published: 21.11.2018
The essence: Concourse CI can be integrated with anything using resources, is declaratively described, and easy to extend.
Concourse is written in Go. It consists of a server with a web interface, API, scheduler, and worker registrar. An SSH tunnel creates a connection between the workers (cross-platform) and the web. The web stores data in Postgres, which stores pipelines, logs, etc. Concourse cannot store artifacts.
There is a utility called fly that allows you to interact with the web. Manage commands, projects, pipelines, etc. As a rule, everyone has enough of the web interface for interaction. For debugging, you can execute a task locally or log in to an existing one.
The guys have microservices, for each of which Concourse runs tests, and then makes a mark in Bitbucket that everything is okay and it’s time to review. When a PR is approved, merged, tagged, tests are rerun, a container build is done, and a notification is sent to Slack. Then comes the deployment to test by clicking a button in the web or changing the Kubernetes manifest. Then the tester will make a note in the task that everything is okay, and you can deploy to production. Now we are negotiating to automate deployment to test and production completely.
There is a pipeline consisting of resources and jobs. Resources are cubes with which Concourse interacts with something. For example, the Slack resource allows you to work with Slack. Jobs in which commands are executed. There is a built-in dashboard with the status of tasks.
There are built-in resources, there are about 20 of them. For example, Git, which constantly checks for new commits and downloads code. There are community resources. For example, one that can create pull requests in Bitbucket. We declare what external resources we need in the pipeline and use them.
Resources can fetch secrets from different sources (Vault, Kubernetes, etc.) or a separate YAML file.
Each resource consists of 3 binary files (check, in, out). Through check, the resource checks what it needs to do. In is used within the job, where it is described what needs to be done. Out sends the result.
Contains instructions. When triggered, which tasks to perform, which resources to use, what to do if everything falls / passes successfully. Tasks can be outsourced to files/repositories and reused in different projects. Tasks can be executed in parallel and aggregate the result.
The author is currently addressing resource requirements and rewriting built-in resources in Go.
I consult about what I write about, you can contact me via telegram @aladmit or via email [email protected]