pipeline

KO | EN

gitlite — search

#typescript #ai-agents #deepseek-harness #dsh-plugin #open-source #ai #claude-code #cli #windows #codex #developer-tools #dsh

★ 80 Open GitHub ↗

Concurrent iterators and pipeline operations.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# amphp/pipeline

AMPHP is a collection of event-driven libraries for PHP designed with fibers and concurrency in mind.
`amphp/pipeline` provides concurrent iterators and collection operators.

## Installation

This package can be installed as a [Composer](https://getcomposer.org/) dependency.

```bash
composer require amphp/pipeline
```

## Requirements

This package requires PHP 8.1 or later.

## Usage

Using fiber-based coroutines, asynchronous sets can now be created and consumed within a single fiber using PHP's built-in `Iterator`. Attempting to consume an `Iterator` instance from multiple fibers is problematic though, as one fiber may modify the state of the iterator while another is suspended.

This library provides a `ConcurrentIterator` interface which provides a fiber-safe iterator that may be consumed by multiple fibers concurrently, as well as tools for creating asynchronous sets.

### Concurrent Iterators

A `ConcurrentIterator` may be used in place of an `Iterator`, meaning it can be used with `foreach`, `yield from`, `iterator_to_array()`, argument unpacking, and more!

Like an `Iterator`, a `ConcurrentIterator` may also be iterated manually, with separate methods for advancing and retrieving the current value.

```php
use Amp\Pipeline\Pipeline;

// Pipeline::getIterator() returns a ConcurrentIterator
$concurrentIterator = Pipeline::fromIterable([1, 2, 3])->getIterator();
while ($concurrentIterator->continue()) {
    $position = $concurrentIterator->getPosition();
    $value = $concurrentIterator->getValue();

// ...
}

// Equivalently, multiple fibers may consume a single ConcurrentIterator
// instance using foreach.
$concurrentIterator = Pipeline::fromIterable([1, 2, 3])->getIterator();
foreach ($concurrentIterator as $position => $value) {
    // ...
}

```

`continue()` suspends the current fiber until a value becomes available or the iterator completes, returning `true` or `false` respectively. An exception is thrown from `continue()` the source of the iterator throws an exception while generating the next value.

`getValue()` returns the last value emitted on the iterator within the current fiber. The return value of this function will not change within the current fiber until `continue()` is called again. `continue()` must be invoked and return before this method can be called.

`getPosition()` returns the current 0-indexed position within the current iterator. If consuming from multiple fibers, this value may not be sequential within a single fiber. Similar to `getValue()`, `continue()` must be invoked and return before this method can be called.

> **Note**
> In general, it is not necessary to call these methods directly within application code. Concurrent iterators typically should be used with `foreach`.

### Queue

A `Queue` is used to create an asynchronous set of values with the ability to await consumption of the values produced, providing back-pressure to the production of more values, so consumption and production can be synchronized.

Values may be added to a `Queue` in two ways.
- `push()` adds the value to the queue, only returning once the value has been consumed from the queue.
- `pushAsync()` adds the value to the queue, returning a `Future` immediately which is completed only once the value has been consumed from the queue.

```php
use Amp\Pipeline\Queue;
use function Amp\async;
use function Amp\delay;

$queue = new Queue();

$start = \microtime(true);
$elapsed = fn () => \microtime(true) - $start;

// Generate values in a separate fiber
async(function () use ($queue, $elapsed): void {
    printf("Starting production loop at %.3fs\n", $elapsed());

foreach (range(1, 10) as $value) {
        delay(0.1); // Production of a value takes between 100ms
        $queue->push($value);
    }

printf("Completing production loop at %.3fs\n", $elapsed());

// Queue must be completed, otherwise foreach loop below will never exit!
    $queue->complete();
});

foreach ($queue->iterate() as $value) {
    printf("Iterator yielded %d at %.3fs\n", $value, $elapsed());
    delay(0.5); // Listener consumption takes 500 ms
}
```

Once all values have been pushed into a `Queue`, the producer must call `complete()` to end the concurrent iterator. Failure to do so will leave the consumer suspended indefinitely. Alternatively to indicate an error, the producer may use `error()` to throw an exception to the concurrent iterator consumer and end the concurrent iterator.

#### DisposedException

If the consumer of the concurrent iterator generated by the `Queue` is destroyed, `push()` will throw a `DisposedException` (or the future returned from `pushAsync()` will error with a `DisposedException`). This indicates that no additional values need to be generated since consumption of those values has ended. If for some reason the producer wishes to continue (e.g., to consume bytes from a buffer), either catch the exception or ignore the future. (The `DisposedException` instance is created only once for each queue.)

### Pipeline

A `Pipeline` represents an asynchronous set and provides operations which can be applied over the set.

```php
use Amp\Pipeline\Pipeline;
use function Amp\delay;

$pipeline = Pipeline::fromIterable(function (): \Generator {
    for ($i = 0; $i < 100; ++$i) {
        yield $i;
    }
});

$pipeline = $pipeline
    ->concurrent(10) // Process up to 10 items concurrently
    ->unordered() // Results may be consumed eagerly and out of order
    ->tap(fn () => delay(random_int(1, 10) / 10)) // Observe each value with a delay for 0.1 to 1 seconds, simulating I/O
    ->map(fn (int $input) => $input * 10) // Apply an operation to each value
    ->filter(fn (int $input) => $input % 3 === 0); // Filter only values divisible by 3

foreach ($pipeline as $value) {
    echo $value, "\n";
}
```

Alternatively, `Pipeline` also has methods which consume the set, such as `forEach()` or `reduce()`, which return only once the set is complete or throws an exception.

```php
use Amp\Pipeline\Pipeline;

Pipeline::generate(function (): int { static $v = 0; return ++$v; })
    ->take(10) // Take only 10 values from the generation function.
    ->concurrent(3) // Process 3 values concurrently
    ->delay(1) // Delay for 1 second to simulate I/O
    ->forEach(function (int $value): void {
        echo $value, "\n";
    });
```

#### Disposal

Iterators derived from another iterator, such as the iterator returned from `Pipeline::getIterator()` or created by intermediate operations like `map()`, are independent of their source: disposing the derived iterator does not dispose the source iterator, since the source may be shared with other consumers. For example, a limited number of values may be consumed from a source using `take()`, with the remaining values consumed from the source afterward.

Use `Pipeline::dispose()` to explicitly dispose the source of a pipeline, signaling any producers pushing to the source with a `DisposedException`. A source is also disposed automatically once the last reference to it is destroyed, so producers pushing to an abandoned source may remain suspended until the source is garbage collected.

## Versioning

`amphp/pipeline` follows the [semver](http://semver.org/) semantic versioning specification like all other `amphp` packages.

## Security

If you discover any security related issues, please email [`contact@amphp.org`](mailto:contact@amphp.org) instead of using the issue tracker.

## License

The MIT License (MIT). Please see [`LICENSE`](./LICENSE) for more information.

pipeline

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README