Architecture

How it works

To solve the complex application orchestration problem Orkestra builds a Directed Acyclic Graph using the application, and it’s dependencies and submits it to Argo Workflow. The Workflow nodes use workflow-executor nodes to deploy a HelmRelease object into the cluster. This HelmRelease object signals Flux’s HelmOperator to perform a “Helm Action” on the referenced chart.

Sequence

1. Submit an ApplicationGroup custom resource object
2. For each “application” in ApplicationGroup download the Helm chart from “primary” Helm Registry
3. For each dependency in the Application chart, if subcharts found in charts/ directory, push the subcharts and the application chart to the ”staging” Helm Registry (Chart-museum).
4. Generate and submit the Argo (DAG) Workflow
5. In parallel,

• (Executor nodes aka workflow pods will) submit and probe the status of the deployed HelmRelease CR (.Status.Phase)
• (helm-controller will) watch and deploy Helm charts referred to by each HelmRelease CR to the Kubernetes cluster

Key Pieces

• The ApplicationGroup custom resource type, on which workflow definitions are based. Orkestra basically uses this as its own deployment template, wrapping the Helm releases within.
• Orkestra’s own operator, which interprets the ApplicationGroup input, initiating the actual workflow steps within Argo Workflow.
• Helm charts are referenced in ApplicationGroup documents. Orkestra caches/stages the required Helm charts in a local repository for which it uses ChartMuseum.
• The actual Helm releases as per workflow steps triggered by Argu are executed through a Helm operator that is part of Orkestra.

---