Note: this post is a customer story by Alexander Malaev, a software developer.
Well, we are actively using Phoenix/Elixir in our projects for backend development, we also have a RoR project as a frontend-service for our Admin UI. Our project consists of a bunch of microservices written in Elixir/Erlang, and we are running it in production with Docker-containers linked together and composed by Docker-compose.
On every push to a project's branch on [GitLab], [GitLab CI] runs tests, style checking, and other tasks. These tasks are configured using
.gitlab-ci.yml. On every merge to
master GitLab builds a release image for us and uploads it to [GitLab Container Registry][registry]. After all, we run
docker-compose pull && docker-compose up -d on the servers to download the latest release images and upgrade our containers.
So, in the following I will describe our release pipeline for Elixir services, using snippets from our project’s
We are using
docker:latest image for our Runner, and several stages:
image: docker:latest stages: - build - styles - test - release - cleanup
Passing some variables:
variables: APP_NAME: project APP_VERSION: 0.0.1 CONTAINER_RELEASE_IMAGE: gitlab.example.org/example/project:latest POSTGRES_HOST: postgres POSTGRES_USER: postgres POSTGRES_PASSWORD: password
These variables are used during the release's build, so they will be available for all the stages. E.g.,
CONTAINER_RELEASE_IMAGE is used on the release stage, as a link to push the release image to. The
POSTGRES_* variables are used to configure postgres service, and to connect later from containers.
Our build stage:
build: before_script: - docker build -f Dockerfile.build -t ci-project-build-$CI_PROJECT_ID:$CI_BUILD_REF . - docker create -v /build/deps -v /build/_build -v /build/rel -v /root/.cache/rebar3/ --name build_data_$CI_PROJECT_ID_$CI_BUILD_REF busybox /bin/true tags: - docker stage: build script: - docker run --volumes-from build_data_$CI_PROJECT_ID_$CI_BUILD_REF --rm -t ci-project-build-$CI_PROJECT_ID:$CI_BUILD_REF
Before running this stage, we create a container which provides volumes for building artifacts. By the way, GitLab CI has a cache volume itself for similar purposes, but I couldn’t make it working correctly with GitLab Runner using Docker image.
test: services: - postgres tags: - docker stage: test script: - env - docker run --rm --link $POSTGRES_NAME:postgres -e POSTGRES_HOST=$POSTGRES_HOST -e POSTGRES_PASSWORD=$POSTGRES_PASSWORD -e POSTGRES_USER=$POSTGRES_USER -e MIX_ENV=$MIX_ENV --volumes-from build_data_$CI_PROJECT_ID_$CI_BUILD_REF ci-project-build-$CI_PROJECT_ID:$CI_BUILD_REF sh -c "mix ecto.setup && mix test"
Notice that we must pass the variables and link postgres manually, since GitLab Runner is passing the variables only to the first level of Docker, but we go deeply ;)
We could link as many services as we want. For example, we are using Kafka for production, and on our test stage we make Kafka service available for running tests.
styles: tags: - docker stage: styles script: - docker run --rm --volumes-from build_data_$CI_PROJECT_ID_$CI_BUILD_REF ci-project-build-$CI_PROJECT_ID:$CI_BUILD_REF sh -c "mix credo --strict"
Release task; we run it only on pushes to
release: tags: - docker stage: release script: - docker run --volumes-from build_data_$CI_PROJECT_ID_$CI_BUILD_REF -e MIX_ENV=prod --rm -t ci-project-build-$CI_PROJECT_ID:$CI_BUILD_REF sh -c "mix deps.get && mix compile && mix release" - docker cp build_data_$CI_PROJECT_ID_$CI_BUILD_REF:/build/rel/$APP_NAME/releases/$APP_VERSION/$APP_NAME.tar.gz . - docker build -t $CONTAINER_RELEASE_IMAGE . - docker login -u gitlab-ci-token -p $CI_BUILD_TOKEN gitlab.example.org:4567 - docker push $CONTAINER_RELEASE_IMAGE only: - master
We are using Conform to achieve runtime configuration of the release using environment variables. I use the approach described on this [blog post][post-env].
Task to cleanup things:
cleanup_job: tags: - docker stage: cleanup script: - docker rm -v build_data_$CI_PROJECT_ID_$CI_BUILD_REF - docker rmi ci-project-build-$CI_PROJECT_ID:$CI_BUILD_REF when: always
It removes the container with volumes created for build artifacts, and removes the image used during the pipeline. This task is running every time, despite the results of any previous tasks.
Below are our Dockerfiles:
FROM msaraiva/elixir-gcc RUN apk add postgresql-client erlang-xmerl erlang-tools --no-cache WORKDIR /build ADD . /build CMD mix deps.get
This image is used to create a container for running tests and style checks.
FROM alpine:edge RUN apk — update add postgresql-client erlang erlang-sasl erlang-crypto erlang-syntax-tools && rm -rf /var/cache/apk/* ENV APP_NAME project ENV PORT 4000 RUN mkdir -p /app COPY $APP_NAME.tar.gz /app/ WORKDIR /app RUN tar -zxvf $APP_NAME.tar.gz EXPOSE $PORT CMD trap exit TERM; /app/bin/$APP_NAME foreground & wait
This Dockerfile is used to build an actual image with the Elixir release.
- Now we don’t use the "Erlang hot upgrade" feature;
- We don’t test if the release is correctly starting, now we are testing it manually and locally;
- Every container uses its own "epmd" and intercommunication between the services, now made using REST apis, but I’m working on integration of [Erlang-In-Docker approach][approach] to use native erlang messaging between services.
I have a plan to write and publish several articles about our release pipeline, to answer the following questions:
- How do we compile and publish assets?
- How do we run our database migrations, since mix tasks aren’t available from the release image?
- What problems are we facing during the implementation of this pipeline, and what solutions have we found.
Thanks for reading!