Operate directly on the snowbot container (needs the volume flag):

docker build . -t pytest
docker run -it pytest

Operate on docker-compose:

docker-compose up
docker-compose up --build  # to rebuild the underlying container
docker-compose down

Source code changes are picked up in the container without requiring intervention. Changes to requirements.txt are not.

What don't I like about this architecture?

• Have both Setup.py and requirements.txt
• Both direct file entry point and CLI entry point
• Probably doesn't handle vendored well, though we can add that as its own step of the build process
• docker-compose logging is ugly

What do I like about this architecture?

• Can easily map pure docker commands to monolith top-level utility commands
• Going to production dockerfiles seems easy
• Much simpler env setup process
• Internalized VPN handling
• Likely easy config

What do I have to do next?

• Get the roc database dump in here
• Containerize the actual snowbot
• Figure out what top-level API looks like and wire it up (what commands do we have to use to run stuff? Who runs these commands?)

Problems

• What happens with native packages? Scipy, numpy, matplotlib?
  • https://gist.github.com/orenitamar/f29fb15db3b0d13178c1c4dd611adce2
  • https://medium.com/google-cloud/deploying-scipy-numpy-with-docker-205e9afac3b0