SEZ9 commented on PR #11548: URL: https://github.com/apache/seatunnel/pull/11548#issuecomment-5082913303
@DanielLeens Thanks for checking the endpoint against the docs — this one deserves a careful answer because the two sources genuinely disagree, so I re-verified empirically against the live endpoint (us-east-1, `openai.gpt-5.6-terra`) before responding: - `https://bedrock-mantle.us-east-1.api.aws/openai/v1` (current code) → **works**, returns a completed response. - `https://bedrock-mantle.us-east-1.api.aws/v1` (proposed) → **400**: `"The model 'openai.gpt-5.6-terra' does not support the '/v1/responses' API"`. This matches the model card for GPT-5.6 Terra, which calls it out explicitly: *"This model is available on the `openai/v1/responses` path on the bedrock-mantle endpoint. This is different from the `v1/responses` path used by other models on the responses endpoint."* The generic `/v1` path you referenced exists, but the OpenAI-family models this provider targets are only served under `openai/v1` — routing them to `/v1` reproduces exactly the 400 above. So the runtime URL stays as-is, but your underlying point — that the unit tests mock the OpenAI client and would never catch a wrong URL — is fair and now addressed: added `test_client_uses_documented_mantle_openai_path` (head `cd24684`), which asserts the exact `base_url` constructed by `_get_client()` (including region substitution) and documents in its docstring why it must be the `openai/v1` path. If AWS ever changes the routing, that test plus the model-card reference will make the fix obvious. Happy to add a README footnote linking the model card if you think users pointing other Responses-API models at this provider need the distinction spelled out. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
