Docker is a technology stack for running things in virtualized environments. In Keboola Connection (KBC), we use a limited set of Docker features. Their description follows. For a full technical description of Docker, consult its official documentation.
At first sight, Docker is similar to other virtualization technologies (such as VMware or VirtualBox). However, there are some fundamental differences, the main one being that Docker runs only virtualized applications, not the entire OS.
Docker has Docker Images and Docker Containers. To create a Docker image, create a Dockerfile. Dockerfiles contain instructions on how the Docker image should be built, and this represents the environment (OS + modifications) in which an application runs.
A Docker image contains everything required to run an application. It usually has an entrypoint, which is a single command executed when the image is run.
When you run an image (start an application in it), a Docker container is created. The container is a sandbox isolated from the image itself and cannot make permanent changes to it. Maybe somewhat surprising, this is very important. When you run the image again (and create a new container), it won’t be affected in any way by the previous container. The Docker image is therefore stateless and acts as a template. The state is stored only in the container.
Docker Images are created by executing the instructions written in a Dockerfile. It is a simple text file consisting mostly of shell commands which must be executed to prepare the application for running. Docker images can be based on other images. So if you need minor modification to a system, you do not have to build the whole thing from scratch. If you want Images to be reused, push your Dockerfile to a Docker registry. The registries (Dockerhub, Quay) will build the image; anyone interested in using it can download it. AWS ECR is a private repository and has no build triggers. You need to push the images manually or using a deploy script in your CI pipeline.
Docker image names are based on the following scheme:
registry-name/account-name/image-name:tag where registry-name
and account-name can sometimes be omitted. For example, you can refer to a Docker hello-world image as:
docker.io refers to the Docker Hub registry,
library refers to account (common library is default),
hello-world refers to the image name,
latest refers to the tag.
Image tags work similarly to Git tags as they refer to a specific build of the image. However, Docker tags can be moved easily, so they do not always need to refer to the same build. The general convention is that the latest tag points to the same (latest) build and is movable.
We have wrapped Docker in our Docker Runner component. The component
runs registered Docker images. Docker Runner
has an API
which allows to run Docker Images and encrypt arbitrary values.
Docker Runner takes
care of injecting the right data, creating, running, and terminating the container, and uploading
the result data to KBC Storage. All images to be run in KBC must have an
Before you run Docker applications in KBC, make sure to set up your Docker environment. Before you develop a dockerized application for KBC, you should be able to create and run dockerized applications in your own environment.
If you are already familiar with Docker, jump straight into our sample application
code in PHP.
The demo application itself starts with a single
can exist independently (without Docker), and contains unit and functional tests.
The repository includes also the Docker image definition in the
Dockerfile. The Docker environment including the application
is prepared by the Docker image definition. The Travis CI service is used to builds Docker image automatically on every commit and
deploy it to KBC and public registries.
A similar application is also available in Python.