A Factory’s containers.git repo holds the source for Docker containers, as well as Compose Apps. When changes are made to this repo, new Targets will be built. The CI logic for building a Target is based on simple naming rules:

  • Ignores top-level directories that ends with .disabled.
  • A container image will be built for every top-level directory containing a file named Dockerfile.
  • A compose app will be built for every top-level directory containing a file named docker-compose.yml.

Both a container image and compose app will be built if both a Dockerfile and Docker-compose file are included.

Advanced Container Usage

A docker-build.conf file enables advanced functionality. This is sourced by the CI shell script and can set build time options:

  • TEST_CMD: Adding TEST_CMD="<some command>" directs the CI builder to run the command inside the container to verify that it is functioning correctly.
  • SKIP_ARCHS: A container will not be built for the specified architecture, e.g., SKIP_ARCHS=arm64.
  • EXTRA_TAGS_$ARCH: This can work with SKIP_ARCHS. If builds are skipped for arm64, an arm32 container could be tagged for it with EXTRA_TAGS_arm=arm64.
  • DOCKER_BUILD_CONTEXT: Set an alternative directory for the Docker build context.


# Only build for amd64 and arm
# Use a 32-bit arm container for a 64-bit host:
# Use container.git as the build context

Passing Arguments to Build Context

Containers may require Dockerfile ARG support for including build time variables. If the file .docker_build_args exists in a container director, the contents are used as --build-arg options. These are passed to the docker build command.

Static information can be done by defining a file like:

# <container>/.docker_build_args
KEY2="Value with spaces"

This produces a build command that includes --build-arg KEY=Value --build-arg KEY2="Value with spaces".

The need for dynamic arguments means some values must be generated at build time. This can be accomplished is by taking advantage of docker-build.conf; as it is sourced by the build script, it can be used to generate content dynamically.


A common case is including Git commit information in a container:

# $TAG is set by the build script as the Git short hash of the
# containers.git commit being built.
# $x is the path to the container.
cat <<EOF >$x/.docker_build_args
GIT_MSG="$(git log --format=%s -1)"
# NOTE - These ARG's change *every* build. In order to maximize
# Docker build caching, they should be as close to the end of the
# file as possible so that the steps after these lines don't have
# to get re-run *every* build.

Advanced Container Dependencies

If you need custom code to be executed before the Docker build, add pre-build.conf in the top-level directory of containers.git.


Here are examples of what can be done in pre-build.conf:

# Create a file with build environment for container "shellhttpd":
env > shellhttpd/envvars
# Allow containers in factory to use a common base image

# First: Make our images build in a predictable order.
# This ensures 0base is built first so other containers can inherit it:
export IMAGES=$(find ./ -mindepth 2 -maxdepth 2 -name Dockerfile | cut -d / -f2 | sort)

# Second: Modify each container to use the locally build arch-specific base image:
for x in $IMAGES ; do
    echo "Prebuild checking $x for FROM override"
    sed -i "s|${FACTORY}/0base|${_base_img}|" $x/Dockerfile


If there is shared files between containers, the recommendation is to put the common folder in the base image. The containers can inherit the files using Multi-Stage Container Builds.