[Bug]: fopen(https://172.66.0.243/v1/chat/completions): Failed to open stream: HTTP request failed!

[tw-ict]

Description

When doing the request, i sometimes get a error thrown "fopen(https://172.66.0.243/v1/chat/completions): Failed to open stream: HTTP request failed!"

Steps To Reproduce

public static function chat(string $systemChat, string $userChat, string $responseType = 'text')
  {

    $parameters = [
      'model' => 'gpt-4o-mini',
      'messages' => [
        ['role' => 'system', 'content' => $systemChat],
        ['role' => 'user', 'content' => $userChat],
      ],
      'temperature' => 0.002,
    ];

    if ($responseType == 'json') {
      // Tell OpenAI to return the response as a JSON object
      $parameters['response_format'] = ['type' => 'json_object'];
    }
    $client = self::getClient();
    $result = $client->chat()->create($parameters);

    if ($responseType == 'json') {
      return json_decode($result->choices[0]->message->content, true);
    }


    return $result->choices[0]->message->content;
  }

OpenAI PHP Client Version

V0.10.3

PHP Version

8.3.1

Notes

No response

[iBotPeaches]

What are you looking to achieve with the report? The library adapts PSR18 and finds a valid client - so you can design retry logic around operations.

Your URL is also odd, it looks like the Cloudflare IP associated with OpenAI - is that intentional? Either way I don't think this is a bug and instead a support request for handling retries.

[tw-ict]

To be completely honest, i don't know. I assumed it was the client taking the IP adress of the server and trying to access it directly, but that doesn't seem to be the case.

Where do you think this error should be resolved? Is it with OpenAI sending a wrong IP back? Or is it my computer which is doing weird things?

I just would like this error not happening, obviously :)

[iBotPeaches]

I believe its on your side/OpenAI. We send probably 20-30k requests a day and maybe get 10 or so a day that just timeout/idle from OpenAI. So the service isn't perfect so you have to design some logic around outbound calls to retry/timeout at levels that make sense for your system.

I'm not quite sure why your system has the URL logged, but I use Guzzle as my HTTP backing which probably prefers curl under the hood. With your output have fopen, I guess you use a stream handler.