Package PyPI Dependencies in the Image During Build Time or Defer to Container Runtime?

Image Builds
1

So, I am packaging a batch inference job (written in python) as a docker image to execute periodically. There are many dependencies of the job, which are captured in a requirement.txt like this

tensorflow==2.12.0
polars==1.12.0

I have to build the environment using this requirement file, straightforward enough. But I have a design choice to make (prompting this question) between two approaches

  1. Build the environment (install all the packages after downloading from pip) during image building phase, then during container runtime (once a day) just run the main.py inside the container
  2. Defer the environment building to the container runtime (just before main.py is run), which can be done as requirement.txt is available inside the container.

So far as I can see, each has its own advantage

  1. Quicker container runtime, as the environment is already built, saving the time to download and unpack all the pypi packages.
  2. Smaller image size, as the libraries taken together take a good amount of space

The job is not really latency sensitive, as in, I save about 4-5 minutes a day by prepackaging the dependencies instead of repeating at every run, but it does not practically matter.

But is there any other trade of I am missing, or does the docker community have any general recommendation, considering security and all? Any other parameter I should look at to make a decision or is there a best practice covering the scenario?

3

Why does the answer sound like from ChatGPT?

4

Definitely looks like an AI.

Anyway, put the installations needed within the image, instead of having the user download those each and every tiem they want to start a container from that image

Look at multi-stage builds to slim up your image

5

The post is removed now, as AI generated content is not allowed in this forum. Furthermore, the post had a self-promoting link to website, which is also not allowed.

If you see such kind of posts again: feel free to flag those posts, so that they get moderator attention.