otlptracehttp otlpmetrichttp: Handling 502 Bad Gateway and 504 Gateway Timeout

[pellared]

Description

From specification:

The requests that receive a response status code listed in following table SHOULD be retried.

HTTP response status code
429 Too Many Requests
502 Bad Gateway
503 Service Unavailable
504 Gateway Timeout

Actual behavior

opentelemetry-go/exporters/otlp/otlpmetric/otlpmetrichttp/client.go

Lines 141 to 198 in fc0a461

	return c.requestFunc(ctx, func(iCtx context.Context) error {
	select {
	case <-iCtx.Done():
	return iCtx.Err()
	default:
	}
	request.reset(iCtx)
	resp, err := c.httpClient.Do(request.Request)
	if err != nil {
	return err
	}
	var rErr error
	switch sc := resp.StatusCode; {
	case sc >= 200 && sc <= 299:
	// Success, do not retry.
	// Read the partial success message, if any.
	var respData bytes.Buffer
	if _, err := io.Copy(&respData, resp.Body); err != nil {
	return err
	}
	if respData.Len() != 0 {
	var respProto colmetricpb.ExportMetricsServiceResponse
	if err := proto.Unmarshal(respData.Bytes(), &respProto); err != nil {
	return err
	}
	if respProto.PartialSuccess != nil {
	msg := respProto.PartialSuccess.GetErrorMessage()
	n := respProto.PartialSuccess.GetRejectedDataPoints()
	if n != 0 || msg != "" {
	err := internal.MetricPartialSuccessError(n, msg)
	otel.Handle(err)
	}
	}
	}
	return nil
	case sc == http.StatusTooManyRequests, sc == http.StatusServiceUnavailable:
	// Retry-able failure.
	rErr = newResponseError(resp.Header)
	// Going to retry, drain the body to reuse the connection.
	if _, err := io.Copy(io.Discard, resp.Body); err != nil {
	_ = resp.Body.Close()
	return err
	}
	default:
	rErr = fmt.Errorf("failed to send metrics to %s: %s", request.URL, resp.Status)
	}
	if err := resp.Body.Close(); err != nil {
	return err
	}
	return rErr
	})

502 Bad Gateway and 504 Gateway Timeout is not be retried.

Expected behavior

Follow the specification.