566E PHP包仓库
首页 包列表 提交包 常见问题 关于

定制 hiblaphp/http-server 二次开发

按需修改功能、优化性能、对接业务系统，提供一站式技术支持

邮箱：yvsm@zunyunkeji.com | QQ：316430983 | 微信：yvsm316

hiblaphp/http-server

Composer 安装命令：

composer require hiblaphp/http-server

包简介

High Performance Pure PHP HTTP Server Library

README 文档

README

A high-performance, pristine, async-first HTTP server for PHP built on native PHP Fibers.

The Hibla HTTP Server is a low-level, strict RFC-compliant protocol engine. It is designed to be the foundational layer for framework builders, APIs, and microservices. It is intentionally unopinionated, meaning there is no built-in router, middleware pipeline, or PSR-7 mapping. It simply gives you raw speed, asynchronous non-blocking I/O, multi-core clustering, and absolute control over the HTTP request/response lifecycle.

The Hibla Ecosystem: This library is a core component of the broader Hibla Ecosystem. For advanced information on the underlying asynchronous primitives, check out the documentation for the Socket, Stream, and Promise libraries. Understanding Promise Cancellation is especially useful for building highly resilient, resource-efficient servers.

Latest Release Tests Total Downloads MIT License

Contents

Getting started

Server Configuration

Protocol-Level Security Hardening

Handling Requests

Sending Responses

Error Handling

Advanced Low-Level Control

Lifecycle Hooks

Clustered Mode (Multi-core)

API Reference

Meta

Installation

This package is currently in beta. Before installing, ensure your composer.json allows beta releases.

composer require hiblaphp/http-server

Requirements:

  • PHP 8.4+ (the library relies on native PHP Fibers and PHP 8.4 property hooks / asymmetric visibility).
  • The pcntl and posix extensions are required to support signal handling and graceful shutdown.
  • The proc_open function must be enabled in your php.ini if you intend to use multi-core clustering.

Quick start

use Hibla\HttpServer\HttpServer;
use Hibla\HttpServer\Message\Request;
use Hibla\HttpServer\Message\Response;
use function Hibla\await;

HttpServer::create('127.0.0.1:8000')
    ->withMaxBodySize(10 * 1024 * 1024) // 10MB
    ->onRequest(function (Request $request) {

        if ($request->method === 'POST' && $request->uri === '/api/echo') {
            // Await safely inside the handler; the server runs it in an async Fiber!
            $data = await($request->getJson());
            return Response::json(['you_sent' => $data]);
        }

        return Response::plaintext("Hello from the Server! You requested: {$request->uri}");
    })
    ->start();

Design Philosophy

This library is built like Node.js's node:http or Golang's net/http module. It operates as a strictly-scoped protocol engine.

  1. No Routing/Middleware: Use this library to build your own router or wrap it in a PSR-7 adapter.
  2. Strict RFC Compliance: Defends against Request Smuggling (TE.TE/TE.CL), malformed headers, and protocol downgrade attacks (RFC 9110/9112/9931 compliant).
  3. Async-First: Every request handler is executed inside an isolated Fiber. You can await() database queries, HTTP calls, or sleep() without blocking the server from accepting thousands of other concurrent connections.

Production Deployment (Reverse Proxies)

While the HTTP Server is highly secure and capable of binding directly to a public port, it is highly recommended to run this server behind a reverse proxy like Nginx, Caddy, or HAProxy in production environments.

Using a reverse proxy provides several architectural advantages:

  1. Privilege Dropping: It allows your PHP server to run safely on an unprivileged high port (like 8000) while the proxy binds to port 80 and 443 as root.
  2. Protocol Translation (HTTP/2 & HTTP/3): The HTTP Server natively speaks HTTP/1.1. Native HTTP/2 support is planned for a future release. However, proxies can accept HTTP/2 and HTTP/3 connections from the public internet and effortlessly translate them to HTTP/1.1 for your PHP backend, giving you immediate performance and multiplexing benefits.
  3. SSL/TLS Termination: Proxies are deeply optimized for SSL handshakes and certificate management.
  4. Static Asset Serving: Proxies can serve your static CSS, JS, and image files instantly without waking up the PHP runtime.

Streaming-by-Default Architecture

Unlike traditional PHP setups (such as Nginx + FPM or Apache) which block execution and pre-buffer the entire request body onto your disk or RAM before your script even starts, this server is streaming-by-default.

The exact millisecond the server finishes parsing the HTTP headers, it fires your onRequest handler. The request body is still coming over the TCP wire.

1. Client sends headers ──▶ [Server Parses] ──▶ 2. onRequest() fires immediately!
                                                     │
                                                     ├── (Body is still streaming over TCP)
                                                     │
                                                     └── 3. Your code decides:
                                                            ├── Abort immediately (unauthorized)
                                                            └── Or consume the stream asynchronously

Your Request->body is an active, unbuffered RequestBodyStream which implements the Hibla\Stream\Interfaces\ReadableStreamInterface. If you do not read it, the bytes are never buffered into memory, protecting your server from memory exhaustion attacks. If your handler returns without ever attaching a listener to a readable body, the connection manager automatically marks the response Connection: close and closes the stream for you. This ensures a client can't hold the socket open indefinitely with an unconsumed body.

Convenience Helpers

While the server is strictly streaming under the hood, writing raw chunk-reading loops for every request is tedious. The Request object provides highly optimized, asynchronous convenience helpers that automatically handle the stream buffering and parsing for you on-demand:

// 1. Buffers the streaming body into a string asynchronously
$rawBodyString = await($request->getBufferedBody());

// 2. Buffers the stream and parses it as a JSON array on-the-fly
$decodedJson = await($request->getJson());

// 3. Streams and parses multipart/form-data asynchronously
$form = await($request->getParsedBody());

Using these helpers gives you the simplicity of synchronous-looking code with the memory safety and speed of a non-blocking streaming engine. Cancelling any of these returned promises mid-stream will instantly cleanly abort the transfer and free up resources.

Server Configuration

The fluent builder

HttpServer::create() returns an immutable fluent builder. Every configuration method returns a new cloned instance, so you can safely chain configuration methods to build your ideal server state before calling ->start().

Security & Limits

Protect your server from memory exhaustion and abuse:

HttpServer::create()
    ->withMaxConnections(limit: 10000, pauseOnLimit: true) // Backpressure control
    ->withMaxBodySize(15 * 1024 * 1024)                    // 15MB max request body
    ->withHeaderLimits(maxSize: 8192, maxCount: 100)       // Prevent header bloat
    ->withMultipartLimits(maxFiles: 20, maxFields: 1000)   // Prevent hash collisions

The pauseOnLimit flag in withMaxConnections controls what happens once the connection cap is hit. Passing true applies backpressure at the OS level so new connections wait, while false accepts and immediately drops connections past the limit.

Timeouts & Slowloris Protection

Drop connections that are intentionally trickling data to tie up your sockets. All of these timeout features are disabled (null) by default to accommodate long-running streams, unless explicitly configured:

HttpServer::create()
    // Drop if headers take > 5s (Default: null / disabled)
    ->withHeaderTimeout(5.0)   
    
    // Drop if 10s pass between body chunks (Default: null / disabled)
    ->withBodyTimeout(10.0)    
    
    // Hard 60s absolute limit for the whole request (Default: null / disabled)
    ->withRequestTimeout(60.0)

Graceful Shutdown

When the server receives a termination signal (SIGINT or SIGTERM) during deployments or scaling events, it does not abruptly sever active user connections. Instead, the server stops accepting new connections and waits for all currently executing requests to finish cleanly.

You can configure the absolute maximum time the server will wait for active requests to finish before it forcefully exits:

HttpServer::create()
    // Max time to drain requests on SIGTERM (Default: 15.0 seconds)
    ->withGracefulShutdownTimeout(15.0)

Connection Persistence (Keep-Alive)

HttpServer::create()
    // Close idle connections after 5s (Default: null / disabled)
    ->withKeepAliveTimeout(5.0)       
    
    // Force reconnect after 100 requests (Default: null / unlimited)
    ->withKeepAliveMaxRequests(100)   
    
    // Set the HTTP/1.1 Pipelining depth queue (Default: 128)
    ->withMaxConcurrentRequestsPerConnection(128)

Important Proxy Configuration Note: If you are running this HTTP Server behind a reverse proxy like Nginx or HAProxy, you must ensure that your withKeepAliveTimeout() value is strictly greater than the proxy's configured keep-alive timeout.

If the PHP server closes the connection before the proxy expects it to, the proxy will encounter a closed socket when attempting to forward the next user's request. This race condition results in intermittent 502 Bad Gateway errors in production. Let the proxy manage the idle timeouts, and set the PHP server timeout slightly higher as a safety net.

HTTP/1.1 Pipelining

The HTTP Server supports HTTP/1.1 request pipelining out of the box. Multiple requests can arrive on the same connection before earlier ones have finished, and each is processed concurrently in its own Fiber. Responses are flushed back to the client strictly in the order the requests were received per RFC 9112. The server queues completed responses internally and only writes them once every earlier response in the pipeline is ready. Calling withMaxConcurrentRequestsPerConnection() bounds how deep that queue is allowed to grow per connection before the socket is paused.

HTTPS / TLS

HttpServer::create('0.0.0.0:443')
    ->withTls([
        'local_cert' => '/path/to/cert.pem',
        'local_pk'   => '/path/to/key.pem',
    ])

TLS metadata (negotiated protocol, cipher, and client certificate subject if mutual TLS is used) is exposed on $request->serverParams as SSL_PROTOCOL, SSL_CIPHER, and SSL_CLIENT_CERT_SUBJECT.

Custom Sockets & Testing

For unit/integration testing, or to plug in an already-configured transport, you can inject your own socket server directly. This bypasses withCluster(), single-process binding, and TLS configuration, giving you complete ownership of the socket lifecycle:

use Hibla\Socket\SocketServer;

$socket = new SocketServer('127.0.0.1:0'); // ephemeral port, useful in tests

HttpServer::create()
    ->withSocketServer($socket)
    ->onRequest(fn ($req) => Response::plaintext('ok'))
    ->start();

Calling withContext(array $context) merges recursively into any existing context on each call (array_merge_recursive), so successive calls accumulate rather than replace. This is useful for composing TCP-level options (e.g. so_reuseport) alongside TLS options.

Protocol-Level Security Hardening

These protections are always active, and there is nothing to opt into.

Request Smuggling Defenses

The protocol parser enforces RFC 9112 framing rules that specifically close known request-smuggling vectors:

  • Rejects requests carrying both Content-Length and Transfer-Encoding (CL.TE/TE.CL).
  • Rejects a chunked coding that appears anywhere except last in a Transfer-Encoding chain (TE.TE).
  • Rejects duplicate or conflicting Content-Length values, including comma-separated lists.
  • Rejects malformed Content-Length values (e.g. "10abc", "-5") instead of silently coercing them with a numeric cast.
  • Forces connection closure whenever an HTTP/1.0 request carries Transfer-Encoding at all, since no compliant 1.0 sender should ever produce one.
  • Enforces the HTTP token grammar on both the request method and every header field name, rejecting control characters and delimiter characters that a lenient upstream proxy might normalize differently.
  • Rejects obsolete line folding (obs-fold) and bare CR characters not followed by LF anywhere in the request line or header block.
  • A rejected CONNECT request (4xx/5xx) forces the connection closed immediately after the response, ensuring optimistically-pipelined bytes can never be misread as a new request.

Slowloris & Trickle-Attack Protection

  • headerTimeout bounds how long a client may take to finish sending headers.
  • bodyTimeout bounds inactivity between body chunks, protecting long uploads from stalling attackers without punishing genuinely slow connections.
  • requestTimeout bounds the entire request (headers + body) end-to-end.
  • Header block size is capped (maxHeaderSize, default 16 KiB) even before the terminating \r\n\r\n is seen. This prevents memory exhaustion from clients that never finish their headers.
  • Chunk-size lines in chunked transfer encoding are capped at 1 KiB to prevent unbounded buffer growth from a chunk-size line with no terminating CRLF.

Upload & Multipart Hardening

  • Multipart parts are structurally validated against RFC 7578 §4.2. Parts without Content-Disposition: form-data are parsed to keep boundary tracking correct, but they are never surfaced as a field or file.
  • Client-supplied filenames are sanitized to a bare leaf name before being exposed as UploadedFile::$clientFilename. Both POSIX (../../etc/evil.txt, /etc/passwd) and Windows-style (C:\evil.txt) path separators are stripped. Never trust this value as a storage path regardless.
  • withMultipartLimits() caps both file count and field count per request to prevent file-bombing and hash-collision attacks against your form-processing code.
  • Individual multipart header blocks are capped independently of the outer HTTP header limit.

Response Header Injection Guard

If your application code sets a response header value containing a raw \r, \n, or \0, the HTTP Server throws an InvalidResponseException rather than writing it to the wire. This prevents HTTP response splitting from user-controlled header values and surfaces the bug to you immediately instead of silently corrupting the response stream.

Handling Requests

Asynchronous Execution (Where can I await?)

Because this library is built on native PHP Fibers, several callbacks execute completely implicitly inside their own isolated Fiber. In these contexts, you can freely use await() to perform asynchronous non-blocking operations without stalling the main event loop.

You can safely await() inside:

  1. The $server->onRequest() callback.
  2. The $server->onError() callback.
  3. The $server->onClientDisconnect() callback (and $request->onClientDisconnect()).
  4. The onFile and onField callbacks provided to $request->streamMultipart().
  5. The callback provided to Response::sse().
  6. The onUpgrade callback provided to Response::upgrade().

CRITICAL: The Golden Rule (Never Block)

Because the Hibla Event Loop runs on a single thread utilizing cooperative multitasking, you must never use standard PHP blocking functions.

If you use a blocking function, the entire OS thread halts. The server will completely freeze and will not be able to accept new connections, read TCP data, or resume other Fibers until that blocking function finishes. You must always use the non-blocking async equivalents provided by the Hibla ecosystem.

THE WRONG WAY (Will freeze the entire server):

$server->onRequest(function (Request $request) {
    // FATAL: This completely halts the PHP process thread. 
    // No other users can connect or receive data for 2 seconds!
    sleep(2); 
    
    // FATAL: This blocks the thread waiting for network I/O.
    $data = file_get_contents('https://api.example.com/data'); 
    
    // FATAL: Standard PDO blocks the thread while waiting for the database.
    $pdo = new PDO('mysql:host=localhost;dbname=test', 'user', 'pass');
    
    return Response::plaintext($data);
});

THE RIGHT WAY (Non-blocking, cooperative multitasking):

$server->onRequest(function (Request $request) {
    // SAFE: This suspends only this specific request's Fiber.
    // The server instantly switches to serving other connected users.
    await(\Hibla\delay(2.0)); 
    
    // SAFE: Use the async HTTP client from the Hibla ecosystem.
    $response = await(\Hibla\HttpClient\Http::get('https://api.example.com/data'));
    
    return Response::plaintext($response->body());
});

If you really need to execute a blocking call without stalling the event-loop, you can use Hibla Parallel to offload blocking task to worker process, or use a dedicated non blocking queue library.

The Request Object

The Request object represents an incoming HTTP request.

$method = $request->method; // e.g. "POST"
$uri    = $request->uri;    // e.g. "/api/v1/users?active=true"
$version = $request->protocolVersion; // e.g. "1.1"

// Retrieve headers case-insensitively
$token = $request->getHeaderLine('Authorization');

The 100-Continue Handshake

If a client sends Expect: 100-continue, the server automatically writes HTTP/1.1 100 Continue back the moment your handler starts reading the body. This happens the first time something attaches a data listener to $request->body, or the first time you await() one of the buffered body helpers. You do not need to detect or respond to the Expect header yourself. If you never read the body, no 100-Continue is ever sent.

Body parsing

// Buffers the entire body into memory, rejecting with PayloadTooLargeException
// if it exceeds the server's max body size.
$raw = await($request->getBufferedBody());

// You can also override the global limit on a per-request basis
$largeRaw = await($request->getBufferedBody(maxBytes: 50 * 1024 * 1024));

// Buffers and json_decode()s the body; throws MessageParsingException on invalid JSON.
$data = await($request->getJson());

Multipart & Upload Handling

The HTTP Server provides two distinct, highly optimized models for handling multipart/form-data uploads. Both models defend against file-bombing and resource-exhaustion attacks out of the box.

Option A: Auto-Buffered Disk Uploads

This is the standard approach. It behaves like traditional PHP uploads but operates completely asynchronously without blocking the thread during disk writes.

Calling await($request->getParsedBody()) streams the multipart fields and writes file uploads into secure temporary files on disk (stored in sys_get_temp_dir()).

An Honest Note on Local Disk I/O (Fake Async): At the operating system level, standard local files do not support true non-blocking kernel watchers. When the server writes a file to your local disk, it uses cooperative time-slicing (writing the file in micro-chunks and yielding to the Event Loop instantly) to ensure the operation never blocks the server from processing other incoming requests. However, because PHP runs on a single thread, concurrent writes to the same physical disk will execute at roughly the same total speed as sequential writes due to hardware bounds. For massive high-traffic applications, consider Option B below to pipe data directly to cloud storage.

use Hibla\HttpServer\Message\MultipartForm;
use Hibla\HttpServer\Message\UploadedFile;

$form = await($request->getParsedBody());

// 1. Get standard form fields
$username = $form->get('username');

// 2. Access uploaded files
$file = $form->getFile('avatar'); // returns ?UploadedFile

if ($file instanceof UploadedFile) {
    echo $file->clientFilename;  // e.g. "profile_pic.jpg" (path-stripped, see hardening notes)
    echo $file->clientMediaType; // e.g. "image/jpeg" (client-supplied, unverified)
    echo $file->size;            // Size in bytes, counted while streaming to disk

    // Move the file asynchronously using pure stream pipes (non-blocking)
    await($file->moveTo('/var/www/uploads/' . $file->clientFilename));
}

Automatic Garbage Collection

To prevent disk leakages, the HTTP Server cleans up after itself. If an UploadedFile object goes out of scope and is garbage collected without you calling moveTo(), the underlying temporary file is automatically and instantly deleted from your disk.

Option B: Zero-Disk Async Streaming

This is the ultimate high-performance model. If you are uploading massive files (e.g. 5GB video files) and need to pipe them directly to S3 or an object storage service, writing them to your local disk first is a massive waste of I/O.

streamMultipart() bypasses the local disk entirely. It parses the incoming TCP stream on-the-fly and immediately hands you raw, async-readable streams for each file part as it arrives:

use Hibla\Stream\Interfaces\PromiseReadableStreamInterface;

await($request->streamMultipart(
    onFile: function (string $name, string $filename, string $mime, PromiseReadableStreamInterface $fileStream) {
        echo "Streaming file '{$filename}' directly to cloud storage...\n";

        // Read chunks asynchronously directly from the TCP socket as they land!
        // This callback is executed in an isolated Fiber, so await is safe.
        while (($chunk = await($fileStream->readAsync(8192))) !== null) {
            // Write directly to your S3 or Cloud stream
            await($s3UploadStream->writeAsync($chunk));
        }

        echo "Finished streaming {$filename}.\n";
    },
    onField: function (string $name, string $value) {
        // Standard form fields are delivered here as soon as they are parsed
        echo "Received field: {$name} = {$value}\n";
    }
));

Note: streamMultipart() actively enforces the exact same file and field limits configured via withMultipartLimits() to protect your server from resource exhaustion and hash collision attacks, completely mirroring the safety of the buffered getParsedBody() approach.

The MultipartForm Value Object

Returned by getParsedBody(). Per RFC 7578 §5.2, fields with duplicate names are never coalesced. Every submitted value is retained in submission order. Files sharing the same field name (multi-file upload) are likewise stored as a list.

$form->get('tags');       // first value only, or null
$form->getAll('tags');    // every value submitted for "tags", in order
$form->getFile('photos'); // first uploaded file for "photos", or null
$form->getFiles('photos');// every uploaded file for "photos", in order
$form->all();             // array<string, list<string>> of every field

Charset note: Values are returned as raw, undecoded bytes exactly as received on the wire. The parser does not interpret the RFC 7578 §4.5/§4.6 charset hints (per-part charset or a _charset_ field). No mainstream browser does either, since modern browsers submit UTF-8 regardless of page encoding. If you expect a non-UTF-8 charset, decode explicitly with a tool like mb_convert_encoding().

Sending Responses

The Response object represents an outgoing HTTP response. It is a lightweight value object.

use Hibla\HttpServer\Message\Response;

// Explicit response creation
$response = new Response(
    statusCode: 200,
    headers: ['Content-Type' => 'text/html'],
    body: '<h1>Welcome</h1>'
);

Response Factories

For standard workloads, always use the highly optimized factory methods:

return Response::plaintext('Hello');
return Response::json(['status' => 'created'], 201); // pretty-printed, unescaped slashes/unicode
return Response::html('<h1>Not Found</h1>', 404);
return Response::redirect('/login', 302);

If $response->body is set to any Hibla\Stream\Interfaces\ReadableStreamInterface instead of a string, the server automatically streams it out using chunked transfer encoding unless you have explicitly set a Content-Length header yourself.

Serving Static Files

The server provides a highly optimized, asynchronous static file serving factory. It automatically:

  • Detects the file's Content-Type safely from its extension.
  • Injects precise Content-Length headers.
  • Parses the request's Range header to serve HTTP 206 Partial Content. This allows browsers and mobile clients to scrub, seek, and buffer video or audio files perfectly with zero lag.
  • Returns a 404 Response::plaintext automatically if the file doesn't exist or isn't readable, so you don't need to check beforehand.

An Honest Note on Static File Serving: Just like local file uploads, serving static files directly from PHP relies on cooperative time-slicing. While this protects the event loop from stalling, reading hundreds of files concurrently on a single thread is bottlenecked by your server's local disk I/O. For production, it is strongly recommended to let your reverse proxy (Nginx/Caddy), a CDN, or an Object Storage service (S3) serve static files rather than tying up your PHP workers.

return Response::file('/var/www/public/video.mp4', $request);

Server-Sent Events (SSE)

Real-time streaming is natively supported. The SSE factory automatically configures non-buffering headers (Cache-Control: no-cache, X-Accel-Buffering: no), keep-alive parameters, and manages the background execution loop inside an isolated Fiber.

The SseStream provided to your callback implements Hibla\Stream\Interfaces\ReadableStreamInterface, making it fully compatible with the underlying stream engine.

use Hibla\HttpServer\Message\SseStream;

return Response::sse(function (SseStream $stream) {
    while ($stream->isReadable()) {
        $data = json_encode(['metrics' => getSystemMetrics()]);

        // Push the formatted SSE frame onto the wire
        $stream->send($data, event: 'metrics_update');

        // Sleep for 1 second asynchronously without blocking the event loop!
        await(\Hibla\delay(1.0));
    }
});

Calling $stream->ping($comment = 'ping') emits a spec-compliant comment line (: ping\n\n) to keep proxies and load balancers from timing out an idle connection. Both send() and ping() transparently suspend the emitter Fiber under TCP backpressure and resume once the client drains.

Protocol Upgrades (WebSockets)

For protocols like WebSockets or HTTP CONNECT tunneling, you can hijack the raw TCP connection entirely. Once you call upgrade(), the server detaches the socket from the HTTP loop and hands it directly to your callback:

return Response::upgrade(
    status: 101,
    headers: ['Upgrade' => 'websocket', 'Connection' => 'Upgrade'],
    onUpgrade: function ($connection, string $trailingBytes) {
        // You now own the raw Hibla\Socket\Interfaces\ConnectionInterface socket!
        $connection->write("Switched to WebSocket protocol.\n");

        $connection->on('data', function ($chunk) use ($connection) {
            $connection->write("Echo: " . $chunk);
        });
    }
);

Error Handling

Exception Hierarchy

All exceptions extend Hibla\HttpServer\Exceptions\HttpServerException (which is a RuntimeException), so you can catch broadly or narrowly depending on your needs.

HttpServerException
├── InvalidConfigurationException     // e.g. start() called with no onRequest handler
├── InvalidResponseException          // bad Response from handler/error-handler, or CRLF/NUL in a header value
├── JsonEncodingException             // Response::json() couldn't encode the given data
├── MessageParsingException           // malformed request line / headers (→ 400)
│   ├── RequestHeaderFieldsTooLargeException  // too many header fields (→ 431)
│   └── UnsupportedTransferCodingException    // unrecognized Transfer-Encoding coding (→ 501)
├── MultipartException                // generic multipart parser error
│   ├── MalformedMultipartException   // missing/invalid boundary
│   └── MultipartPartTooLargeException// a part's headers exceed the configured limit
├── PayloadTooLargeException          // body exceeds withMaxBodySize() (→ 413)
├── FileAlreadyMovedException         // UploadedFile::moveTo() called twice
├── UploadedFileNotFoundException     // UploadedFile::moveTo() but the temp file is gone
└── (Stream-related, thrown from underlying stream operations)
    ├── StreamClosedException
    ├── StreamNotWritableException
    └── StreamTransferException

These map directly onto the HTTP status codes noted above when they escape unhandled from protocol parsing. You generally only need to catch them yourself around the application-level helper calls (getJson(), getParsedBody(), UploadedFile::moveTo(), etc.):

use Hibla\HttpServer\Exceptions\PayloadTooLargeException;
use Hibla\HttpServer\Exceptions\MessageParsingException;

try {
    $data = await($request->getJson());
} catch (PayloadTooLargeException $e) {
    return Response::plaintext('Body too large', 413);
} catch (MessageParsingException $e) {
    return Response::plaintext('Invalid JSON', 400);
}

onError Behavior

Uncaught exceptions thrown inside your onRequest handler are routed to your onError callback (if registered). Otherwise, a generic 500 Internal Server Error plaintext body is sent.

Because an uncaught exception means internal state or streams may be unstable, any response returned from onError automatically gets Connection: close appended unless you already set it. The connection is not reused for a following request, even if you do not ask for that explicitly. If your onError callback itself throws an exception, or returns something other than a Response or null, the server falls back to the generic 500 response rather than propagating the failure.

Advanced Low-Level Control

The ProtocolHandlerInterface

For framework builders, the HTTP Server exposes the extreme low-level HTTP transport layer. Your onRequest() callback actually receives a second argument which is the ProtocolHandlerInterface.

This interface is the actual state machine driving the raw TCP socket. Accessing it directly lets you bypass the standard request-response loop for advanced networking, debugging, or custom protocols.

use Hibla\HttpServer\Interfaces\ProtocolHandlerInterface;
use Hibla\HttpServer\Message\Request;
use Hibla\HttpServer\Message\Response;

$server->onRequest(function (Request $request, ProtocolHandlerInterface $protocol) {
    // 1. Inspect the underlying socket connection directly
    $socket = $protocol->connection;
    $remoteIp = $socket->getRemoteAddress();

    // 2. Check active pipeline concurrency on this specific socket
    $concurrentRequestsOnSocket = $protocol->activeRequestsCount;

    // 3. Write responses with explicit event callbacks
    $protocol->writeResponse(Response::plaintext('Direct Write'), function () {
        // This callback is executed the exact millisecond the bytes leave the OS buffer!
        echo "Response fully transmitted to client.\n";
    });
});

Manual Connection Throttling (Backpressure)

One of the major benefits of accessing the raw $protocol->connection is the ability to apply manual backpressure (TCP flow control). If a client is uploading a massive file or flooding the server with requests faster than your database or downstream consumers can process them, you can tell the OS kernel to stop reading packets from the TCP socket by calling pause(). Once your queue clears, call resume() to start reading again:

$server->onRequest(function (Request $request, ProtocolHandlerInterface $protocol) {
    $socket = $protocol->connection;

    // Stop reading any new TCP packets from this client!
    // The client's OS will buffer data locally (TCP Window saturation)
    $socket->pause();

    Hibla\async(function () use ($socket, $request) {
        // Process a slow, database i/o task asynchronously
        await(slowDatabaseWrite(await($request->getBufferedBody())));

        // We are ready for more data. Tell the kernel to resume reading!
        $socket->resume();
    });

    return Response::plaintext('Processed');
});

Hijacking the Connection (Socket Detachment)

If you need to transition the socket into an entirely different protocol (such as WebSockets, SSH tunnelling, or custom binary framing), you can explicitly detach the protocol handler.

Detaching stops all HTTP parsing, cancels any associated HTTP timeouts, and returns any unparsed bytes currently sitting in the receive buffer:

$server->onRequest(function (Request $request, ProtocolHandlerInterface $protocol) {
    if ($request->getHeaderLine('Upgrade') === 'my-custom-protocol') {

        // Send the HTTP protocol switch response
        $protocol->writeResponse(new Response(101, [
            'Upgrade' => 'my-custom-protocol',
            'Connection' => 'Upgrade'
        ]));

        // Hijack and detach!
        $rawSocket = $protocol->connection;
        $unparsedBytes = $protocol->detach(); // Cleans up and returns trailing bytes

        // The HTTP server has completely forgotten about this socket.
        // You are now writing raw TCP data:
        if ($unparsedBytes !== '') {
            processCustomFraming($unparsedBytes);
        }

        $rawSocket->on('data', function (string $chunk) use ($rawSocket) {
            $rawSocket->write("Echo: " . $chunk);
        });

        return; // Return null so the server knows not to try to send a response
    }

    return Response::plaintext('Standard HTTP');
});

Lifecycle Hooks

onRequest

The primary hook. Called every time a complete HTTP header block is parsed. Executed safely in its own dedicated Fiber.

onError

Catch unhandled exceptions thrown by your onRequest callback and format a proper response. See onError Behavior above for the automatic Connection: close caveat.

$server->onError(function (\Throwable $e, Request $request) {
    return Response::json(['error' => $e->getMessage()], 500);
});

onClientDisconnect & Promise Cancellation

Triggered if the client drops the TCP connection before your onRequest handler has finished responding. Because Hibla is built on the hiblaphp/promise library, you can safely cancel long-running promises to save CPU cycles and database connections if a user abandons their request early.

$server->onRequest(function (Request $request) {
    // Start a long running downstream API call or database query
    $downstreamPromise = fetchAnalyticsDataAsync();

    // If the user closes their browser tab, instantly cancel the query!
    $request->onClientDisconnect(function () use ($downstreamPromise) {
        $downstreamPromise->cancel();
    });

    try {
        $data = await($downstreamPromise);
        return Response::json($data);
    } catch (\Hibla\Promise\Exceptions\CancelledException $e) {
        // The promise was cancelled cleanly.
        return null;
    }
});

onStart

Fired exactly once, right before the server binds to the socket and starts accepting connections. In clustered mode, this fires inside each isolated worker subprocess. See the Serialization Trap below for why this matters.

Clustered Mode (Multi-core)

Node.js and traditional PHP share a limitation in that an event loop only runs on a single CPU core. The HTTP Server solves this natively using SO_REUSEPORT and hiblaphp/parallel.

You can fork multiple worker processes that natively share the TCP port load, maximizing CPU utilization:

// Spawns 8 independent worker processes handling requests simultaneously
HttpServer::create('0.0.0.0:8000')
    ->withCluster(8)
    ->onRequest(...)
    ->start();

Clustering is not supported on Windows because there is no SO_REUSEPORT functionality. The HTTP Server detects this automatically and falls back to single-process mode with a warning, rather than failing to start.

How Clustering Works

When you call withCluster(8), the Master process serializes your onRequest, onError, and onClientDisconnect closures, sends them over IPC pipes, and forks 8 child workers. The children deserialize those closures and run the actual HTTP servers.

The Serialization Trap (Dos & Don'ts)

Because your closures are serialized and sent across process boundaries, they cannot capture active OS resources like open PDO database connections, File handles, or Redis clients from the Master process.

THE WRONG WAY (Will crash in Cluster Mode):

// Executed in the MASTER PROCESS
$logger = new FileLogger('/var/log/app.log');

HttpServer::create('127.0.0.1:8000')
    ->withCluster(4)
    ->onError(function (\Throwable $e, Request $request) use ($logger) {
        // ERROR: The Parallel library cannot serialize the $logger file resource
        // to send it to the child workers! This will throw a TaskPayloadException.
        $logger->log($e->getMessage());
    })
    ->start();

THE RIGHT WAY (Using onStart): To make this work cleanly, initialize your connections inside the worker process using the onStart hook.

HttpServer::create('127.0.0.1:8000')
    ->withCluster(4)
    ->onStart(function () {
        // This runs INSIDE the child worker right after it boots.
        // It is perfectly safe to open Database/File resources here.
        global $workerLogger;
        $workerLogger = new FileLogger('/var/log/app.log');
    })
    ->onError(function (\Throwable $e, Request $request) {
        // Safe to serialize: captures no external scope!
        global $workerLogger;
        $workerLogger->log($e->getMessage());
        return Response::plaintext('Internal Error', 500);
    })
    ->start();

Note: Framework builders can also use Dependency Injection containers or Static Facades initialized via ClusterOptions::withClusterBootstrap(), as static calls do not capture scope.

Cluster Options & IPC

Pass a ClusterOptions object to configure worker environments and Inter-Process Communication (IPC):

use Hibla\HttpServer\ClusterOptions;

$options = ClusterOptions::make()
    ->withWorkerMemoryLimit('256M')
    ->withWorkerRestartLimit(10) // Prevent fork-bomb crash loops
    ->withClusterBootstrap('/path/to/autoload.php') // Preload legacy code
    ->onWorkerMessage(function ($message) {
        // Receive messages emitted by workers to the Master process
        echo "Worker {$message->pid} says: {$message->data}\n";
    });

HttpServer::create()
    ->withCluster(4, $options)
    ->onRequest(function(Request $request) {
        \Hibla\emit('I just served a request!'); // Send message to Master via IPC
        return Response::plaintext('OK');
    })
    ->start();

If a worker dies unexpectedly, the Master automatically respawns a replacement worker (bounded by withWorkerRestartLimit() to avoid crash-loop fork bombs) and logs the active PID list.

API Reference

HttpServerInterface API

All fluent configuration methods return a new cloned instance of the server.

Method Return Type Description
HttpServer::create(string|int $address) HttpServerInterface Named constructor. Spawns a new server instance. Accepts a port (8080) or full URI (127.0.0.1:8000).
withSocketServer(ServerInterface $socket) static Inject a custom pre-bound socket server instance (useful for testing). Disables clustering.
withContext(array $context) static Configure raw stream/socket context options. Merges recursively across calls.
withTls(array $tlsOptions) static Configure SSL/TLS parameters.
withCluster(int $workers, ?ClusterOptions $opts) static Enable multi-process worker clustering via SO_REUSEPORT (no-op on Windows).
withoutCluster() static Run entirely inside the parent thread (disables clustering).
withoutLogging() static Disable built-in CLI starting logs.
withMaxBodySize(int $bytes) static Maximum allowed request body size (Default: 10MB).
withMaxConnections(int $limit, bool $pause) static Set concurrency backpressure bounds.
withHeaderLimits(int $size, int $count) static Maximum allowed header block size and count limits.
withHeaderTimeout(?float $seconds) static Slowloris protection timeout for request headers.
withBodyTimeout(?float $seconds) static Timeout between receiving active request body chunks.
withRequestTimeout(?float $seconds) static Absolute timeout allowed for a complete request (Headers + Body).
withKeepAliveTimeout(?float $seconds) static Max idle duration allowed for persistent connections.
withKeepAliveMaxRequests(?int $limit) static Max requests allowed over a single connection.
withGracefulShutdownTimeout(float $sec) static Max time allowed to drain in-flight requests on SIGTERM.
withMaxConcurrentRequestsPerConnection(int $n) static Limit pipelining queue depth per TCP socket.
withMultipartLimits(int $files, int $fields) static Limit files and fields inside multipart/form-data.
onRequest(callable $handler) static Register the primary async HTTP request handler.
onError(callable $handler) static Register a global error interceptor callback.
onClientDisconnect(callable $callback) static Register a callback for aborted requests.
onStart(callable $callback) static Register a late-stage worker initialization callback.
start() void Binds the sockets and starts the Event Loop (blocking). Throws InvalidConfigurationException if no handler is set.

ClusterOptions API

Method Return Type Description
ClusterOptions::make() ClusterOptions Named constructor for fluent chain building.
withWorkerMemoryLimit(string $limit) static Set memory limits (ini_set) inside spawned workers.
withWorkerRestartLimit(?int $limit) static Max worker restarts allowed per second (Default: 10).
withClusterBootstrap(string $file, ?callable $cb) static Specify a bootstrap file to run before workers boot.
onWorkerMessage(callable $handler) static Register an IPC handler to receive worker emit() payloads.

ProtocolHandlerInterface API

Exposed as the second argument to onRequest(). Controls raw TCP transport-to-HTTP mapping.

Property / Method Return Type / Signature Description
$connection ConnectionInterface Property (Read-only). The underlying raw socket.
$activeRequestsCount int Property (Read-only). Active concurrent requests on this socket.
isUpgraded() bool Checks if the connection has been hijacked/upgraded.
writeResponse (Response $response, ?callable $onComplete = null): void Serializes and writes a Response directly onto the wire.
detach() (): string Halts HTTP parsing and releases the raw socket. Returns unparsed buffer bytes.
gracefulShutdown() (): void Signals the handler to cleanly finish and disconnect.

Request API

The value object representing an incoming HTTP Request.

Property / Method Return Type / Signature Description
$method string Property (Read-only). The HTTP method (e.g. "POST").
$uri string Property (Read-only). Raw target URI path and query string.
$protocolVersion string Property (Read-only). Protocol version (e.g. "1.1").
$serverParams array<string, string> Property (Read-only). Environment parameters (REMOTE_ADDR, REMOTE_PORT, SERVER_ADDR, SERVER_PORT, REQUEST_SCHEME, HTTPS, SSL_PROTOCOL, SSL_CIPHER, SSL_CLIENT_CERT_SUBJECT).
$body RequestBodyStream Property (Read-write). The unbuffered body stream. Can be manually reassigned or piped.
hasHeader (string $name): bool Check if a specific header exists (case-insensitively).
getHeader (string $name): string[] Get list of values for a specific header name.
getHeaderLine (string $name): string Get header values flattened into a comma-separated string.
getBufferedBody (?int $maxBytes = null): PromiseInterface<string> Asynchronously buffer the entire request body into a string. Throws PayloadTooLargeException if exceeded.
getJson (?int $maxBytes = null): PromiseInterface<mixed> Asynchronously buffer and decode the body as JSON. Throws MessageParsingException on invalid JSON.
getParsedBody (): PromiseInterface<MultipartForm> Asynchronously parse multipart data, buffering files to temp disk.
streamMultipart (callable $onFile, ?callable $onField): PromiseInterface<void> Stream multipart on-the-fly directly to memory (zero-disk).
onClientDisconnect (callable $callback): static Register a callback to fire if the TCP socket drops early.
isDisconnected (): bool Checks if the client has already disconnected.

Response API

The value object representing an outgoing HTTP Response.

Factory / Method Return Type / Signature Description
Response::plaintext (string $text, int $status = 200): Response Factory. Create a plaintext response.
Response::json (mixed $data, int $status = 200): Response Factory. Create a JSON response. Throws JsonEncodingException if the data can't be encoded.
Response::html (string $html, int $status = 200): Response Factory. Create an HTML response.
Response::redirect (string $url, int $status = 302): Response Factory. Create a redirect response.
Response::file (string $path, ?Request $req = null, array $headers = []): Response Factory. Async, stream-based file serve with HTTP 206 support. Returns a 404 automatically if unreadable.
Response::sse (callable $emitter): Response Factory. Ergonomic Server-Sent Events stream.
Response::upgrade (int $status, array $hdr, callable $onUpgrade): Response Factory. Hijack the TCP socket for custom protocols (WebSockets).
hasHeader (string $name): bool Checks if a header exists.
getHeader (string $name): string[] Get values of a specific header.
getHeaderLine (string $name): string Get comma-separated string of header values.
setHeader (string $name, string|array $val): void Overwrite or set a header. Throws InvalidResponseException on CR/LF/NUL in the value.
addHeader (string $name, string|array $val): void Append values onto an existing header.

SseStream API

Passed to the callback of Response::sse(). Implements Hibla\Stream\Interfaces\ReadableStreamInterface.

Method Return Type / Signature Description
send (string $data, ?string $event = null, ?string $id = null, ?int $retry = null): void Safely formats and pushes an SSE message to the client. Suspends under backpressure.
ping (?string $comment = 'ping'): void Emits a standard-compliant SSE comment block to keep the connection alive.
isReadable (): bool Checks if the stream is still open and readable.
close (): void Forcefully closes the SSE stream.
end (): void Ends the stream gracefully.

MultipartForm API

Returned by Request::getParsedBody().

Method Return Type / Signature Description
get (string $name): ?string First value submitted for the field, or null.
getAll (string $name): list<string> Every value submitted for the field, in order.
getFile (string $name): ?UploadedFile First uploaded file for the field, or null.
getFiles (string $name): list<UploadedFile> Every uploaded file for the field, in order (handles both name and name[]).
all (): array<string, list<string>> Every field and its submitted values.

UploadedFile API

Represents a parsed file upload buffered asynchronously to your temp disk.

Property / Method Return Type / Signature Description
$tmpPath string Property (Read-only). Temporary storage path on the local disk.
$clientFilename string Property (Read-only). Client-supplied file name, directory path stripped (see hardening notes).
$clientMediaType string Property (Read-only). Client-supplied MIME type. This is unverified and spoofable, so use content-based detection if it matters.
$size int Property (Read-only). Total size of the uploaded file in bytes.
moveTo (string $destinationPath): PromiseInterface<void> Asynchronously move the temporary file to its final target path. Throws FileAlreadyMovedException or UploadedFileNotFoundException.

Exceptions Reference

All under Hibla\HttpServer\Exceptions\, all extend HttpServerException (a RuntimeException).

Exception Typically thrown when Resulting status (if unhandled)
InvalidConfigurationException start() called with no request handler; invalid limit values passed to builder methods
InvalidResponseException Handler returns a non-Response; header value contains CR/LF/NUL 500
JsonEncodingException Response::json() given data json_encode can't serialize 500
MessageParsingException Malformed request line, headers, or Content-Length 400
RequestHeaderFieldsTooLargeException Too many header fields, or a field name too long 431
UnsupportedTransferCodingException Unrecognized Transfer-Encoding coding 501
MultipartException Generic multipart stream/parser failure, or too many files/fields
MalformedMultipartException Missing or invalid multipart boundary
MultipartPartTooLargeException A multipart part's headers exceed the configured limit
PayloadTooLargeException Body exceeds withMaxBodySize() 413
FileAlreadyMovedException UploadedFile::moveTo() called a second time
UploadedFileNotFoundException UploadedFile::moveTo() called after the temp file is gone
StreamClosedException Read/write attempted on an already-closed stream
StreamNotWritableException Write attempted on a non-writable destination during a pipe
StreamTransferException Destination closed before a stream transfer completed

Development

If you would like to contribute to the HTTP Server or run the test suite locally, you can clone the repository and use Composer to install the development dependencies:

git clone https://github.com/hiblaphp/http-server.git
cd http-server
composer install

Running Tests The project uses Pest for testing.

./vendor/bin/pest

Static Analysis The project strictly enforces PHPStan at the maximum level.

./vendor/bin/phpstan analyse

Code Formatting The codebase follows PSR-12 standards enforced by Laravel Pint.

./vendor/bin/pint

License

MIT License. See LICENSE for more information.

统计信息

  • 总下载量： 0
  • 月度下载量： 0
  • 日度下载量： 0
  • 收藏数： 2
  • 点击次数： 1
  • 依赖项目数： 0
  • 推荐数： 0

GitHub 信息

  • Stars： 2
  • Watchers： 0
  • Forks： 0
  • 开发语言： PHP
访问仓库

其他信息

  • 授权协议： MIT
  • 更新时间： 2026-07-10

承接程序开发

PHP开发

VUE

Vue开发

前端开发

小程序开发

公众号开发

系统定制

数据库设计

云部署

网站建设

安全加固