Minimal Docker Images

Docker is a great tool. One of my favorite aspects of building, shipping, and running docker containers is the lightweight nature of deployment. Only that which is needed for the container should be included in the image. The problem is that many running containers include more than what is absolutely necesary for the application to run.

Many of the base images that I see people using for their deployable images have a full OS such as Ubuntu, Debian, or Alpine (although Alpine does have a much smaller footprint). Docker image layering is great and strongly encouraged. I can create my own MySQL image based on the community supported MySQL image from DockerHub. I can then add any other customization I need in my Dockerfile (see below):

FROM mysql:5.6

RUN ./do-custom-stuff.sh

I recently pulled the node:4 image from DockerHub. It was 656.9 MB and I had not even added my application! If we trace the history of the this image, we will find that it is based on the debian:jessie image. To run a node application, I do not need a full OS.

At work, I work with some really smart people. We write and deploy many Java Spring Boot applications. We compile them into a fat jar and then we have a docker image that only has the libc package needed for the JVM and then on top of that, we trimmed out the portions of the JVM that we do not need. The base image for our deployable images is around 100 MB. We will add roughly another 60 MB to the image with the Spring Boot jar that includes all our dependencies. So our deployable images are right around 160 MB. This still seems large when I compare to other images and see that a statically compiled Go application can be right around 20 MB.

These minimal images are nice for deploying to production as they do not have any unnecesary items in the image and will result in faster deployments as it is a smaller image to pull from the registry. With all things though, there are pros and cons each way. The minimal images are not nice for local development if you need to do any sort of debugging. Having a shell in the container to investigate is super handy but is most likely unnecessary in production. As for me, I am always to going to strive to keep my deployable images to a minimum and include only what is absolutely necessary to run my application.