Download the PHP package webmozarts/console-parallelization without Composer

Parallelization for the Symfony Console

This library supports the parallelization of Symfony Console commands.

• How it works
• Installation
• Usage
• The API
  • The ParallelCommand and the Parallelization trait
  • Items
  • Segments
  • Batches
  • Configuration
  • Hooks
• Subscribed Services
• Differences with Amphp/ReactPHP
• Contribute
• Upgrade
• Authors
• License

How it works

When you launch a command with multiprocessing enabled, a main process fetches items and distributes them across the given number of child processes over the standard input. Child processes are killed after a fixed number of items (a segment) in order to prevent them from slowing down over time.

Optionally, the work of child processes can be split down into further chunks (batches). You can perform certain work before and after each of these batches (for example flushing changes to the database) in order to optimize the performance of your command.

Installation

Use Composer to install the package:

Usage

Add parallelization capabilities to your project, you can either extend the ParallelCommand class or use the Parallelization trait:

You can run this command like a regular Symfony Console command:

Or, if you want, you can run the command using parallelization:

The API

The ParallelCommand and the Parallelization trait

This library offers a ParallelCommand base class and a Parallelization trait. If you are looking for a basic usage, the ParallelCommand should be simpler to use as it provides the strictly required methods as abstract methods. All other hooks can be configured by overriding the ::configureParallelExecutableFactory() method.

The Parallelization trait on the other hand implements all hooks by default, requiring a bit less manual task. It does require to call ParallelizationInput::configureCommand() to add the parallelization related input arguments and options.

Items

The main process fetches all the items that need to be processed and passes them to the child processes through their Standard Input (STDIN). Hence, items must fulfill two requirements:

• Items must be strings
• Items must not contain newlines

Typically, you want to keep items small in order to offload processing from the main process to the child process. Some typical examples for items:

• The main process reads a file and passes the lines to the child processes
• The main processes fetches IDs of database rows that need to be updated and passes them to the child processes

Segments

When you run a command with multiprocessing enabled, the items returned by fetchItems() are split into segments of a fixed size. Each child processes process a single segment and kills itself after that.

By default, the segment size is the same as the batch size (see below), but you can try to tweak the performance of your command by choosing a different segment size (ideally a multiple of the batch size). You can do so by overriding the getSegmentSize() method:

Batches

By default, the batch size and the segment size are the same. If desired, you can however choose a smaller batch size than the segment size and run custom code before or after each batch. You will typically do so in order to flush changes to the database or free resources that you don't need anymore.

To run code before/after each batch, override the hooks runBeforeBatch() and runAfterBatch():

You can customize the default batch size of 50 by overriding the getBatchSize() method:

Configuration

The library offers a wide variety of configuration settings:

• ::getParallelExecutableFactory() allows you to completely configure the ParallelExecutorFactory factory which goes from fragment, batch sizes, which PHP executable is used or any of the process handling hooks.
• ::configureParallelExecutableFactory() is a different, lighter extension point to configure the ParallelExecutorFactory factory.
• ::getContainer() allows you to configure which container is used. By default, it passes the application's kernel's container if there is one. This is used by the default error handler which resets the container in-between each item failure to avoid things such as a broken Doctrine entity manager. If you are not using a kernel (e.g. outside a Symfony application), no container will be returned by default.
• ::createErrorHandler() allows you to configure the error handler you want to use.
• ::createLogger() allows you to completely configure the logger you want.

Hooks

The library supports several process hooks which can be configured via ::configureParallelExecutableFactory():

*: When using the Parallelization trait, those hooks can be directly configured by overriding the corresponding method.

Subscribed Services

You should be using subscribed services or proxies. Indeed, you may otherwise end up with the issue that the service initially injected in the command may end up being different from the one used by the container. This is because upon error, the ResetServiceErrorHandler error handler is used which resets the container when an item fails. As a result, if the service is not directly fetched from the container (to get a fresh instance if the container resets), you will end up using an obsolete service.

A common symptom of this issue is to run into a closed entity manager issue.

Differences with Amphp/ReactPHP

If you came across this library and wonder what the differences are with Amphp or ReactPHP or other potential parallelization libraries, this section is to highlight a few differences.

The primary difference is the parallelization mechanism itself. Amphp or ReactPHP work by spawning a pool of workers and distributing the work to those. This library however, spawns a pool of processes. To be more specific, the differences lies in how the spawn processed are used:

• An Amphp/ReactPHP worker can share state; with this library however you cannot easily do so.
• A worker may handle multiple jobs, whereas with this library the process will be killed after each segment is completed. To bring it to a similar level, it would be somewhat equivalent to consider the work of handling a segment in this library as a Amphp/ReactPHP worker task, and that the worker is killed after handling a single task.

The other difference is that this library works with a command as its central point. This offers the following advantages:

• No additional context need to be provided: once in your child process, you are in your command as usual. No custom bootstrap is necessary.
• The command can be executed with and without parallelization seamlessly. It is also trivial to mimic the execution of a child process as it is a matter of using the --child option and passing the child items via the STDIN.
• It is easier to adapt the distribution of the load and memory leaks of the task by configuring the segment and batch sizes.

Contribute

Contributions to the package are always welcome!

• Report any bugs or issues you find on the issue tracker.
• You can grab the source code at the package's Git repository.

To run the CS fixer and tests you can use the command make. More details available with make help.

Upgrade

See the upgrade guide.

Authors

• Bernhard Schussek
• Théo Fidry
• The Community Contributors

License

All contents of this package are licensed under the MIT license.