This is a shim/template repo for an Azure DevOps pipeline to build multi-architecture images:
At the Azure Pipelines level, this creates:
- One independent stage for each CPU architecture
- A "wrap-up" stage that runs after all the others that builds and publishes the Docker manifest file
Because I needed a simple, re-usable example of how to build multi-architecture images (in, this case, Linux for several different CPU architectures) that I could both re-use and share with customers.
Images are created atop a local base (in this case Ubuntu) with a corresponding qemu-user-static binary embedded, which allows most CI systems to build ARM images atop an amd64 CPU.
The sample src/Dockerfile only installs zsh into that image, but the intent is that you can build this up from there.
Most of the image generation logic (including embedding QEMU and building the final manifest that allows for automatic discovery of architecture-specific tags) lives inside the Makefile.
This is done because:
- Having a
Makefileallows for easy local testing - It also allows for easy movement between different CI systems
- The actual architecture tagging (and mapping between different styles of architecture references) can be maintained inside the
Makefile - Encapsulating that logic makes the CI YAML files considerably more readable
- This does not cover multi-OS (Linux/Windows) images--the principles are the same, but that needs to split on the CI side to allow for different stages inside different VM agents.
- This only uses Docker Hub (and requires
DOCKER_USERNAMEandDOCKER_PASSWORDto be set as private global variables for the pipeline--I recommend using linked variables at the organization level). - This does not use the built-in Docker actions in Azure Pipelines (because they do not yet support the experimental mode settings that are still required, as of June 2019, to build manifests).