Investigate how to decouple AIP from Conda, and IF it brings improvements/flexibility

[verovaleros]

The current AIP docker image is 1.92GB, which begs the question 'is all of this needed?'.

A big component on this docker size can be related to the use of Conda to manage the dependencies and requirements.

The task would be to investigate if this is the case, also if it's possible to not use conda and use virtual environments instead and how much would the size reduction would be.

All in all, the key question is: is this change worth it?

[verovaleros]

There are three key files that need modifying to decouple AIP from the use of Conda:

• etc/docker/Dockerfile (same for M1)
• etc/docker/entrypoint.sh
• Additionally, the introduction of a .dockerignore file can help reduce even further the size of the container.
• There may also be some packages in requirements.txt that may not be needed, such as ipython.

Ongoing tests show that with these are some other modifications the docker size can be reduced cca 70%.

[verovaleros]

A complementary approach would be to use multi-stage builds. This would allow to have a base image that can be use for development and a reduced smaller size build with only the required packages for production.

[verovaleros]

Ongoing tests show that a multi-stage build is what works best, using python-vevn instead of Conda, and removing unwanted requirements. Final size shrinks from 1.92GB to 371M.

[verovaleros]

Proposed changes can be found currently in https://github.com/stratosphereips/AIP/tree/40-decouple-aip-from-conda