using python script plus shell to install dependencies

[EBallou]

The script example is a bit simplistic - I'm curious of recommendations where I need to modify the OS and the Python libraries before the script is executed. Below is a snippet of my attempt as an example, but it doesn't seem to take in the actual 'source'. What would be better approach?

(I've also posted this in slack argo-workflows.)

      script:
        name: main
        image: 'python:3'
        command:
          - sh
          - '-c'
        args:
          - >-
            apk add git gcc libc-dev && python -m pip install -r /work/requirements.txt && python
        source: |
           import os

           key = 'MY_ENV_VAR'
           value1 = os.getEnv(key)

           print( "Got my env: ", value

[EBallou]

In chatting in slack - it was noted I'm combining the script and container template syntax ... My issue may more reside in my attempt to combine them. It was suggested by @mac9416 that instead I use a shell script to embed the needed python code instead.

[agilgur5]

Note that building a custom image was suggested in the original Slack thread as first option, similar to the other answer here. A shell script with embedded Python was the second option.

[terrytangyuan]

It might be better to build a new image based on the python3 image with the additional dependencies pre-installed before running your Python script.

[dberardo-com]

is it possible to per-build such a dependency as part of a workflow?

or should one create an ad-hoc init task that checks for an image existence and if it does not exist then builds it and pushes it, so that the next workflow step can use that image ?

what is the best practice here?

[agilgur5]

You could do that if you wanted to, but I wouldn't recommend that.

what is the best practice here?

Production should be static, so you should know all your scripts etc ahead of time and pre-build images for all of them; no dependencies installed at run-time.