"Failed to return job results" and 400 bad request with known good code
I've been trying to get a serverless endpoint working with a Stable Diffusion script. When I test locally (or with the same hardware on pods) with
--rp_serve_api or --test_input, it works perfectly fine. I can also use the same functions in jupyter or a bare python script and it works as expected. But when I deploy the same code to serverless, I get (...) {"requestId": "(...)", "message": "Failed to return job results. | 400, message='Bad Request', url=URL('(...)')", "level": "ERROR"} with no other information.3 Replies
yeah, I have the same confusion as you, it can run all through locally, but it will report an error on runpod, and I have not solved this problem now
How large is the data in your response? There is a limit of 20MB for for /run and 10MB for /runsync. This applies to both the request and the response.
Responses are around ~15.5-16MiB; I'm not using runsync, so it should fit. I'll switch to uploading to an s3 bucket and returning the address and see if that fixes the issue.