Documentation Packages Getting Started FAQ Support

Http Server ⋅ Performance Tips

Bottlenecks

The HTTP server won’t be the bottleneck. Misconfiguration, use of blocking I/O, or inefficient applications are.

The server is well-optimized and can handle tens of thousands of requests per second on typical hardware while maintaining a high level of concurrency of thousands of clients.

But that performance will decrease drastically with inefficient applications. The server has the nice advantage of classes and handlers being always loaded, so there’s no time lost with compilation and initialization.

A common trap is to begin operating on big data with simple string operations, requiring many inefficient big copies. Instead, streaming should be used where possible for larger request and response bodies.

The problem really is CPU cost. Inefficient I/O management (as long as it is non-blocking!) is just delaying individual requests. It is recommended to dispatch simultaneously and eventually bundle multiple independent I/O requests via Amp’s combinators, but a slow handler will slow down every other request too. While one handler is computing, all the other handlers can’t continue. Thus it is imperative to reduce computation times of the handlers to a minimum.

Disconnecting Clients

This section will discuss disconnecting clients with streamed response bodies.

This section is currently undocumented or incomplete. Please help improving this situation by filing a pull-request.

Body

RequestBody provides buffered and streamed access. Use the streamed access to handle large messages, which is particularly important if you have larger message limits (like tens of megabytes) and don’t want to buffer it all in memory. If multiple people are uploading large bodies concurrently, the memory might quickly get exhausted.

Hence, incremental handling is important, accessible via the read() API of Amp\ByteStream\InputStream.

In case a client disconnects, the read() fails with an Amp\Http\Server\ClientException. This exception is thrown for both the read() and buffer() API.

ClientExceptions do not need to be caught. You may catch them if you want to continue, but don’t have to. The Server will silently end the request cycle and discard that exception then.

Instead of setting the generic body limit high, you should consider increasing the body limit only where needed, which is dynamically possible with the increaseMaxSize() method on RequestBody.

<?php

use Amp\Http\Server\RequestHandler\CallableRequestHandler;
use Amp\Http\Server\ClientException;
use Amp\Http\Server\Request;
use Amp\Http\Server\Response;
use Amp\Http\Status;
use Amp\ByteStream;
use Amp\File;

$handler = new CallableRequestHandler(function (Request $request) {
    $path = "test.txt";
    
    try {
        /** @var File\Handle $file */
        $file = yield File\open($path, "w+");
        $body = $request->getBody();
        $body->increaseMaxSize(10 * 1024 ** 2); // 10 MB

        yield ByteStream\pipe($body, $file);

        return new Response(Status::OK, [], "OK, saved.");
    } catch (ClientException $e) {
        // Writes may still arrive, even though reading stopped
        if ($e->getCode() === Status::PAYLOAD_TOO_LARGE) {
            return new Response(Status::PAYLOAD_TOO_LARGE, [], "Too big, aborting.");
        }
        
        if (isset($file)) {
            // explicit close to avoid issues when unlink()'ing
            yield $file->close();
            yield File\unlink($path);
        }

        throw $e; // Don't care to return a response
    }
});