A collection (and build system) of neatly optimized containers to run any workload (some day, incomplete yet).
Warning! To get all of the benefits you have to use only images with the same tag. When you upgrade to new
tag then change tag for all mcc based image that you use.
- The main idea is to share exactly the same image layer as base.
- Reduces memory footprint due to overlayfs exposing inodes from the same image.
- Reduces the IO overhead of running the containers (no need to read the same files for every container).
- Reduces the amount of data to be transferred and stored.
- Because of the above we can pull more handy packages into the base without thinking too much. It just does not have to be so size-optimized because it is fetched only once. 🎉
- Also the builds should be more deterministic - very specific versions shall be pulled.
- Each container has a component stage which shall contain only a single layer with only
the files that need to be overlayed over base image to use that feature.
- This allows to easily compose containers with multiple features by pulling every file from any chosen set of components.
- Ideally we should make sure that the exact component layer is overlayed in the resulting image - this gives us the same benefit as using the same layer for base. I cannot guarantee that yet with docker, but I will keep searching for a clean and universal solution.
Obviously some of the benefits of this approach will be nullified if you don't use overlayfs, however,
being the default nowadays it would be a rare case.
Read about this overlayfs inode magic in the original kernel devel article and the official Docker Documentation.
The images are auto-built and pushed to NimbleCo org at Docker Hub.