Deployment on Kubernetes/RHOS fails: Cannot connect to the Image Caption Generator REST endpoint

[ptitzler]

[04/06/20 15:47:23] INFO: Connecting to ML endpoint at http://localhost:5000/model/predict
[04/06/20 15:47:23] ERROR: Cannot connect to the Image Caption Generator REST endpoint at http://localhost:5000

[ptitzler]

To fix this we need to add the ability to define the service URL at runtime using an environment variable, making it possible to adjust the setting as needed. Otherwise users will have to either edit the source code https://github.com/IBM/MAX-Image-Caption-Generator-Web-App/blob/master/app.py#L38 or change the Dockerfile https://github.com/IBM/MAX-Image-Caption-Generator-Web-App/blob/master/Dockerfile#L24. Neither of these alternative is a viable solution.

[ptitzler]

To make this work I had to manually set the endpoint to http://max-image-caption-generator:5000, where max-image-caption-generator is the service name in RHOS.

[ajbozarth]

Nice catch, this would be fixed by updating the Dockerfile to conditionally set the port and ml-endpoint args based on ENV_VARS or args to the docker command. Those args to app.py already address this issue for cases directly calling the app