Connection Interruption Troubleshooting

Image generation is a long request (10–60 seconds), and connection interruptions are mostly caused by client-side or intermediary timeouts. Troubleshoot in the following order.

Troubleshooting Checklist

1. HTTP client timeout: The default timeout for requests / axios / fetch is generally 30 seconds, which is shorter than the generation time and will definitely cause failures. Set it uniformly to ≥120 seconds:

   client = OpenAI(base_url="...", api_key="...", timeout=180)

2. Self-hosted reverse proxy timeout: nginx defaults to proxy_read_timeout 60s, which also needs to be increased:

   proxy_read_timeout 180s;
   proxy_send_timeout 180s;

3. Response body size limit: When returning base64, the response can reach several MB. Check the proxy and gateway client_max_body_size / body limits;

4. Company network / VPN: Some corporate proxies will cut off long connections; compare with a direct network connection;

5. Too much concurrency: Under heavy concurrency, some requests are queued. After the total time is prolonged, timeouts are triggered— reduce concurrency or add a queue.

Correct Handling After Interruption

• Failed requests are not billed (the pre-authorization is automatically refunded), so it is safe to retry;
• Retry with exponential backoff (5s → 15s → 45s), do not resend immediately in quick succession;
• Occasional 5xx errors during peak periods are upstream instability. Retrying 1–2 times usually recovers; see Error Codes and Retries.

Still Not Resolved

Check the status of the request in the console “Logs”; attach the request time and model name, and contact support to investigate channel-side issues.