I didn't have any prior experience with Go, gRPC, docker or kubernetes, so a working docker image with an image scaling service is as far as I got. There's a few hacks due to my lack of Go expertise:
- I edited api.pb.go directly as the service couldn't compile with the auto-generated code from
protoc, likely due to some missing flags or version differences. - I forward requests from service.go to the python server service.py and I chose to connect to the client as part of service.go. Ideally this would be handled in api.go, but the only way I managed to get that working was to remake the connection for every request, and I opted for the former solution.
Run
python -m grpc_tools.protoc -I./proto --python_out=./proto --grpc_python_out=./proto ./proto/api.proto
to generate python source code for gRPC server interface.
This is symlinked in the python service folder.
The image will be processed from the bytes stream if available, else it will be read from a given URL.
If Greyscaling: true is given, the processed image will be grey-scaled.
Resize: true with the options Height, Width and AllowCrop resizes the image using the following logic.
If only Height or Width are given, the other value will be set to keep the aspect ratio of the image.
If neither is given, 1024x768 will be used as default.
If the requested image ratio doesn't match the given image, the image will be squished or stretched during resizing, unless AllowCrop: true is given.
In that case, the same behavior is used if the requested aspect ratio is within a 5% margin of the given image.
Outside this margin, a random crop will be performed that keeps as much of the original image as possible.