I’m still curious about that you have trained the models with the data from AWS S3. It seems not supported in the upstream TesnorFlow. Have your implemented this backend storage internally?
And you extend the Custom Source of TensorFlow Serving for S3. If you have implemented the S3 storage in TensorFlow, there is nothing to do in TensorFlow Serving to read the files in S3.
It would be great if you have done something. Is there any Github issue about supporting S3?