Agree. The issue with reproducibility comes up mostly once you are building your own containers, many of them, and run them in production but then it can hit very hard because it becomes too expensive then to work around all of the following: Once you are hit by a problem (sometimes years after building the original one), it’s hard to debug bc you can’t be certain to be able to reproduce what you have. It’s even difficult to just find out what is in an existing Docker container, companies are running costly scans for that. And finally you can’t compose (you can mostly layer) as well one Docker container with another. Using Nix with flakes gets rid of all of that and you can still have Docker or other containers as output.
Agree. The issue with reproducibility comes up mostly once you are building your own containers, many of them, and run them in production but then it can hit very hard because it becomes too expensive then to work around all of the following: Once you are hit by a problem (sometimes years after building the original one), it’s hard to debug bc you can’t be certain to be able to reproduce what you have. It’s even difficult to just find out what is in an existing Docker container, companies are running costly scans for that. And finally you can’t compose (you can mostly layer) as well one Docker container with another. Using Nix with flakes gets rid of all of that and you can still have Docker or other containers as output.