Get involved with the Airship community at the Open Infrastructure Summit Shanghai!

View Sessions

Airship Blog Series 3 - Airship 2.0 Architecture High Level

By Alan Meadows and Rodolfo Pacheco on 2019-8-10.

In order to achieve the goals of vanishing complexity, as well as broaden the variety of supported use cases for Airship 2.0, we shifted far more of the process to the left. We will accomplish this by introducing airshipctl. The airshipctl command line interface is really the heart of the Airship 2.0 platform. It places an emphasis on a thick client that is effectively able to speak to k8s in remote sites and natively understands Argo workflows to drive cluster life cycle management. This contrasts with Airship 1.0 which leveraged a Shipyard API in the remote site, which was a long-lived service and used many Airship specific projects instantiated in the site to accomplish the life cycle management of Airship documents, kubernetes, baremetal nodes, and helm charts.

The goals for this pivot are to reduce the underlying infrastructure required to support updates to an existing site and reduce the number of Airship specific YAML documents required to produce and manage a site, vastly simplifying the overall design. This utility is a net-new GO module. This utility operates on a kubernetes cluster security context and understands how to interpret and generate a skeleton Airship document set. The airshipctl utility is the main entry point for bootstrapping a cluster, collecting and pushing documents, and managing workflows.

To provide greater visibility and ease operations, Airship 2.0 will also introduce an airshipui GO binary that will also operate on the same cluster security context. This user interface will provide users with the ability to drive and visualize workflows, drill into failures, and the ability to manipulate Airship YAML documents and manage them.

These utilities will help drive the four major technologies we will incorporate:

Metal3-IO for Baremetal Provisioning

By incorporating Metal3-IO, we can take advantage of the work being done in that community to treat baremetal machines as proper Kubernetes resources that are declared and reconciled. It also enables us to begin utilizing the Cluster-API framework. This will reduce the number of baremetal related Airship specific documents and replace them with Metal3-IO Kubernetes CRD documents. This transition also allows us to support multiple operating systems as well as advanced baremetal configuration such as firmware versions and RAID configuration.

The Cluster-API for Kubernetes Bootstrapping and LCM

The Cluster-API is a Kubernetes project to bring declarative, Kubernetes-style APIs to cluster creation, configuration, and management. By leveraging the Cluster-API for cloud provisioning, we can take advantage of upstream efforts to build Kubernetes clusters and manage their lifecycle. Most importantly, we can leverage the multitude of providers that already exist, similar to Metal3-IO for baremetal. This will allow Airship deployments to target third-party clouds such as Azure, AWS, and even OpenStack VMs. There are even Cluster-API providers that support Kubernetes deployments on top of already provisioned infrastructure, enabling BYO use cases as well.

Argo for Workflows

Argo is a Container-native workflow engine for orchestrating parallel jobs on Kubernetes. Argo Workflows are implemented as a Kubernetes CRDs. This allows us for the first time to truly realize completely declarative workflows within the Airship document bundle. Each step in the declared workflow is a container and models multi-step workflows as a sequence of tasks or capture the dependencies between tasks using a graph (DAG). The Airship community will provide a set of curated and tested workflows out of the box for the most common deployment and lifecycle needs, but operators and vendors can also introduce their own custom workflow declarations.

Argo will also allow us to vastly simplify the dependencies required to drive workflows in an Airship environment. Where Shipyard required several Airflow components running, message queue infrastructure, and databases, Argo only requires a single operator and stores all state in Kubernetes CRs. Similarly, the Shipyard API was a long-lived process running in the site, whereas the API for Argo is Kubernetes itself. By removing all these dependencies and leveraging Argo, Airship will now be able to simplify the process of upgrading the orchestrator.

Kustomize for Document Layering and Substitution

This tool provides a purely declarative approach to configuration and customization that embraces Airship's approach and adheres to and leverages the familiar and carefully designed Kubernetes API. In short, Kustomize allows for customization of Kubernetes YAML configurations. From an Airship perspective, the introduction of Kustomize will allow us to transform the way we provide layering, substitution, and YAML document management. This functionality was previously provided by Deckhand, an Airship specific tool. Where Deckhand required users to invest in understanding how to perform layering, substitution, and other operations in an Airship specific tool, Kustomize is an example of an entrenched community tool for doing the same.

Besides embracing these technologies, we are also leaning towards a closer relationship with CNCF principles. We are modifying all Airship custom YAML schemas to comply with Kubernetes Custom Resource Definition schemas. These will take us a step closer to the goal of simplifying the cost of entry into the project and leverage all the tools already available to work with these types of schemas.

We're also evolving some existing Airship components, such as Pegleg, and Armada. In Airship 2.0, Pegleg will no longer be a standalone tool, and some of its functionality, such as document rendering, linting, and secret rotation, will be incorporated into airshipctl. Similarly, Armada will continue to help orchestrate collections of helm charts and their associated overrides in Airship 2.0. We will be enhancing it to store and consume Armada documents as proper Kubernetes CRs as well as allowing it to be orchestrated by an Argo workflow.

The picture below represents the target Airship 2.0 architecture:

In the upcoming blog posts, we will elaborate on the evolution of each of the Airship 1.0 components:

  • Shipyard: an evolution on the Front Door
  • Drydock: and its relationship to Metal3-IO
  • Armada: Embracing cloud-native
  • Promenade: Kubeadm and Cluster-API
  • Deckhand: Document rendering the cloud native way, Kustomize!
  • Pegleg: Document management
  • Treasure Map: Supporting Airship 2.0
  • DivingBell: Day 2 approach in Airship 2.0
  • Airship 2.0: Deep dive review
  • What's Next: Beyond Airship 2.0