (specially if you are demanding more Isolate than CPU cores).

SharedData

The class SharedData facilitates and optimizes data shared between tasks. The main advantage of data encapsulated with SharedData is to avoid multiple messages, with the same data, between threads/isolates, avoiding concurrency performance issues and multiple duplicated objects in memory (a GC bottleneck).

Here's an example of a task using SharedData:

// A task that checks if a number is prime:
class PrimeChecker extends AsyncTask<int, bool> {
  // The number to check if is prime.
  final int n;

  // A list of known primes, shared between tasks.
  final SharedData<List<int>, List<int>> knownPrimes;

  PrimeChecker(this.n, this.knownPrimes);

  // Instantiates a `PrimeChecker` task with `parameters` and `sharedData`.
  @override
  PrimeChecker instantiate(int parameters, [Map<String, SharedData>? sharedData]) {
    return PrimeChecker(
      parameters,
      sharedData!['knownPrimes'] as SharedData<List<int>, List<int>>,
    );
  }

  // The `SharedData` of this task.
  @override
  Map<String, SharedData> sharedData() => {'knownPrimes': knownPrimes};

  // Loads the `SharedData` from `serial` for each key.
  @override
  SharedData<List<int>, List<int>> loadSharedData(String key, dynamic serial) {
    switch (key) {
      case 'knownPrimes':
        return SharedData<List<int>, List<int>>(serial);
      default:
        throw StateError('Unknown key: $key');
    }
  }

  // The parameters of this task:
  @override
  int parameters() {
    return n;
  }

  // Runs the task code:
  @override
  FutureOr<bool> run() {
    return isPrime(n);
  }

  // A simple prime check function:
  bool isPrime(int n) {
    if (n < 2) return false;

    // The pre-computed primes, optimizing this checking algorithm:
    if (knownPrimes.data.contains(n)) {
      return true;
    }
    
    // It's sufficient to search for prime factors in the range [1,sqrt(N)]:
    var limit = (sqrt(n) + 1).toInt();

    for (var p = 2; p < limit; ++p) {
      if (n % p == 0) return false;
    }

    return true;
  }
}

The field knownPrimes above will be shared between tasks. In platforms with support for dart:isolate knownPrimes will be sent through an Isolate port only once, avoiding multiple copies and unnecessary memory allocations.

AsyncTaskChannel

An AsyncTask can have a communication channel that can be used to send/received messages during task execution.

To use a task channel just override channelInstantiator, than use channelResolved() inside the task and await channel() outside it:

class YourTask extends AsyncTask<String, int> {
  // ...
  
  @override
  AsyncTaskChannel? channelInstantiator() => AsyncTaskChannel(); // You can extend `AsyncTaskChannel` class if needed.

  // ...

  @override
  FutureOr<int> run() async {
    // ...
    
    // get the resolved channel:
    var channel = channelResolved()!; // The channel is always resolved inside `run()`.
    
    // Send and wait a message in the channel:
    channel.send('some message');
    var result = await channel.waitMessage<String>();
    
    // Send and wait in a single method:
    var result = await channel.sendAndWaitResponse<String, String>('some message');

    // Read a message if available in the queue (non-blocking):
    var msgOptional = channel.readMessage<String>();
    
    // ...
  }
}

Outside communication with the task:

void main() async {
  // ...
  
  // Execute the task:
  asyncExecutor.execute(task);
  
  // ...
  
  // Wait for the channel to be resolved:
  var channel = (await task.channel())!;
  
  // Wait for a message:
  var msg = await channel.waitMessage();
  
  // process msg...
  
  // Send a response:
  channel.send('Some response');
  
  // ...
}

An AsyncTaskChannel is automatically closed when a task finishes (returns its result).

Source

The official source code is hosted @ GitHub:

• https://github.com/eneural-net/async_task

Features and bugs

Please file feature requests and bugs at the issue tracker.

Contribution

Any help from the open-source community is always welcome and needed:

• Found an issue?
  • Please fill a bug report with details.
• Wish a feature?
  • Open a feature request with use cases.
• Are you using and liking the project?
  • Promote the project: create an article, do a post or make a donation.
• Are you a developer?
  • Fix a bug and send a pull request.
  • Implement a new feature, like other training algorithms and activation functions.
  • Improve the Unit Tests.
• Have you already helped in any way?
  • Many thanks from me, the contributors and everybody that uses this project!

If you donate 1 hour of your time, you can contribute a lot, because others will do the same, just be part and start with your 1 hour.

Author

Graciliano M. Passos: gmpassos@GitHub.

License

Apache License - Version 2.0