as someone who does stuff in my lab that can translate to a work context, i absolutely second this opinion.

if i am labbing to learn, then learning the best way to do it is always be the main focus, even if it means restarting what I was doing to change how some prerequisite is setup or functions.

today, OP is working with jellyfin, but as an example, what happens if later they get security cameras and want to use some sort of local ML to analyze events, and don’t want to put a lot cpu utilization to that task during lulls in activity? a solution might be to dynamically create and destroy containers for the analysis tasks, and the background on a network setup in an unrelated container stack that would allow scaling that means one less problem to solve later.