Skip to main content
Version: 0.5

E2E Architecture

The document describes high-level Capact architecture, all components in the system and interactions between them.

NOTE: This document showcases the Capact architecture for initial phase of the development. The target GA architecture may differ, and it will be documented later.

Components#

The following diagram visualizes all components in the system:

Components

OCF#

Open Capability Format (OCF) is not a component per-se, but it's also an important part of the system. OCF is a specification of manifests for every entity in the system. It is stored in a form of multiple JSON Schema files. From the JSON schemas all internal SDK Go structs are generated.

OCF manifests are stored in Hub.

CLI#

CLI is command line tool which makes easier with working the OCF manifests and Actions.

To read the documentation about CLI tool, see the Getting started document.

CLI utilizes SDK.

Gateway#

Gateway is a GraphQL reverse proxy. It aggregates multiple remote GraphQL schemas into a single endpoint. It enables CLI and other clients to have a single destination for all GraphQL operations.

Based on the GraphQL operation, it forwards the query or mutation to a corresponding service:

  • Engine - for CRUD operations on Actions,
  • Local Hub - for create, read, delete operations on TypeInstances,
  • Public Hub - for read operations for all other manifests except for TypeInstance.

It also runs an additional GraphQL server, which exposes single mutation to change URL for Public Hub API aggregated by Gateway.

Engine#

Engine is responsible for validating, rendering and executing Actions. Action describes user intention in platform-agnostic way - which Interface or Implementation to run with a given input parameters and artifacts. For Kubernetes implementation, we use Kubernetes Custom Resources to store Action data.

Engine composes of two modules:

  • GraphQL API server, which exposes platform-agnostic API for managing Actions,
  • Kubernetes operator, which handles Action validation, rendering and execution based on Action Custom Resources spec property.

To resolve Actions, Engine uses GraphQL queries for TypeInstances, Interface and Implementations through Gateway.

Engine utilizes SDK. To execute Actions, it uses Kubernetes Jobs, that executes Argo workflows.

Hub#

Capact Hub stores OCF manifests and exposes API to access and manage them. It uses graph database as a storage for the data.

Hub works in two modes:

  • Local Hub exposes GraphQL API for managing TypeInstances (create, read, delete operations),
  • Public Hub, which exposes read-only GraphQL API for querying all OCF manifests except TypeInstances.

Every manifest for Public Hub has to be signed with CLI. Signed manifests are populated with DB Populator, which populates the graph database directly from a given directory structure.

Hub utilizes SDK.

SDK#

SDK is a Go library with low-level and high-level functions used by Engine, Hub and CLI.

SDK can be used by Users to interact with Capact components in a programmatic way.

Detailed interaction#

The section contains detailed interaction diagrams, to understand how the system works in a higher level of detail.

Executing Action#

On the following diagram, User executes the Mattermost installation using CLI.

NOTE: To make the diagram more readable, some details have been omitted, such as TypeInstance injection into workflow and details about watching built-in Runner status. Also, the Gateway component was excluded. Every operation proxied by Gateway is described with (via Gateway) phrase.

Sequence diagram for Mattermost installation