While the ability to produce the same exact binaries from the corresponding Git commits is a good idea for any open source project, it is a requirement for software that deals with digital tokens of significant value.
The easiest way to guarantee that users are able to replicate
our binaries for themselves is to give them the same software environment we used in CI. Docker
containers fit the bill, so everything starts with the architecture- and
OS-specific containers in
These images contain all the packages we need, are built and published once (to
Docker Hub), and are then reused as the basis for temporary Docker
images where the
nimbus-eth2 build is carried out.
These temporary images are controlled by Dockerfiles in
we're not publishing them anywhere, we can customize them to the system
they run on (we ensure they use the host's UID/GID, the host's QEMU static
binaries, etc); they get access to the source code through the use of external volumes.
It all starts from the GitHub actions in
is a different job for each supported OS-architecture combination and they all
run in parallel (ideally).
Once all those CI jobs complete successfully, a GitHub release draft is created and all the distributable archives are uploaded to it. A list of checksums for the main binaries is inserted in the release description. That draft needs to be manually published.
The build itself is triggered by a Make target. E.g.:
make dist-amd64. This invokes
scripts/make_dist.sh which builds the corresponding Docker container from
docker/dist/ and runs it with the Git repository's top directory as an external
The entry point for that container is
docker/dist/entry_point.sh and that's
where you'll find the Make invocations needed to finally build the software and
create distributable tarballs.
.github/workflows/release.yml (only for Linux AMD64, ARM and
ARM64): we unpack the distribution tarball and copy its content into a third
type of Docker image - meant for end users and defined by
docker/dist/binaries/Dockerfile.amd64 (and related).
We then publish that to Docker Hub.