Dockerfile 与 shell

Question

I am trying to create an ubuntu image with sqlite and postgreSQL installed. I am confused if what I'm doing is the best practice.

I found out that I can build a very simple dockerfile and install sqlite and postgresql via shell (using docker run -it [image_name] ).

So I'll have a dockerfile like such

FROM ubuntu:18.04
ENV TZ=[time zone data]
RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone

and run docker build some_container:1.0 . Then run docker run -it some_container:1.0 , which will give me a shell, and then run install commands like apt-get install postgres-12 inside the shell.

Is this even the right way to build images? I am confused if this is a good way because many online materials don't use shell but use dockerfile to specify what to install and run.

What benefit does using dockerfile have over editing images via shell?

Answer 1

A Dockerfile is a set of instructions to build your image. You should use it to prepare everything you need to work so next time you spin up the container you won't have to do any prep work.

Not have the postgres installation in the Dockerfile means that every time you want to use your container you'll have to open a shell and manually install it instead of just running the container and having a database ready for you to use.

Answer 2

As soon as you quit that machine you're running, it's gone and you need to re-create it. Every command you entered manually is gone. Dockerfiles are a recipe for creating images reproducibly from scratch.

Taking a bigger picture view: virtual machines are useful, we can agree on that yes? They allow you to run complex services in a simple environment, shielded from the complexities of other stuff which you may want to run in parallel. Setting up a virtual machine can be painful though; you need to start the machine itself, and then log into it to install all the stuff you need. That can be complex and time consuming.

Docker simplifies getting virtual machines up and running in a known good state: you define in a Dockerfile what steps are necessary to create exactly the virtual machine you want, and that's basically it. Docker will build the machine and run it with a simple command, and it does so efficiently (with layer caching and tight OS integration), and all you need is a single text file which you can check into your version control system (as opposed to a multi-GB virtual machine image).