Containerized build environment for compiling an executable Golang package and packaging
it in a light-weight Docker container.
One of the (many) benefits of developing with Go is that you have the option of compiling your application into a self-contained, statically-linked binary. A statically-linked binary can be run in a container with NO other dependencies which means you can create incredibly small images.
With a statically-linked binary, you could have a Dockerfile that looks something like this:
FROM scratch COPY hello / ENTRYPOINT ["/hello"]
Note that the base image here is the 0 byte scratch image which serves as the root layer for
all Docker images. The only thing in the resulting image will be the copied binary so the total
image size will be roughly the same as the binary itself.
Contrast that with using the official golang
image which weighs-in at 500MB before you even copy your application into it.
The golang-builder will accept your source code, compile it into a statically-linked binary
and generate a minimal Docker image containing that binary.
In order for the golang-builder to work properly with your project, you need to follow a few simple conventions:
The golang-builder assumes that your "main" package (the package containing your executable command) is at the root of your project directory structure.
. ├─Dockerfile ├─api | ├─api.go | └─api_test.go ├─greeting | ├─greeting.go | └─greeting_test.go ├─hello.go └─hello_test.go
In the example above, the
hello.go source file defines the "main" package for this project and lives at the root of the project directory structure. This project defines other packages ("api" and "greeting") but those are subdirectories off the root.
This convention is in place so that the golang-builder knows where to find the "main" package in the project structure.
If your "main" package does not reside at the root of your project structure, you can accomodate this by specifying the
MAIN_PATH environment variable which references the relative path to your main package.
For example, suppose your structure is like the following (slightly modified from above so that your "main" package resides in the "cli" subfolder):
. ├─Dockerfile ├─api | ├─api.go | └─api_test.go ├─greeting | ├─greeting.go | └─greeting_test.go └─cli ├─hello.go └─hello_test.go
You would then add
MAIN_PATH=cli to your environment specification.
Canonical Import Path
In addition to knowing where to find the "main" package, the golang-builder also needs to know the fully-qualified package name for your application. For the "hello" application shown above, the fully-qualified package name for the executable is "github.com/CenturyLink/hello" but there is no way to determine that just by looking at the project directory structure (during the development, the project directory would likely be mounted at
$GOPATH/src/github.com/CenturyLink/hello so that the Go tools can determine the package name).
In version 1.4 of Go an annotation was introduced which allows you to identify the canonical import path as part of your source code. The annotation is a specially formatted comment that appears immediately after the
package main // import "github.com/CenturyLink/hello"
The golang-builder will read this annotation from your source code and use it to mount the source code into the proper place in the GOPATH for compilation.
There's a good chance that your project imports at least one third-party Go package. The golang-builder obviously needs access to any packages that you've imported in order to compile your code. By default, golang-builder will
go get any packages you've imported which aren't part of your project already.
The problem with doing a
go get with each build is that golang-builder may end up with versions of packages which are different than those you developed against. Depending on the stability of the packages that you are importing this may not be an issue. However, if you want to maintain strict control over your dependency versions you may want to look at the Godep tool.
If you are using Godep to manage your dependencies golang-builder will reference the packages in your
Godeps/_workspace directory instead of downloading them via
If you would like to have golang-builder package your compiled Go application into a Docker image automatically then the final requirement is that your Dockerfile be placed at the root of your project directory structure.
If your Dockerfile resides somewhere else, you can acommodate this by specifying the
DOCKER_BUILD_CONTEXT env var. This will be used to override the default value of
After compiling your Go application, golang-builder will execute a
docker build with your Dockerfile.
The compiled binary will be placed (by default) in the root of your project directory so your Dockerfile can be written with the assumption that the application binary is in the same directory as the Dockerfile itself:
FROM scratch EXPOSE 3000 COPY hello / ENTRYPOINT ["/hello"]
In this case, the hello binary will be copied right to the root of the image and used as the entrypoint. Since we're using the empty scratch image as our base, there is no need to set-up any sort of directory structure inside the image.
If golang-builder does NOT see a Dockerfile in your project directory (or
DOCKER_BUILD_CONTEXT directory) it will simply stop after compiling your application.
There are a few things that the golang-builder needs in order to compile your
application code and wrap it in a Docker image:
- Access to your source code. Inject your source code into the container by mounting it at the
/srcmount point with the
- Access to the Docker API socket. Since the golang-builder code needs to interact with the Docker API in order to build the final image, you need to mount
/var/run/docker.sockinto the container with the
-vflag when you run it. If you omit the volume mount for the Docker socket, the application will be compiled but not packaged into a Docker image.
Assuming that the source code for your Go executable package is located at
/home/go/src/github.com/CenturyLink/hello on your local system and you're currently in the
hello directory, you'd run the
golang-builder container as follows:
docker run --rm \ -v "$(pwd):/src" \ -v /var/run/docker.sock:/var/run/docker.sock \ centurylink/golang-builder
This would result in the creation of a new Docker image named
Note that the image tag is generated dynamically from the name of the Go package. If you'd like to specify an image tag name you can provide it as an argument after the image name.
docker run --rm \ -v "$(pwd):/src" \ -v /var/run/docker.sock:/var/run/docker.sock \ centurylink/golang-builder \ centurylink/hello:1.0
If you just want to compile your application without packaging it in a Docker image you can simply run golang-builder without mounting the Docker socket.
docker run --rm -v $(pwd):/src centurylink/golang-builder
- CGO_ENABLED - whether or not to compile the binary with CGO (defaults to false)
- LDFLAGS - flags to pass to the linker (defaults to '-s')
- COMPRESS_BINARY - if set to true, will use UPX to compress the final binary (defaults to false)
- OUTPUT - if set, will use the
go buildto output the final binary to the value of this env var
- MAIN_PATH - if set, this (relative) path will be used as the location of the "main" package
- DOCKER_BUILD_CONTEXT - if set, this (relative) path will be used as the build context location (where the Dockerfile should reside)
The above are environment variables to be passed to the docker run command:
docker run --rm \ -e CGO_ENABLED=true \ -e LDFLAGS='-extldflags "-static"' \ -e COMPRESS_BINARY=true \ -e OUTPUT=/bin/my_go_binary \ -v $(pwd):/src \ centurylink/golang-builder
An additional image,
centurylink/golang-builder-cross, exists that works identically to
golang-builder save for the presence of the additional options presented above. This uses a larger base image that will build linux and OSX binaries for 32- and 64-bit, named like
mypackage-darwin-amd64. This will use CGO, and you may find that some code – for example things from the
os package – do not behave the same under cross-compilation in a container as they do natively compiled in OSX.
By default it will build Linux and OSX binaries for 32- and 64-bit but you can override this with environment variables.
docker run --rm \
-e BUILD_GOOS="linux" \
-e BUILD_GOARCH="arm amd64" \
-v $(pwd):/src \
This command will build Linux binaries for amd64 and arm.
More information can be found in the Docker Hub page for the official Go images.
If your Go application needs to make calls to SSL endpoints you may find your application failing with a message like:
x509: failed to load system roots and no roots provided
One of the down-sides to using the scratch image is that you no longer have access to the root CA certificates which come pre-installed in most base images. There are a few different options for dealing with this:
- Disable SSL verification. This is not recommended for obvious reasons.
- Bundle the necessary root CA certificates as part of your application.
- Use a different base image which already contains the root CA certificates.
We've created a minimal base image for applications that require SSL verification. The centurylink/ca-certs image is simply the scratch image with the most common root CA certificates pre-installed. The resulting image is only 258 kB which is still a good starting point for creating your own minimal images.