Download the PHP package phasync/server without Composer

phasync/server

This library makes it very easy to create an efficient TCP or UDP server in PHP. It wraps the \stream_socket_server() with sensible configuration details and enables you to handle multiple concurrent connections very easily.

Example HTTP Server

Performance

This implementation performs very well, and is in many cases faster than Node.js at handling concurrent connections (up to 50% more requests per second handled in synthetic benchmarks). This performance is due to leveraging phasync coroutines and non-blocking IO (the phasync::readable() and phasync::writable() calls in the example above), along with PHP 8.3 JIT.

Class Overview

The Server class provides a simplified interface for creating efficient network servers. It handles the complexities of socket creation, configuration, and connection management, allowing you to focus on your application logic.

Class Methods

public static function serve(string $address, ?Closure $socketHandler = null, ?float $timeout = null): Fiber

Creates and starts a TCP or UDP server. For TCP or UNIX socket servers, the handler function must launch a coroutine to handle many connections concurrently.

Parameters:

• $address (string): The address to bind to, in the format protocol://host:port (e.g., tcp://127.0.0.1:8080 or udp://0.0.0.0:9000).
• $handler (Closure): For TCP connections, the closure will be run inside a coroutine receiving the stream resource and the peer address as arguments. For UDP connections, the closure will run inside a coroutine receiving the Server instance as an argument.
• $timeout (float|null): Optional timeout for socket operations.

Returns:

• Fiber

TCP Example

UDP Example

Design Decisions

Disabled Read and Write Buffering and Large Chunk Size

Instead of having PHP manage buffering of reads and writes, developers should try to read and write big chunks of data. Don’t try to read 10 bytes, read 65536 bytes at a time and do the buffering yourself if needed.

When you read big chunks, the benefit of buffering is gone and the end result is slightly faster. Also, the socket works more like a developer would expect.

In short:

• Write big chunks for bandwidth
• Write short chunks for low latency
• Always read big chunks and process or buffer the data yourself

so_reuseport

This is enabled by default. The assumption is that we want to avoid having the network port be busy for a while after the application terminates, for example due to an error. Also, this enables other processes owned by the same user to start listening to the same port, simplifying scaling up the application to multiple processes.

TCP Nagle Algorithm

By default, the Nagle algorithm is disabled. This decision is based on the assumption that if a developer writes a short message to a socket, the developer wants that message delivered quickly. For messages that are longer than a network packet, Nagle has little effect. So generally, we are proponents of letting the developer build larger chunks before writing if the developer intends to, for example, send a full HTML response. See also the chapter above about disabled read/write buffering.

To enable the Nagle algorithm you must pass a custom stream context to the phasync\Server constructor.