`APIError` is suggested to be returned from `stream.Recv()`

[hophacker]

In the example below, it is very helpful to get err type other than err message at some cases. For example, when your quota exceeded, the raw response is

{
    "error": {
        "message": "You exceeded your current quota, please check your plan and billing details.",
        "type": "insufficient_quota",
        "param": null,
        "code": null
    }
}

This error will be returned by stream.Recv() and the error message is generated as error, You exceeded your current quota, please check your plan and billing details.". Error message is changeable due to many reasons, thus it is imprecise to determine that the account has no quota through it. Instead, error type is needed here since it is supposed to be hard to change.

So, an error object to return is needed here, a.k.a, APIError is suggested to be returned from stream.Recv()

		stream, err := tunnel.OpenaiClient.CreateChatCompletionStream(cc, req)
		if err != nil {
			return err
		}
		defer stream.Close()

		for {
			response, err := stream.Recv()
			if errors.Is(err, io.EOF) {
				fmt.Println("\nStream finished")
				return nil
			}

			if err != nil {
				return err
			}
			w.Write([]byte(response.Choices[0].Delta.Content))
			w.(http.Flusher).Flush()
		}
	}

[hophacker]

never mind, I found a way to to that

func getErrType(err error) string {
	unwrappedError := errors.Unwrap(err)
	if unwrappedError != nil {
		if apiErr, ok := unwrappedError.(*openai.APIError); ok {
			return apiErr.Type
		}
	}
	return ""
}

[dannysievers]

This can be solved by using errors.As() as seen in the example README. Check out PR comments #293 for implementation details.

e := &openai.APIError{}
if errors.As(err, &e) {
  switch e.HTTPStatusCode {
    case 401:
      // invalid auth or key (do not retry)
    case 429:
      // rate limiting or engine overload (wait and retry) 
    case 500:
      // openai server error (retry)
    default:
      // unhandled
  }
}

[vvatanabe]

I'm closing this because it appears to have already been resolved.