This is a simple OWIN web application that uses FROM scratch in its Dockerfile. The application packaging is an example of The Build Container pattern where container's build procedure is represented as a Two-Phase process:
- Prepare binary components from which application consists of (
Dockerfile.build) - Build the runtime image by using previous step output (
Dockerfile)
Run build-image.sh and you'll see the described procees in a console.
alpine-mono-sdk is a base image that produces application binaries. They are packed in a single file artifacts/runtime.tar which represents the root filesystem in a runtime image.
This image in turn based on Alpine Linux one and have all needed dependencies to compile .NET application including Mono.
Mono is compiled from sources against musl libc library. It also includes mkbundle patch.
build.sh's role is to restore dependencies and run FAKE script. Unfortunately, FAKE can not instantiate F# interpreter host, so temporary workaround for now just to feed fsi.exe directly with build.fsx. F# interpreter comes from FSharp.Compiler.Tools package (I was unable to build F# from sources on Alpine image; perhaps 80K musl's stack size is a showstopper).
MakeAppBundle creates a single executable file which is under certain circumstances is the only file needed to run the application.
I intentionally introduced Mono.Posix dependency to show how to deal with native dependencies.
It is possible to statically link
glibcandlibgcc— runtime dependencies — with bundle by playing withmkbundleandgcccommand line arguments. It is important to note, that linking should be made with-Wl,-Bdynamicanyway to allow CLR load assemblies even from bundle itself).
PackRuntime archives application bundle with native Linux dependencies. As you can see, it includes Mono runtime deps, as well as Mono.Posix specific (libMonoPosixHelper.so and libz). machine.config could be included in bundle itself but to manage, for instance, machineKey it is better to leave it outside of executable.
Under the hooks directory you could find scripts that help to leverage Two-Phase build within Docker infrastructure.