What tools, tips, and tricks have you used to optimize Dockerfiles?

I wrote a blog post a while ago about optimizing the size (and also build speed) of the containers for AWS Lambdas written in Rust:

Building Super Slim Containerized Lambdas on AWS

Ervin Szilagyi for AWS Community Builders ・ Sep 14 '22 ・ 7 min read

#aws #docker #distroless #rust

Some obvious tricks and tips:

1. Use multi-stage images
2. Cache build folders
3. Use smaller images (distroless or even scratch)

Have you used any non-Docker solutions for efficient container development?

Yes, while I was writing my other blog post:

Containers: Under the Hood

Ervin Szilagyi for AWS Community Builders ・ Oct 15 '22 ・ 15 min read

#containers #docker #linux #beginners

I used Podman and LXC/LXD.

Any triumphs, fail whales, or other stories you'd like to share on this topic?

I guess 2 blog posts should be enough, although, I have even more stories for anybody interested. :)

• put the most changeable layers on the end of docker file. It will speed up building, rebuilding only changed layers.
• analyze weights of your layers using "docker history". It will show you size of each layer. Great hint where to start optimize, eg. not necessary packages
• Add dockerignore file, there you can specify what should not be places in the container. For example some cash file, binary etc. Lack of a dockerignore file is the most common mistake
• put together in a one layer installation of packages (eg. apt-get install) and after that add after installation cache clear in the same layer. It will reduce the final size.
• use alpine base images, it's super light weight OS <3
• When you prepare your production image, copy what is really necessary to run your app

One thing people often forget: installing packages/dependencies will (generally) keep download caches that are totally useless at runtime: npm ci, pip install, etc. but also apt-get.

One thing I like to do: use BuildKit's --mount on RUN stages to put those cache dirs outside the image. You can use a cache mount to keep those between builds, or tmpfs mounts for truely volatile storage.

Do the same with bind mounts instead of COPYing a file to process and delete it in a following RUN (such as copying an archive to then extract and delete it)

Also, use Dive to analyse your images and find bloat. Not only those files overwritten in a subsequent layer (that Dive automatically detects), but also those files you shouldn't have added to your layers (such as those download caches)