Skip to content

Latest commit

 

History

History
111 lines (83 loc) · 3.83 KB

README.md

File metadata and controls

111 lines (83 loc) · 3.83 KB
Raw

Parallel

This is a library for introducing Parallelization into your project. See the example.php file for an example how to use this library.

Theory of Operation

This library will enable the developer to execute a given amount of tasks (workers) in parallel. This is achieved by adding workers onto a manager, optionally defining how many processes to run simultaneously and then execute the manager.

Under Linux this library will try to detect the number of processors and allow a maximum number of processes to run equal to the number of processors. If this cannot be determined or the user is running Windows then a default of 2 is used.

Requirements and graceful degradation

Parallelization has several requirements. But to allow distribution, without adding several requirements to your application, will this library execute the given tasks in serie if the requirements are not met. And throw a E_USER_NOTICE php error that explains to the user that dependencies are missing.

The requirements for this library are:

Workers

Workers are basically wrappers around callback functions or methods. As such you can use anything in your existing project and parallelize it.

Do note that each parallel process is a duplicate of the original. This means that, for example, if you pass an object (or other reference) and change that, that the changes that you have made do not carry over to the caller.

The return value of the given callback is stored as result on the worker and can be read using the getResult() method.

Any exception that is thrown will result in an error, where the getReturnCode() method will return the exception code (be warned: this may be 0!) and the getError() method will return the exception message.

Errors and exceptions

if a task throws an exception it is caught and registered as an error. The exception's code is used as error number, where the message is used as error message.

By using this, instead of dying, you can continue execution of the other parallel processes and handle errors yourself after all processes have been executed.

Examples

Fluent interface

use MehrAlsNix\Parallel\Manager;
use MehrAlsNix\Parallel\Worker;

$mgr = new Manager();
$mgr
  ->addWorker(new Worker(function() { sleep(1); return 'a'; }))
  ->addWorker(new Worker(function() { sleep(1); return 'b'; }))
  ->addWorker(new Worker(function() { sleep(1); return 'c'; }))
  ->addWorker(new Worker(function() { sleep(1); return 'd'; }))
  ->addWorker(new Worker(function() { sleep(1); return 'e'; }))
  ->execute();

/** @var Worker $worker */
foreach ($mgr as $worker) {
    var_dump($worker->getResult());
}

Array interface

use MehrAlsNix\Parallel\Manager;
use MehrAlsNix\Parallel\Worker;

$mgr = new Manager();
$mgr[] = new Worker(function() { sleep(1); return 'f'; });
$mgr[] = new Worker(function() { sleep(1); return 'g'; });
$mgr[] = new Worker(function() { sleep(1); return 'h'; });
$mgr[] = new Worker(function() { sleep(1); return 'i'; });
$mgr[] = new Worker(function() { sleep(1); return 'j'; });
$mgr->execute();

/** @var Worker $worker */
foreach ($mgr as $worker) {
    var_dump($worker->getResult());
}

TODO

  • Improve docs

  • More intelligent process slots; currently only the oldest in a 'set' of slots is waited on but if this runs for a longer time then the other slots than those will not be filled as long as the first slot is occupied.

  • Last parts of IPC (Inter-Process Communication), to be able to return information from Workers to the Manager.

    • STDOUT
    • STDERR