S2i build image with private pip repository

[vincentarthur]

Currently for using S2i to build image for Seldon, it's going through Internet. However, in some cases, the network is only for intranet only which unable to reach Internet. So a new requirement would be -
S2i build image stage should be able to configure whether to use Local private repository for installing python libraries. (Both conda url and folders for .whl)

[vincentarthur]

@jklaise ,Thanks in advance.

[jklaise]

@vincentarthur I've created a pull request #332 which enables you to use local Python wheels to install your runtime dependencies. This solution requires to put the wheels into a "whl" directory where your component source code lives. I understand this is not ideal as you probably want to reuse the same wheels for several components, we have an open issue for this (#324) but presently it seems non-trivial to extend the wrapping process to look for dependencies outside the local directory.

You can test this by using the modified assemble script as a parameter at build time as described here: (https://github.com/openshift/source-to-image/blob/master/docs/user_guide.md)

[vincentarthur]

@jklaise Thanks for the PR. How about this ?

• s2i build . -v /whl/dir:/whl image-name
• (inside assemble) pip install --find-links /whl -r requirements.txt

[jklaise]

@vincentarthur thanks, that looks a lot cleaner. For some reason I couldn't make it work with the -v flag but it does work with the -i (inject directories) flag, so the syntax would be s2i build -i /whl/dir:/whl . builder_image target_image, I think using the -i is preferable to -v as well as it doesn't leave the binaries hanging around after the build process finishes.

[jklaise]

@vincentarthur you can now use the 0.5-SNAPSHOT s2i builder image to achieve the desired functionality, full syntax available here. Closed by #332.